Lecture Notes 2: Ramsey-Cass-Koopmans Model1

Zhiwei Xu ([email protected])
The Solow model does not consider individual optimal decisions. The models dynamic structure
is simply introduced by the capital accumulation rule. The consumption rule is exogenously given,
as a result, the saving rate is exogenous and constant. In this lecture, we will investigate the
Ramsey-Cass-Koopmans model, in which the micro-level optimal behaviors are seriously modelled.
In particular, the saving rate is instead endogenously determined by the household optimization
decisions.
As the Ramsey model is in continuous time, before we formally discuss the details, we rst
introduce the standard solution method Variation of Calculus.

Variation of Calculus

Consider a following integral:

Zb

f (t; x(t); x(t))

(1)

where a and b are some constants. The function x(t); captures the path of x along the time t;
can either be real-valued or Rn valued. For example, x (t) can be the consumption path c (t) ; or
the vector of consumption path and saving path [c (t) ; s (t)] : Let X be the set of all dierentiable
functions dened on the support [a; b]. A typical optimization problem in continuous time is dened
as: choosing the optimal path x (t) 2 X to maximize (or minimize) J subject to the terminal
conditions:
x(a) = ; x(b) = :
(2)
Illustrative examples.
Example 1. Minimum Distance. Find the curve which joints two points on the plane with
the minimum distance. A curve joining A and B can be represented by x(t) with x(a) = ,
and x(b) = .

The distance along each innitesimal segment of x(t) as ds; and we have
s
p
p
dx 2
2 dt:
dt = 1 + [x(t)]
_
ds = (dt)2 + (dx)2 = 1 +
dt

(3)

This note is partly borrowed from the lecture notes of Prof. Yi Wen at Tsinghua University and Prof. Pengfei
Wang at HKUST. I thank them for sharing their materials.

2
The total distance between points

and
J=

is then given by

Zb p

2 dt:
1 + [x(t)]
_

(4)

The problem is then equivalent to nd a curve x(t) 2 X to minimize J subject to x(a) = ,

The budget constraint is given by

c (t) + s_ (t) = rs (t) :

s_ (t) gives us
Z T
J=
e

u(rs (t)

s_ (t))dt:

(6)

(7)

The problem is to nd a saving path s (t) to maximize J:

f (t; x(t); x(t))

has continuous rst and second partial derivatives with respect to all its arguments. Suppose that
there exists a function x (t) 2 X that optimizes J.
Given any path h (t) satisfying h (a) = h (b) = 0 and a small " > 0; we dene a new path x" (t)
such that
x" (t) = x (t) + "h (t) :
(8)
Therefore, the new path x" (t) just deviates from x (t) with a small perturbation. Dene the value
of J on x" (t) as J ("), we then have
J (")

J (t; x" (t) ; x_ " (t)) =

Zb

f t; x (t) + "h (t) ; x_ (t) + "h_ (t) dt:

(9)

As x (t) 2 X optimizes J; we must have

=
"=0

Zb h
a

i
_
fx(t) h (t) + fx(t)
h
(t)
dt = 0;
_

(10)

3
x(t))
_
x(t))
_
Note that, when " = 0; the derivatives fx(t) = @f (t;x(t);
; fx(t)
= @f (t;x(t);
do not depend on
_
@x(t)
@ x(t)
_
h (t) : For the second term in the bracket, according to the integral by parts, we have

Zb

_
fx(t)
_ h (t) dt = fx(b)
_ h (b)

fx(a)
h (a)
_

Zb

dfx(t)
_
dt =
h (t)
dt

Zb

h (t)

dfx(t)
_
dt:
dt

(11)

Putting last equation into (10) yields

dfx(t)
_
h (t) dt = 0:
dt

fx(t)

(12)

Since the function h (t) takes any form with h(a) = h(b) = 0, last equation implies
fx(t) =

dfx(t)
_
; for t 2 [a; b] :
dt

(13)

This condition is called Euler equation which is the necessary condition for the optimal path x (t)
that optimizes J:
Minimum Distance Example (Continued). In this case, since f (t; x (t) ; x(t))
_
=
we have
1
2
2
:
fx(t) = 0; fx(t)
= x(t)
_
1 + [x(t)]
_
_

p
2;
1 + [x(t)]
_
(14)

The Euler equation (13) implies

(15)

or
x (t) =
To determine the

and

2;

1t

2:

(16)

from terminal conditions x (a) = ; x (b) = , we have

a
b

(17)

Therefore, the optimal path x (t) = b a t + b b a a : That is, the curve that minimizes the
distance between point and is a straight line.

Ramsey-Cass-Koopmans Model

Firms and households are representative with unit measure. The market is competitive. There is
only one goods, it can be used either for consumption or investment. The production technology
is given by
Yt = F (Kt ; At Lt ) ;
(18)

4
where L is the total labor input, At is the exogenous technology, growing at rate g > 0;
A_ t = gAt :

(19)

We assume F (:) is concave function and constant return to scale.

2.1

Firms Decision

Each period, from the market the representative rm hires labor Lt with the wage rate wt ; and
rents capital Kt with the rental rate rt : The rm chooses Lt and Kt to maximizes its prot
t

= Yt

rt K t

wt Lt :

(20)

The demand of capital and labor are given by following optimal conditions
rt =

@F (Kt ; At Lt )
;
@Kt

(21)

wt =

@F (Kt ; At Lt )
:
@Lt

(22)

As F (:) is constant return to scale, we have

(23)

Plugging factor demands (21) and (22) into last equation gives us
Yt = rt Kt + wt Lt ;
or

2.2

(24)

= 0: That is, the constant return to scale of production function implies zero prot.

Households Decision

The representative household is a big family, in which there are Mt members. Each individual
inelastically supplies one unit labor. Therefore the total labor supply is Mt : The population Mt is
assumed to grow at the rate n
M_ t = nMt :
(25)
The representative household maximizes following life-time utility
Z 1
U=
e t u(Ct )Mt dt;
0

(26)

5
where Ct is the consumption of a member, u(Ct ) is the corresponding utility level, and
discount rate for the future. In particular, the utility function is assumed to be
u(C) =
where =
and only if

C1
1

is the

(27)

Cu00
u0

captures the coe cient of relative risk aversion. The utility function is concave if
0:

The budget constraint for the household is given by2

) Kt + wt Mt :

(28)

The optimization problem of representative household is to maximize (26) subject to (28).

2.3

Competitive Equilibrium

In the competitive equilibrium, the household and the rm achieve their individual optimum, and
each market clears. In particular, Mt = Lt : The budget constraint (28) and the input demands
(21) and (22) jointly imply the resource constraint is
Ct Lt + K_ t + Kt = Yt

2.4

(29)

Dynamic System

Since there are two potential growth trends, to solve the model we need to rst transform the economy to the stationary one. Dene the detrended variables as xt AXt Lt t ; while for the consumption
Ct
: The life-time utility can be rewritten as
of each individual, we dene ct = A
t
U

Consider a very short time interval dt =

Dividing both sides by

ct1
1
Z 1

At1

= A10

ct1
1

L0

Lt dt

A0 egt
e

ct1
1

L0 ent dt

dt

; the budget constraint is

+ Kt+

Kt = [(rt

] Kt + wt Mt :

yields
Ct Mt +

dKt
= (rt
dt

) Kt + wt Mt :

6
where =
(1
) g n: To guarantee that the utility function is not explosive, we need to
d log Kt
3
t
assume > 0: Without loss of generality, we set A0 = L0 = 1: Note that K_ t = dK
dt =
dt Kt ;
d log Kt
k_ t
where dt = kt + g + n: The budget constraint can be rewritten as
ct + k_ t + (g + n + ) kt = f (kt ) ;

(30)

K
where f (k) = F AL
; 1 : Hence, the competitive equilibrium is equivalent to the solution of following social planners optimization problem

max

fct ;kt g 0

c1t
1

dt

(31)

subject to (30). Denote the Lagrangian multiplier of (30) as t : The social planners problem can
be written as
(
)
Z 1
h
i
1
c
t
max
e t
+ t f (kt ) ct k_ t (g + n + ) kt dt:
(32)
1
fct ;kt g 0
Remember that the optimal ct and kt should satisfy the necessary condition (13). Therefore, we
have
ct = t ;
(33)
d e

t
= e t _t
dt
Combining last two equations, the optimal consumption path is given by

f 0 (kt )

(g + n + ) =

c_t
1 0
=
f (kt )
ct

(g + n + + ) :

(34)

(35)

Moreover, the optimal capital kt satises resource constraint:

ct

(g + n + ) kt :

(36)

The dynamic system is fully described by the ordinary dierential equation (OED) system (35)
and (36). Note that given any initial state (c0 ; k0 ), the system (35) and (36) only provide necessary
conditions for the optimal path.

2.5

The Phase Diagram

According to (35), when c_t = 0; the steady-state capital k satises following equation
f 0 (k) = + g + n + :
3

More rigoriously, to guarantee U is nite, we need the growth rate of e

(37)
1

t ct
1

less than zero; or equivalently

7
As f (k) is a concave function, f 0 (k) is decreasing in k: If kt > k , we have f 0 (kt ) ( + g + n + ) <
0; or c_t < 0: If kt < k ; we have c_t > 0:
According to (36), when k_ t = 0; the steady-state c and k satisfy following equation
c = f (k)

(g + n + ) k;

(38)

which describes a hump-shaped curve in the (c; k) space. For those (c; k) above the curve, we have
k_ t < 0: And for those (c; k) below the curve, we have k_ t > 0:
Note that (37) is the set of all combinations of (c; k) that ensure c_t = 0; and the (38) is the set
of all (c; k) that ensure k_ t = 0: We call these two curves as the equilibrium locus. The intersection
of them is the steady state.
The phase diagram consists of four areas separated by the locus (37) and the locus (38).
Area I: ct " and kt # : Any c k pair in this area indicates that capital stock is relatively high
and consumption is relatively low, thus according to (35) and (36), consumption will increase and
capital will decrease. The above dynamics imply that if the system starts from any point in this
area, consumption and capital will eventually diverge from the steady-state point (c ; k ).
Area II: ct # and kt # : Any c k pair in this area indicates that both capital stock and
consumption are relatively high, thus according to (35) and (36), both of them will monotonically
decrease as long as (ct ; kt ) stays in this area. The above dynamics imply that in this area there
exist some c k pairs as initial points, from which consumption and capital will eventually converge
to the steady-state point (c ; k ).
Area III: ct " and kt " : The dynamics in this area is just opposite to those in Area II.
Area IV: ct # and kt " : The dynamics in this area is just opposite to those in Area I.
According to the above dynamic analysis, there exists a unique path (solid line with arrows in
Figure 1) such that starting from any points in this path, the system will eventually converge to
the steady state. We call this unique equilibrium path as the "saddle path".
Note that as the ordinary dierential equations (35) and (36) just describes the dynamics of
c_t and k_ t instead of the levels of ct and kt ; these two equations only provide necessary conditions
for the optimal path. To get the optimal path, we need to impose initial condition and terminal
condition. Specically, the stationarity condition (both ct and kt cannot diverge) gives the initial
condition (will discuss late), and the non-ponzi-game condition,
lim e

s!1

Rs

Ks = A0 L0 lim e
s!1

Rs (g+n)s

ks

0;

(39)

denes the terminal condition. The non-ponzi-game condition means that one cannot constantly
borrow from outside to cover its consumption. It can be shown that as long as the budget constraint

8
is satised, the non-ponzi-game condition always holds (see Romers textbook, page 53). Therefore,
in the Ramsey model, we just need to consider the initial condition.

Figure 1. Phase diagram of Ramsey model

2.6

Modied Golden Rule and Balance Growth Path

Remember that the golden rule is dened as the steady-state (S-S) consumption at the maximum
level. From (38), the capital stock at implied by the golden rule, k GR ; satises
f 0 k GR = + g + n:

(40)

However, the optimal steady-state capital k satises

f (k ) =

+ + g + n:

(41)

Therefore, only if the discounting rate = 0; the optimal S-S capital k is identical to the golden
rule capital k GR : It is easy to see that if > 0; k GR > k : Moreover, the S-S saving rate in the

9
Ramsey model is
s

= 1

c
y
f (k )

(g + n + )k
f (k )
k
= (g + n + )
f (k )
g+n+
=
< :
+g+n+
= 1

(42)

Last equality is due to f f(k(k )k) = : The optimal S-S saving rate is less than the golden rule saving
rate if > 0: The intuition is that keeping the maximum consumption cGR at each period is not
optimal because the household cares more about current period than the future ( > 0). Therefore,
starting at k = k GR (higher than k ); the household always has incentive to consume more than
cGR in the current period.

2.7

Transitional Dynamics: Algebraic Analysis

Now we provide a rigorous discussion about the optimal solution implied by (35) and (36). As the
system is nonlinear, we need to rst linearize the system around the steady state (c ; k ) : Dene a
new function (ct ; kt ) as
#
"
1
ct [f 0 (kt ) (g + n + + )]
(ct ; kt ) =
:
(43)
f (kt ) ct (g + n + ) kt
Then the system (35) and (36) can be expressed as
"
#
c_t
= (ct ; kt ) :
k_ t
The rst-order Taylor expansion around the steady state gives us
"
#
c_t
' (c ; k ) + c (c ; k ) (ct c ) + k (c ; k ) (kt
k_ t

(44)

k ):

(45)

#
#
"
00
1
c f (k )
0
: Dene
Note that (c ; k ) = 0; c (c ; k ) =
; k (c ; k ) =
0
f (k ) (g + n + )
1
00
0
= 1 c f (k ) > 0 and joint with (37), we have f (k ) (g + n + ) = : The linearized system
then takes the form
"
#
"
#
c_t
ct c
=
:
(46)
k_ t
kt k
"

10
"

#
0
where matrix =
: From the above ordinary dierential equations, we can derive the
1
linearized equilibrium locus c_t = 0; k_ t = 0: In particular, c_t = 0 implies
kt = k :

(47)

And k_ t = 0 implies
ct

c =

(kt

k ):

(48)

The gure below depicts the phase diagram of the linearized system, which is quite similar to the
phase diagram of the nonlinear system (Figure 1).

Figure 2. Phase diagram in the linearized system

is full-rank, it can
(49)

is the eigenvalue matrix

"

0
2

q
1
1
+
2
2q
1
1
2
2

+4 ;

+4 :

(50)

11
Obviously,

> 0 and

< 0: Furthermore, the eigenvector matrix P is

1+

(51)

The dierential equations can be further expressed as

c
k

(52)

It can be shown that the solution of the above system is given by

c
k

(53)

or more compactly,
ct

kt

= a1 e

1t

2 a1 e
1t

1 a2 e

+ a2 e

2t

2t

(54)

(55)

where a1 and a2 are functions of initial state (c0 ; k0 ) and given by

a1 =

1 (k0

k )

(c0

c )] ;

(56)

a2 =
1

2 (k0

k )

(c0

c )] :

(57)

Note that (54) and (55) are just the solution derived from the necessary condition of the optimal
decisions. Indeed, they are not the optimal solution (i.e., saddle path) of the problem (31). This
is because given any pair of (c0 ; k0 ) ; there exists a combination of paths fct ; kt g correspondingly.
That is, there are potentially innite combinations of paths fct ; kt g that solve (54) and (55). As a
result, to nd out the unique optimal path (saddle path), we need to impose extra conditions.
The rst condition we have to impose is the stationarity condition. As you will see, this condition
indeed is equivalent to the initial condition. Since we have 1 > 0 and 2 < 0; to make sure that
ct and kt are not explosive, we have to set a1 = 0: Then (56) implies
c0

c =

1 (k0

k ):

(58)

Obviously, last equation is essentially the initial condition. Besides, in the Ramsey model, we
do not need the terminal condition (non-ponzi-game condition), because the terminal condition is
identical to the budget constraint.
Plugging (58) into (57) gives us
a2 = k0

k :

(59)

12
From (54) and (55), together with a1 = 0 and a2 = k0

k ; the saddle path is given by

kt

= (k0

k )e

ct

= (c0

c )e

2t
2t

(60)

(61)

In addition, last two equations and (58) imply

c =

1 (kt

k );

(62)

the slope of saddle path is 1 : As 1 > (see equation 50), the saddle path is steeper than the
locus k_ t = 0. This explains why in the phase diagram, the saddle path is in the middle of c_t = 0
and k_ t = 0:

2.7.1

Transitional Dynamics: Unexpected Changes of

So far, we have discussed how to obtain the saddle path, along which the economy will eventually
converge to the steady state (c ; k ). Given the xed fundamental (no shocks, no changes of the
values of deep parameters), the saddle path is a unique path that solves the optimization problem.
Any initial (c0 ; k0 ) o the saddle path will eventually diverge, and the corresponding path is not
optimal to the economy.
In this section, we will discuss the scenario that if there is an unexpected change of the fundamental, what will the economy respond. To take a concrete example, we discuss the change of the
discounting rate :
Suppose that in the period 0, the economy is at the old steady state: (c ; k ) : In the same
period, there is a sudden permanent increase in . You may take the change of as a surprise or
an exogenous shock. We denote the new as new ; and the new steady state as (c ; k ) : Under
the new new ; the dynamic system will be changed to
ct
kt

new new
2 a1 e

= anew
1 e

new
1

new
1

+ anew
2 e

new new
1 a2 e
new
2

new
2

(63)
(64)

where

q
1 new 1
=
+
( new )2 + 4 new > 0;
(65)
2
2q
1 new 1
new
=
( new )2 + 4 new < 0:
(66)
2
2
2
new
To see how the economy transits to the new steady state, we need to solve the coe cients fanew
1 ; a2 g :
Note that according to (56) and (57), we have
1
new
anew
=
(k
k ) (c +
c )] ;
(67)
1
new
new [ 1
new
1

anew
=
2

1
new
1

new
2

new
2

(k

k )

(c +

c )] :

(68)

13
where
is the change of consumption (or impulse response) in period t = 0: The above two
new
equations show that fanew
1 ; a2 g determine the change of consumption in the shock coming period,
thus to get the transition path we need to solve these two coe cients.
Since

new
1

> 0; to make sure the path is not explosive, we have to set anew
= 0; or
1
=

new
1

(k

k )

(c

c ):

(69)

The describes the jump in consumption in the period 0 corresponding to the permanent change
in : Therefore, the transition path is fully described by
kt
ct

k
c

= (k
= (c +

k )e

new
2

c )e

new
2

(70)
t

(71)

Note that the k0 (= k ) does not change in the period 0, because k is a predetermined (or state)
variable. Figure 3 illustrates the transition dynamics: A ! B ! C.
When the fundamental changes, the consumption adjusts such that (c; k) will jump towards the
new saddle path.4 Thus, we call the variable ct as control variable or jump variable.

Figure 3. Transition dynamics under unexpected change of

In this case, the consumption immediately jumps to the saddle path, this is mainly due to the change of
unexpected and permanent.

is

14
2.7.2

Transitional Dynamics: Expected Changes of

Suppose that in the period 0, the economy is at the old steady state (c ; k ) : In the same period,
there is an announcement saying that there will be a permanent increase in starting from t = T .
That is, = old when t < T; and = new ; when t
T: We denote the new steady state as
(c ; k ) :
For those periods t < T; the dynamic system is still the old one:
ct

kt

= a1 e

1t

2 a1 e
1t

1 a2 e

+ a2 e

Since we assume k0 = k ; from (56) and (57), we have

2t

2t

(72)

(73)

+4 ;

(74)

+4 :

(75)

and

a1 =

(c0

a2 =

(c0

For the periods t

kt

c
k

new new
2 a1 e

=
=

anew
1 e

(76)

c ):

(77)

2
new

T; the dynamic system under

c );

new
1

new
1

will be changed to
new new
1 a2 e

anew
2 e

new
2

new
2

(78)
(79)

where
new
1

new
2

1
2
1
2

and
anew
=
1
anew
=
2
Since

new
1

q
1
+
(
2q
1
new
(
2
new

1
new
1

new
2

1
new
1

new
1

new
2

new 2

new

> 0;

new 2

) +4

new

< 0:

k )

(cT

) +4

(kT

new
2

(kT

k )

(cT

(80)

c )] ;
c )] :

(81)
(82)

> 0; to ensure that the path is not explosive, we have to set anew
= 0; or equivalently,
1
(cT

c )=

new
1

(kT

k ):

(83)

15
Therefore, the transition path after t
kt
ct

T is fully described by
k
c

= (kT
= (cT

k )e
c )e

new
2
new
2

t
t

(84)

(85)

To get the full transition path, we need to solve the part for t < T: From (72) and (73), you may
nd the transition path for t < T is determined by the coe cients fa1 ; a2 g ; which are functions of
c0 : To solve c0 ; we rst, according to (72) and (73), write cT and kT as functions of c0 :
cT

2 a1 e

kT

= a1 e

1T

1T

+ a2 e

1 a2 e
2T

2T

+c ;

+k :

Putting last two equations into the initial condition (83) for the period T , we can nally solve c0 .
Once we have c0 ; the transition path during the period t < T; is described by (72) and (73).
The gure below illustrates the transition dynamics under an expected increase in : It shows
that even though the change of will occur in the future, the optimal transition for the economy
is to jump in the initial period (remember that the is still unchanged for all t < T ), and keep
moving under the old system until the announcement is realized, at that time (ct ; kt ) just arrives
the new saddle path. The transition path in the Figure 4 is: A ! B ! C:

Figure 4. Transition dynamics under expected change of