1 Introduction
The growth model developed by Ramsey (1928) and subsequently adapted by
Cass (1965) and Koopmans (1968) is arguably the single most important
theoretical construct in modern macroeconomics. It is the foundation of both
growth theory and business cycle theory. It takes pride of place in every graduate
text in macroeconomics and growth theory.1
The geometry of the dynamics of the Ramsey model is well understood;
every graduate student ritually ponders its phase diagram. There is a paucity of
analytical solutions, however. Long and Plosser (1983) developed a discrete-time
example with log utility, Cobb-Douglas technology, and 100 percent depreciation
which is often used to exposit RBC models.2 It is also possible to derive an
explicit solution when there is a constant gross saving rate [Kurz (1968), Barro,
Mankiw, and Sala-i-Martin (1995), Barro and Sala-i-Martin (2004, Chapter 2),
and Wälde (2005)]. Recently, Mehlum (2005) has provided a closed form solution
for the case of a Leontieff technology. While these are illuminating examples,
each involves some unappealing features. The assumption of 100 depreciation in
Long and Plosser is highly unrealistic, as is the restriction to log preferences. At
any rate, this example only yields a solution in discrete time; it doesn’t work in
continuous time [Chang (1988)] . The constant saving rate formulation requires
the elasticity of intertemporal substitution to be tied to the rates of depreciation
and time preference in an unintuitive way. The Leontieff technology has long
been known to be a razor-edge case. It would be nice to have an example with a
reasonable rate of depreciation, a non-unitary elasticity of intertemporal
substitution, and diminishing returns. Absent such a solution, empirical
implementation of the model has required either a linear approximation around
the steady state [Barro and Sala-i-Martin (2004)] or numerical simulation
[Mulligan and Sala-i-Martin (1991)].
This paper will derive an explicit solution to the canonical version of the
Ramsey model with constant elasticity preferences and Cobb-Douglas
technology. The solution can only be obtained when capital’s share of GDP is
equal to the reciprocal of the intertemporal elasticity of substitution. The
distinguishing feature of this constellation of parameters is that it generates a
1 Some examples are Blanchard and Fischer (1989), Romer (2005), Sargent (1989, 2002),
Turnovsky (1999, 2000), Obstfeld and Rogoff (1999), and Barro and Sala-i-Martin (2004).
2 In this case both the utility function and production function have a unitary elasticity of
substitution. Benhabib and Rustichini (1994) show that solutions can be found for a more general
class of models, where (given CRRA preferences and CES technology) the elasticities of
substitution of the utility function and production function are equal, but not necessarily equal to
one.
linear saddle path. This permits an explicit solution for the dynamics of the
capital stock, along with an exact expression for the speed of convergence.
Although this is admittedly a special case, it provides a concrete example
that makes the dynamic workings of the model transparent. It is not intended to
supplant graphical and numerical analysis, but to serve as a complementary tool
that helps reveal the economic properties of the model. As Boucekkine and RuizTamarit (2004b) put it when motivating the same parametric device
in studying
the Lucas (1988) model, “. . . there is an obvious need to complement this
computational and qualitative literature with full-fledged theoretical proofs . . .”
Two elements of this solution have been anticipated in the literature. First,
what makes the model tractable is the fact that the consumption/capital ratio is
constant when labor’s share equals the intertemporal elasticity. Barro and Sala-iMartin (2004) are aware of this fact in the Ramsey model. Xie
(1991) exploits this
property in the Romer (1986) model, while Xie (1994) and Boucekkine and RuizTamarit (2004a) use it to shed light on the Lucas (1988) model. The
tractability
of this case has also been put to good use in some stochastic growth models. For
example, Chang (1998) solved the closely related “inverse optimal” problem for
this case, deriving the utility function and rate of time preference that generate a
constant equilibrium consumption/capital ratio. More recently, Wälde (2005) has
used this combination of parameters to solve a Poisson model of endogenous
growth cycles. The innovation here is to use the constancy of the
consumption/capital ratio to arrive, via a Bernoulli transformation, at closed form
solutions for the capital stock and the speed of convergence in the non-stochastic
Ramsey model.
Second, the derivation centers around the hypergeometric function.
Boucekkine and Ruiz-Tamarit (2004b) have recently, and independently, applied
the hypergeometric function to the Lucas (1988) model. The recursive structure of
the Lucas model allows them to derive analytical results without restricting the
intertemporal elasticity. The Ramsey model, however, is generally not recursive;
the only case where it is recursive seems to be the example studied here, where
the intertemporal elasticity is equal to the reciprocal of capital’s share.
The outline of the paper is as follows. Section 2 recapitulates the Ramsey
model. Section 3 sketches the solution, leaving the details to an appendix.
Section 4 discusses the properties of the solution. It shows vividly how the
speed of convergence varies with time, a feature of the model that is masked by
estimates of the speed of convergence calculated from a Taylor-series
approximation around the steady state. It also suggests a pedagogical application
of the model: students are often asked to solve the Cobb-Douglas version of the
Solow (1956) model by using a Bernoulli transformation. This model could be
adapted to serve as a complementary example of a tractable solution to the
Ramsey model.
Section 5 extends the model to study two problems that have heretofore
lacked analytical solutions. The first is the dynamic effects of a temporary change
in government expenditures; the second is to the transition dynamics of the Jones
and Manuelli (1990) model.
Section 6 calibrates the model. This is obviously a very special case, so I
do not wish to press the issue. However, I suggest that this calibration may not be
terribly unrealistic. If we adopt a broad notion of capital and embrace a degree of
agnosticism about the elasticity of intertemporal substitution it may be a
reasonable first approximation to the real world.
Section 7 offers some concluding thoughts.
2 The Model
Since the model is so familiar, my description of it will be succinct. Imagine an
economy inhabited by a large number of identical, infinitely-lived consumers.
For simplicity there is neither population growth nor technological growth. I
adopt the functional forms commonly used in empirical, and in many theoretical,
applications: technology is Cobb-Douglas, with a capital share ofα ∈[0,1]; the
rate of depreciationδ ∈[0,1]is constant; there is a constant rate of time preference
ρ ∈[0,∞]and preferences are time-separable, with a constant elasticity of
intertemporal substitution, σ ∈[0,∞].
It is well known that the dynamics of per capita consumption c and capital
k are determined by the following pair of differential equations:
k = k α -δk – c
= σ (αk α -1 – δ – ρ) (1)
(2)
cc
Equation (1) is the resource constraint; Equation (2) is the Euler equation. There
are two boundary conditions: the initial capital stock ko and the transversality
condition
.
lim 0
1
→ ∞
=
–
–
t
e ρtc σ k (3)
For future reference, the steady state is
1
1 –
+
=
α
ρ α
δ
k (4)
( )
.
1
c k k k
α
ρ δ α
= α – δ = + – (5)
The system is saddle-path stable. Given the initial capital stock k0 there is a
unique value of initial consumption c0 that will satisfy the transversality
condition.
3 A Solution
The system of equations (1) – (2) is dauntingly nonlinear. To approach a solution,
define the capital-output ratio (a Bernoulli transformation) by z = k 1-α and the
consumption-capital ratio by x = c k . Using these transformations the system can
be rewritten as
z = (1-α )[1- (δ + x)z] (6)
( )
+ – –
1 x
+ 1 (7)
x z
x
–
= σα δ σ ρσ The Bernoulli transformation converts the capital accumulation equation
[Equation (1)] into the linear, albeit non-autonomous, differential equation in
Equation (6). Despite this simplification, the system still does not allow an
analytical solution. Equation (7) is still non-linear.
Suppose, however, that σ =1 α . In that case, z disappears from equation
(7), and the system becomes recursive: Equation (7) reduces to
( )
,
1
x
xx
+
– +
= –
α
δ α ρ
(8)
while z evolves according to Equation (6), given the forcing process x determined
in Equation (8). This is a simple, autonomous logistic equation.
The solution to Equations (6) and (8), subject to the initial condition
k0 and the transversality condition [Equation (3)] is given by the following result.
PROPOSITION 1: If σ = 1 α the transitional paths of consumption and capital
are
( )
,
1
c k
α
ρ + δ -α
= (9)
-α ( -α -α ) -( -α )ρα+δ -α
= + –
1
1
1
1 1
0
1 t
k k k k e . (10)
Since to some extent the method is the message of this result, I will sketch its
derivation before discussing its properties.
PROOF:
Consider the logistic equation in Equation (8). This has the general
solution
xAe xt
x
x
+
=
1
(11)
where A is an arbitrary constant of integration and ( )
α
ρ +δ -α
=
1
x . Substitute
this back into Equation (6) to arrive at
( )
= – – +
z
1 α 1 δ
+
z
xAe
x
1 xt (12)
In the appendix I demonstrate that the general solution to Equation (12) is:
1 1(1 α,1, ;ω) α 1 (1 α )( δ ) (1 )1 α ,
– – – + –
– + +
=
F d Bx e x t xAe xt
z (13)
2 δ
+
x
where B is another arbitrary constant of integration, d = 1+ (1-α)(δ + x) x ,
,
1 xt
xt
xAe
xAe
+
ω = and
( ) ( )( )
( )
( )( )( )
(1 )(2 ) . . .
1 2 3
1
1
1 2
d d
2 1 1 ,1; ; 1 2 3 +
+ +
– – –
+
+
– –
+
–
– = + ω
α α α
ω
α α
ω
α
α ω
d d d d
F d
(14)
is a hypergeometric function.3
The following lemma pins down the constants A and B .
LEMMA. Given the initial condition
k0 the transversality condition [Equation
(3)] will be satisfied if and only if A = 0 and
.
1
0
1
= –
–
δ
B x z (15)
+
α
x
The proof of this lemma is sufficiently involved to be relegated to the appendix.
The intuition, however, is straightforward. Notice that the transversality
condition (3) can be written as
.
lim 0
→ ∞
=
– –
–
t
e ρt x σ k σ (16)
1
1 1
Consider the solution for x in Equation (11). If A > 0 then x would fall to zero
over time, driving marginal utility and k to infinity. If A < 0 then x would
become negative in finite time. If A = 0 x is constant and, given Equation (10),
Equation (16) is satisfied. Therefore, the transversality condition is satisfied if and
only if A = 0.
Applying this Lemma to Equations (11) and (13) completes the proof:
Since A = 0, Equation (11) implies that x = x,∀t. This yields Equation (9). Next,
evaluating Equation (13) at the values of A and B described in the Lemma yields
( )(x )t
e
x
α δ
δ
– – +
+
1
0
1
. z
=
1 δ (17)
z
x
+ –
+
Using the definition of z we can then back out the solution for the capital stock in
Equation (10).
4 Discussion
Note several interesting features of the equilibrium. First, Equation (9) asserts
that in every period the ratio of consumption to capital must be constant,
3 The function 2 F1(a, b, d;ω)is convergent for arbitrary a,b, and d for real -1 < ω < 1. See
Weisstein (2004).
c k = x = x. In other words, the saddle-path is linear, with a slope equal to x.
This is depicted in Figure 1. This property is now fairly well known, for a range
of models [Chang (1988), Xie (1991, 1994), Barro and Sala-i-Martin (2004),
Boucekkine and Ruiz-Tamarit (2004a, b), Wälde (2005)].
The Phase Diagram
Figure 1
The Phase Diagram
Second, Equation (10) provides an exact solution for the dynamics of the
capital stock. The growth rate of the capital stock is
( )
–
+
=
+
– –
+ –
= = –
–
+
–
– –
– –
1
.
1
1
1 1
0
1 1
0
α
δ α
ρ
α
α α
α α
δ α
ρ
ρ δ
α α
ρ δ
γ
k k
k k e
k k
kk
t
k
(18)
On inspection, the growth rate decreases over time: The growth rate depends upon
the deviation between net output and consumption. However, consumption grows
linearly with the capital stock, while gross output increases at a decreasing rate,
due to diminishing returns.
We can also obtain an exact, analytical calculation of the speed of
convergence to the steady state:
(1 ) .
ln
-1
+
= –
∂ ∂
= –
α
δ α
ρ
α
γ
β
k k
k
k
k (19)
Barro and Sala-i-Martin (2004, chapter 2), in contrast, resort to a calibration of a
linear approximation around the steady state. The accuracy of their estimate
diminishes rapidly the further away capital is from its steady state. The
numerical calculations of Mulligan and Sala-i-Martin (1991) are exact, but lack
the transparency of the solution in Equation (19).
Third, the model yields a clean expression for the dynamics of the gross
saving rate, s =1- c k α . From Equations (9) and (10) it follows that
( ) ( ) ( )
( ) ( ) ,
1
1
1
0
1
1 1
0
1
t
t
s s s e
s k k k e
δ α
ρ
α
δ α
ρ
α
α α α
α
ρ δ α
+
– –
+
– –
– – –
= + –
+ –
+ –
= –
(20)
where s =1- xα (δ + ρ) is the steady state saving rate. The model predicts that, if
k0 < k , the saving rate will decrease monotonically over the transition to the
steady state. The intuition is simple. In general, the increase in the capital stock
exerts offsetting income and substitution effects on saving [Barro and Sala-iMartin (2004, Chapter 2, Section 6.3)]. As interest rates fall the
substitution effect
tends to reduce saving. However, as the gap between current income and
permanent income diminishes, there is less incentive to save. Because people like
intertemporal substitution (σ > 1) in this calibration of the model the substitution
effect is strong, and dominates the income effect: Since the consumption-capital
ratio is constant it follows that s =1- xk 1-α =1-αx R, where R =αk α -1 is the
gross interest rate. The saving rate therefore decreases as the interest rate falls.
This is not consistent with the empirical evidence [Barro and Sala-i-Martin
(2004)], since the saving rate actually increases over the transition.
Finally, the solution suggests a useful pedagogical exercise. It is well
known that the Solow (1956) model with Cobb-Douglas technology is an
autonomous Bernoulli equation [Chiang (1967, p. 454), Obstfeld and Rogoff
(1999), Barro and Sala-i-Martin (2004, p. 44)].4 Students are often asked to solve
4 Barro and Sala-i-Martin (2004, appendix 2.C) attribute the use of this transformation in the
Solow model to Jaume Ventura. However, it was known at least as early as Chiang (1967).
the Solow model explicitly using this “trick.” It is non-trivial to show – using
either the methods here, or those in Barro and Sala-i-Martin (2004, Appendix 2b)
— that the saddle-path is linear when σ =1 α . However, students may as an
exercise be told to assume this result – or tell an intuitive story to motivate it —
and then be invited to solve the rest of the model. Once they have determined the
steady-state capital-output ratio, they can write Equation (6) as an autonomous
Bernoulli equation, which can be solved just like the Solow model.
5 Extensions
The model can be adapted to provide explicit solutions to a range of problems in
growth theory. I will consider two here.
5.a A Temporary Increase in Government Expenditures
A traditional thought exercise [Romer (2005), Turnovsky (1999)] is to trace out
the dynamic effects of a temporary increase in government expenditures in the
Ramsey model. This problem has been addressed graphically [Blanchard and
Fischer (1989), Romer (2005)] and solved using linear approximations
[Turnovsky (1999)]. I provide an explicit solution for the case where σ = 1 α. 5
First we need to incorporate the government into a decentralized version
of the model. Suppose that that government issues debt b , collects lump-sum
taxes τ, and engages in expenditures G. Given an initial debt of b0 and the
interest rate r = αk α -1 – δ , its flow budget constraint is then
b = rb + G -τ. (21)
To obtain an analytical solution I assume that government expenditures are
proportional to the capital stock, G = gk. While this assumption is just a
convenience here, it is standard in endogenous growth models [see Turnovsky
(2000), for example].6
Imagine a temporary increase in g from time zero to time T :
5 It is also possible to consider expected future changes in government expenditure, as well as
temporary and expected future changes in distortionary taxes on capital.
6
The traditional exercise involves an increase in government expenditures that is independent of
the capital stock or output [Blanchard and Fischer (1989), Romer (2005)]. This makes no
substantive difference. Alternatively, the model could be solved by making government
expenditures proportional to output.
g g t T
g t g t T
= < ≥
= ≤ <
2 1
1, 0
(22)
The dynamics of capital and consumption are now given by
, .
, 0
1 2
k k c g k t T
k k k c g k t T
= – – – >
= – – – ≤ ≤
δ
δ
α
α
(23)
= σ (αk α -1 -δ – ρ)
cc
(24)
Define ( )
1 1
1
x – g
+ –
=
α
ρ α δ
,
( )
2 2
1
x – g
+ –
=
α
ρ α δ
and .
∆g = g1 – g 2 In the
appendix I prove
PROPOSITION 2: Assuming that σ =1 α the equilibrium paths of consumption
and capital in response to a temporary increase in government expenditures from
time 0 to time T are:
( )
, .
, 0 ,
1
2
2
1
1
x k t T
k t T
e
x
xg
c
x t T
= >
≤ ≤
∆
–
=
–
(25)
( )
( )
( ) ( )( )
( )
, .
0 ,
1 ,
1
1 ,1, ;
1
1
1
1
1
1
1
1
2
2
2
2 1 1
1
1
1
k e t T
t T
e e
x
g
e
x
e g
x
g
k F d
t
T
x t T x t
x t T
x t T
>
= + + – +
≤ ≤
+ Ω – ∆
∆
∆ –
– –
+
=
–
+
– –
–
–
– – +
–
–
–
–
α
δ α
ρ
α
α
α
α δ
α
ρ δ
α
ρ δ
α
α
ρ δ
α
(26)
where
1 .
1
1 1 ,1, ; 1
2
2
2
0 2 1 1
+ ∆
∆
∆ –
– –
+
Ω = – –
–
–
x T
xT
xT
e
x
g
e
x
e g
x
g
F d
x
z α
δ
(27)
Equation (25) makes the dynamics of consumption per unit capital
transparent: As shown in Figure 2, consumption initially falls, given the initial
capital stock, so that c k drops.7 Then, as the capital stock falls, c k increases
until the economy regains the saddlepath at timeT ; thereafter it remains constant.
Intuitively, people respond to the increase in government expenditures by
reducing consumption. Since the change in government expenditures is
temporary, however, they reduce consumption by less than the increase in
government expenditures. This causes the capital stock to decrease over time, so
that consumption per unit capital has to accelerate as it approaches the saddlepath.
Effects of a Temporary Increase in Government Expenditures Figure 2
Effects of a Temporary Increase in Government Expenditures
7 Figure 2 is identical to Figure 2.9 in Romer (2005) except that the k = 0 locus rotates around the
origin rather than making a parallel shift.
5.b Transition Dynamics in an Endogenous Growth Model
Another illuminating application of the model is to the transitional dynamics of
endogenous growth. This too has been analyzed graphically [Barro and Sala-iMartin (2004)], but never solved explicitly.
I adopt a Cobb-Douglas version of the a-k technology suggested by Jones
and Manuelli (1990): y = k α + ak. Since the marginal product of capital is
bounded away from zero, this is sufficient to induce long run growth. The
dynamics of k and c are now
k = k α + ak -δk – c
= σ (αk α -1 + a -δ – ρ). (28)
(29)
cc
In order to generate an asymptotically balanced growth path I assume that
a > δ + ρ. In order to satisfy the transversality condition I also assume that
ρ > (1-α )(a -δ ).
In the appendix I derive the closed form solution to this model:
PROPOSITION 3: If σ = 1 α the transitional paths of consumption and capital in
the endogenous growth model are
( )( )
,
1
c a k
α
ρ – -δ -α
= (30)
( )
( δ ρ ) α
α α
α
δ ρ
α
δ ρ
α – – –
–
–
= – – – + + – – 1
1
1
10
a t
e
a
k
a
k . (31)
Equation (30) asserts that the consumption/capital ratio is again constant over
time. Equation (31) implies that the growth rate of capital is
( )
(a )t
δ ρ
– –
–
–
1 a k
α
–
+ a
ρ
– – k
10 a
α
δ ρ
– –
=
e
ρ
a
kk
α α
α
δ α
δ ρ
α
δ α
–
– –
–
– –
+
10
. (32)
This makes it clear that the growth rate falls over time, tending asymptotically
to(a – δ – ρ) α .
6 Calibration
We can only solve the model in the special case where the intertemporal elasticity
of substitution σ is equal to the reciprocal of capital’s share of income α. Is this
likely to be a reasonable description of the real world?
At first glance, the answer to this question would seem to be no. The
traditional estimate of physical capital’s share of GDP is about one third [Simon
(1990)]. This would require the intertemporal elasticity of substitution to be as
large as three. However, the intertemporal elasticity of substitution is usually
thought be quite small, certainly less than one [Hall (1988), Attanasio and Weber
(1993), Epstein and Zin (1991), Campbell, Lo, and MacKinlay (1997)].
Recent empirical work casts this conventional wisdom in doubt. Barro and
Sal-i-Martin (2004) point out that if α = .3the Ramsey model does a poor job at
explaining a range of empirical features of growth. However, “all of these
shortcomings are eliminated” [Barro and Sal-i-Martin (2004); see also Mankiw,
Romer, and Weil (1992) and Obstfeld and Rogoff (1994)] if we adopt a broader
notion of capital (so that it includes human as well a physical capital), and
increase capital’s share: Barro and Sala-i-Martin suggest α = .75, Obstfeld and
Rogoff (1994) α = .66. If we set α = .75, we are only restricting the elasticity of
substitution to be about σ =1.33. The last decade has seen more and more
empirical evidence that σ is in fact greater than one, with estimates ranging from
1.5to as much as 3[Bufman and Leiderman (1990), Attanasio and Weber (1989),
Koskievic (1999), Vissing-Jorgensen (2003)]. There is also the interesting
empirical fact that, in a cross section of countries, growth rates are negatively
correlated with production risk [Ramey and Ramey (1995)]. In the canonical
stochastic A-k growth model [Smith (1996)] this can only occur if the
intertemporal elasticity is greater than one.
What of the model’s prediction that the consumption-capital ratio should
be constant? It is a traditional “stylized fact” of growth, enunciated by Kaldor
(1963) and later supported by Maddison (1982), that the ratio of consumption to
physical capital is roughly constant over time. What can we say about the
consumption-capital ratio if we adopt a broader notion of capital, one that that
encompasses human as well as physical capital? Empirical estimates [Jorgenson
and Fraumeni (1989), Lettau and Ludvigson (2004)] suggest that human capital
represents between two thirds and 90 percent of total wealth. In a recent paper
Zhang (2004) establishes that the ratio of consumption to total wealth is roughly
constant. He estimates the steady-state consumption-capital ratio to be .05739, in
annual terms. Now compare this with the predicted consumption-capital ratio of
the model: Following Barro and Sala-i-Martin (2004), set the rate of depreciation
at .05 and the rate of time preference at .02 per year. Let capital’s share be .66 [as
suggested by Obstfeld and Rogoff (1994)], which restricts the intertemporal
elasticity of substitution to the fairly low value of 1.5. The consumption-capital
ratio predicted by Equation (9) is then .055. This matches the data remarkably
well.
In conclusion, if we are willing to embrace a broad notion of capital and
entertain the possibility that the elasticity of substitution may be even a bit larger
than one, this calibration may work fairly well.
7 Conclusion
This paper provides an explicit solution to the Ramsey model when capital’s share
is equal to the reciprocal of the intertemporal elasticity of substitution. This is
admittedly a special case. However, it illuminates the dynamics of the model in a
way that approximations and numerical methods cannot: it helps us to “inspect
the mechanism” [Campbell (1994)]. In principle it could be used as a benchmark
from which to undertake perturbation analyses of less restrictive cases [Judd
(1997)]. In addition, it offers a useful pedagogical exercise, providing a closedform solution to the Ramsey corresponding to the Bernoulli
solution of the Solow
model.
In another paper [Smith (2005)] I derive an explicit solution to a stochastic
version of the Ramsey model. Applying the methods of Merton’s (1975) classic
paper on stochastic growth I can then derive an explicit solution for the ergodic
distribution of the capital stock. The nonlinearities of the model (in both the utility
and production functions) interact with uncertainty — via Jensen’s inequality — to
affect the steady state in complicated and even surprising ways. The presence of
uncertainty makes the “mechanism” much more complicated than generally
appreciated.
Appendix
DERIVATION OF THE GENERAL SOLUTION FOR z
Consider first the homogeneous equation
( )
1
z
= – -α δ + .
1
z
xAe
x
xt
+ (A.1)
This can be integrated to find the complementary solution
z c = x α -1e -(1-α )(x+δ )t (1 + xAe xt )1-α . (A.2)
To find the particular solution to Equation (12), I will use the method of
variation of parameters. Conjecture that the particular solution is
z p = zcΨ, where Ψ is an unknown function of time. Substituting this conjecture
into Equation (12), it follows that
Ψ = 1 – α = (1 – α ) 1-α (1-α )(x+δ )t (1 + xt )α -1.
c
x e xAe
z
(A.3)
Integrating Equation (A.3) leads to1
( )( ) ( )( ) ( )( )
; .
1
1 1 2 1 1 , 1 ,1
–
– +
+
– +
–
+
Ψ = – +
–
x t xAe xt
x
x
x
x
e F
xx
α δ α δ
α
δ
α δ
α
(A.4)
It follows that
( ) ( )( ) ( )( )
; .
1
1 1 2 1 1 , 1 ,1
–
– +
+
– +
–
+
+
=
–
xt
xt
p xAe
x
x
x
x
F
x
xAe
z
α δ α δ
α
δ
α
(A.5)
However, a useful property of the hypergeometric function [Weisstein (2004)] is
that ( ) ( )
–
= – –
–
1
2 1 , , ; 1 2 1 , , ,
z
z
F a b c z z a F a c b c . Therefore Equation (A.4) can
be written as
.
1
1 2 1 1 ,1, ;
z p = x +δ F -α d +xAe xAe xt xt (A.6)
where d = 1+ (1-α)(δ + x) x . To verify that this is a solution; differentiate
Equation (A.6) with respect to time, using the definition of the hypergeometric
function given by Equation (14) of the text:
( )
( )( )
– –
+
–
1 2
1 2
2
α α α
x Ae
xt
xt =
1
z
p
( )
( )( )( )
( )( )
( ) ( )( )
( )
( )( )( )
( )( )
+
+ +
– – –
+
+
– –
+
–
–
+
=
+
+ +
– – –
+
+
+ +
. . .
1 2
1 2 3
3
1
1 2
1 1 2
. . .
1 2
1 2 3
3
1
1
2
2
2
ω
α α α
ω
α α α
ω ω
δ
ω
α α α
ω
δ
x d d d d d d
x
xAe d d d d d d
x
(A.7)
Substitute this back into Equation (12) in the text to find
( ) ( )( )
( )
( )( )( )
( )( )
( ) [ ( )] ( )( )
( )
( )( )( )
( 1)( 2) . . .
1 2 3
1
1 1 2
1 1 1 1
. . .
1 2
1 2 3
3
1
1 2
1 1 2
2 3
2
+
+ +
– – –
+
+
– –
+
–
+
+
+ –
– –
=
+
+ +
– – –
+
+
– –
+
–
–
+
ω
α α α
ω
α α
ω
α
ω δ
δ
α
ω
α α α
ω
α α α
ω ω
δ
x d d d d d d
x
x d d d d d d
x
(A.8)
Collect the terms inω i (i = 0,1,2,…). Some tedious algebra reveals that the
coefficient attached to each ω i in the left hand side of Equation (A.8) equals the
coefficient attached to the corresponding ω i on the right hand side. Therefore
Equation (A.6) solves Equation (14) in the text.
Using Equations (A.2) and (A.6) leads to the general solution
α ( α )( δ ) α ( α )( δ ) ( ) α
δ
x + x 1+xAe
Bx e xAe
– – – + –
+ +
z = 1 2 F1 1- ,1,1+ 1- x + ;xAe xt xt 11 x t 1 xt 1
(A.9)
where B is an arbitrary constant.
PROOF OF THE LEMMA
Using the initial condition k0 and Equations (13) and (14), it follows that
( )
+
–
+
= –
–
xA
xA
F d
x
B A x z
1
1 0 1 2 1 1 α,,1, ;
δ
α
. (A.10)
Now consider the transversality condition in Equation (3) of the text. Recalling
that σ =1 α , use Equations (11) and (13) to write this limit as
( )
( )( )
1 1 1
– – – +
x t
α α δ lim 1 1 ,1; ;
–
xAe
F a d
xt
α
1
+
xAe 2
+
x
δ 1
+
xAe
xt (1 ) .
+
–
Ae
xt
α + xt
→ ∞
– t
xx e
B A
x
e
t
ρ
(A.11)
We need to show that this limit is zero if and only if A = 0.
First, suppose that A = 0. Note that, from Equation (14),
(0) 1 1 .
= –
–
B x z (A.12)
0 +
δ
α
x
Use this fact to evaluate the limit in Equation (A.11) when A = 0 :
( )( )
0
1 1
lim 1 1
0
=
→ ∞
+
+ –
+
– – – – +
t
e
x
k
x
e
ρt α α δ x t
δ δ (A.13)
This establishes sufficiency.
To establish necessity first rewrite Equation (A.11) as
t ( xt ) xt xt ( ) (
xAe e
– + – + x xAe δ → ∞
α ρ
–
xt )[ ( )(x )]t
AB
t
xAe
e xAe F a d
x
x
ρ α δ
+ +
lim + 1+ 2 1 ,1; ;1+ 1 1
(A.14)
Now consider the limit of the second term:
( )( ) [ ( )( )]
.
lim 1 1
→ ∞
+ – + – +
t
xAe e
Ax
B
xt ρ α x δ t
(A.15)
Since ρ + (1-α)(δ + x) > 0 and – ρ + x -(1-α)(δ + x) = 0 ,
( )( ) [ ( )( )] ( )
.
lim 1 1
→ ∞
+ – + – + =
t
xAe e AB A
Ax
B
xt ρ α x δ t
(A.16)
If B(A) ≠ 0 this will be zero if and only if A = 0.
Next consider the first expression in Equation (A.14):
( )
1 .
1 2 1 ,1; ;
++ +
xAe
xAe
e xAe F a d lim
–
ρt xt
xt α
→ ∞
–
t
xx
xt
α
δ (A.17)
First consider the limit of the hypergeometric function. Note that
lim 1 =1.
→ ∞
+
t
xAe
xAe
xt
xt
(A.18)
[Lebedev (1972, p. 244, Equation (9.3.1))] provides a limit result about
hypergeometric functions which then establishes
( ) ( )
( ) ( )
→ ∞
= Θ >
Γ – + Γ –
Γ Γ – +
=
+
t
d d
d d
xAe
xAe
F a d
xt
xt
0.
1 1
2
1
lim 2 1 ,1; ;
α
α
(A.19)
Since the hypergeometric function converges to this finite limit as t → ∞ it follows
that
( )
=
++ +
xAe
xAe
e xAe F a d
xt
t xt
lim
1 2 1 ,1; ;
( )
+ Θ
+
– –
e xAe
t xt
1 .
ρ α
α – xt lim α
→ ∞
→ ∞
– t
xx
t
xx
1
ρ
α
δ δ
(A.20)
Now rewrite this as
( ) ( ) ( ) ( )
→ ∞
t → ∞
lim
Θ
+
+ Θ =
+
–
–
– –
–
e xA
xx
t
e e xA
xx
x t xt x t
lim α ρ α .
α
α
α ρ
α
δ δ (A.21)
This will go to zero if and only if A = 0. This establishes necessity.
Given A = 0, B is given by Equation (A.12). This proves the lemma.
SOLUTION FOR THE MODEL WITH A TEMPORARY INCREASE IN
GOVERNMENT EXPENDITURES
The dynamics of x and z are given by
( )
( )
, ;
1
1 , 0 ,
1 2
g x t T
g x t T
xx
+ + >
– +
= –
+ + ≤ ≤
– +
= –
α
δ α ρ
α
δ α ρ
(A.22)
( )[ ( ) ]
(1 )[1 ( ) ], .
1 1 , 0 ,
2
1
g x z t T
z g x z t T
= – – + + >
= – – + + ≤ ≤
α δ
α δ
(A.23)
First, define xi = +( – ) – g i
α
ρ 1 α δ
, and =1+ (1- )( + ), i =1,2.
x
x
d
i
i
α δ
Let
0.
∆g = g1 – g 2 > Note in passing that
x + g = x + g = x =
1 α
1 2 2 1 .
()ρ +δ -α
Following the derivation of the basic model without government, the general
solutions for x and z are
, ;
1
, 0 ,
1
1 2
2 2
2
1 1
1
t T
x A e
x
t T
x A e
x
x
x t
x t
>
+
=
≤ ≤
+
=
(A.24)
( )( ) ( )
( )( ) (1 ) , .
1
1 1 ,1, ;
1 , 0 ,
1
1 1 ,1, ;
1
2 2
1 1
2 2
2 2
2 2
2 1 2
1
1 1
1 1
1 1
1 1
1 1
2 1 1
21
2
1
1
1
B x e x A e t T
x A e
x A e
F d
x
B x e x A e t T
x A e
x A e
F d
x
z
x t x t
xt
x t
x t x t
x t
x t
+ + >
+
–
+
=
+ + ≤ ≤
+
–
+
=
–
– – – +
– – – + –
α α δ α
α α δ α
α
δ
α
δ
(A.25)
There are four arbitrary constants, A1, A2 , B1, and B2. To pin down these
constants, we have four boundary conditions:
(a)
(b)
(c)
(d) the initial capital stock, k0,
the transversality condition [Equation (3) in the text],
continuity of c at T,
continuity of k at T.
Consider the A’ s first. Using the argument developed in the previous
appendix, the transversality condition (b) requires A2 = 0. Using the continuity
condition (c) along with Equation (A.24) and the fact that A2 = 0, we find
.
1
1 2
x T
e
x x
g
∆ – (A.26) 1
A = –
To find
B1, substitute Equation (A.26) into Equation (A.25) evaluated at t = 0 and
using the initial condition (a), z0 = k01-α :
∆
∆ –
– –
–
F d
z
0 2 1 1
1 1 α,1, ; = – ∆
– –
x T
e
x
g
B x
2
11
1
1 1
α –
xT (A.27)
+
x
δ 1
–
–
xT
e
x
e g
x
g
2
2
1
α
This implies, from Equation (A.25), that for the interval 0 ≤ t ≤ T
( )
( )
( )
( )( )
.
1
1
1
1 1 ,1, ;
1
1 1 ,1, ;
1
1
2
2
2
2
0 2 1 1
2
2
2 1 1
1
1
1
1
x t
x T
x t T
xT
xT
x t T
x t T
e
e
x
e g
x
g
e
x
e g
x
g
F d
x
z
e
x
e g
x
g
F d
x
z
α δ
α
α
δ
α
δ
– – +
–
–
–
–
–
–
–
∆
∆ –
–
∆
∆ –
– –
+
–
+
∆
∆ –
– –
+
=
(A.28)
Note that
( )( )
.
1
1
1
1 1 ,1, ;
1
1 1 ,1, ;
1
1
2
2
2
2
0 2 1 1
2
2
2 1 1
1
x T
xT x T
xT
T
e
e
x
g
x
g
e
x
e g
x
g
F d
x
z
x
g
x
g
F d
x
z
α δ
α
α
α δ
δ
– – +
–
– –
–
∆
–
∆
–
∆
∆ –
– –
+
–
+
∆
∆ –
– –
+
=
(A.29)
Finally, to find B2 substitute Equation (A.29) to the second interval (for t > T ) of
Equation (A.26), evaluated at t = T. Recalling that z =1/(x +δ )this yields
B2 = x21-α (zT – z)e(1-α )(x+δ )T
Substituting this back into Equation (A.26) yields z for t > T :
z z (zT z)e (1)(x)(tT ). (A.30)
= + – –α +δ – (A.31)
SOLUTION FOR THE ENDOGENOUS GROWTH MODEL
Assuming thatσ = 1 α , Equations (28) and (29) in the text imply that the
dynamics of z and x are
z = (1-α )[1+ (a -δ – x)z]
(a )(1 ) x.
= – -δ -σ – ρσ + (A.32)
(A.33)
xx
I assume that a > δ + ρ in order to generate long-term growth.
Begin with the logistic equation (A.33). This has the general solution
,
1 ˆ
ˆ
x x
= (A.34)
xAe xˆt
+
where ( )( )
α
ρ – -δ -α
=
1
xˆ a . The transversality condition will require that xˆ > 0.
Substitute Equation (A.34) back into the linear equation (A.32):
( )
z
+
= – + – – xAe
x
z a
1 ˆ xˆt
ˆ
1 α 1 δ . (A.35)
Following the derivation of Equation (A.9), the general solution to Equation
(A.35) is
α ( α )( δ ) α ( α )( δ ) ( ) α
δ
xˆ + – a x 1+xˆAe xt
2 Bxˆ e 1 xˆAe
–
– – – + –
+ +
z = 1 F1 1- ,1,1+ 1- xˆˆ+ – a ;xˆAe xˆt ˆ11 xˆ a t xt 1
.
(A.36)
The initial condition
z0 requires
( )
1
ˆ
1
= –
–
B A x z
α ,
ˆ
1 ,,1, ˆ;
–
xA
F d
α
1 ˆ
+
xA ˆ
0 2 1
+ –
x a
δ
(A.37)
where ( )( ).
ˆ
1 ˆ
ˆ 1
x
x a
d = + -α +δ – The transversality condition requires
( )
( )( )
ˆ
ˆ
,1; ˆ;
1
ˆ 1
1 1 ˆ
ˆ
ˆ
2 1
– – – + –
x e
B A
xt
x a t
xt
xt
(1 ˆ ) .
+
–
xAe 1 ˆ
+
+
xAe ˆ
+ –
x a
δ 1 ˆ
ˆ
+
Ae
ˆ lim
e –
xt
α xAe
F a d α α
→ ∞
– t
xx
t
α
δ
ρ
(A.38)
Following the reasoning in the previous lemma, the transversality condition will
be satisfied if and only if (1) A = 0 and (1) ρ > (1-α)(a -δ ). The latter condition
implies that xˆ > 0, which yields Equation (30) in the text.
Setting A = 0 , Equation (A.37) yields
.
ˆ
1
ˆ
0
1
+ –
= –
–
x a
B x z
δ
α (A.39)
Substituting this into Equation (A.36) yields
( )( )
( )
( )
.
ˆ
1
ˆ
1
1
0
1 ˆ
0
a t
x a t
e
a
z
a
e
x a
z
x a
z
δ ρ
δ α α
α
δ ρ
α
δ ρ
α
δ δ
– –
–
– – + –
= – – – + + – –
+ –
+ –
+ –
=
(A.40)
Substituting z = k 1-α into Equation (A.40) yields Equation (31) in the text.
References
Attanasio, Orazio and Guglielmo Weber (1992) “Intertemporal Substitution, Risk
Aversion, and the Euler Equation for Consumption,” Economic Journal, 99,
59-73.
Attanasio, Orazio and Guglielmo Weber (1993) “Consumption Growth, the
Interest Rate, and Aggregation,” Review of Economic Studies, 60, 411-445.
Barro, Robert, N. Gregory Mankiw, and Xavier Sala-i-Martin (1995) “Capital
Mobility in Neoclassical Models of Growth,” American Economic Review 85,
103-115.
Barro, Robert and Xavier Sala-i-Martin (2004) Economic Growth, 2nd edition,
M.I.T. Press, Cambridge.
Benhabib, Jess and Aldo Rustichini (1994) “A Note on a Class of Solutions to
Dynamic Programming Problems Arising in Economic Growth,” Journal of
Economic Dynamics and Control, 18, 808-813.
Blanchard, Olivier and Stanley Fischer (1989) Lectures on Macroeconomics,
M.I.T. Press, Cambridge.
Boucekkine, Rauof and José Ramoń Ruiz-Tamarit (2004a) “Imbalance Effects in
the Lucas Model: an Analytical Exploration,” BE Contributions to
Macroeconomics vol. 4 (1), Article 15:
http://www.bepress.com/bejm/topics/vol14/iss1/art15l.
Boucekkine, Rauof and José Ramoń Ruiz-Tamarit (2004b) “Special Functions for
the Study of Economic Dynamics: the Case of the Lucas-Uzawa Model,”
Working Paper.
Bufman, Gil and Leonardo Leiderman (1990) “Consumption and Asset Returns
Under Non-Expected Utility,” Economics Letters, 34, 231-235.
Campbell, John Y. (1994) “Inspecting the Mechanism: An Analytical Approach to the
Stochastic Growth Model,” Journal of Monetary Economics 33, 463-506.
Cass, David (1965) “Optimum Growth in an Aggregative Model of Capital
Accumulation,” Review of Economic Studies, 32, 233-240.
Chang, Fwu-Rang (1988) “The Inverse Optimal Problem: A Dynamic
Programming Approach,” Econometrica 56, 147-172.
Chiang, Alpha (1967) Fundamental Methods of Mathematical Economics,
McGraw-Hill, New York.
Epstein, Larry and Stanley Zin (1991) “Substitution, Risk Aversion, and the
Temporal Behavior of Consumption and Asset Returns: An Empirical
Analysis,” Journal of Political Economy, 9, 263-268.
Hall, Robert (1988) “Intertemporal Substitution in Consumption” Journal of
Political Economy, 96, 339-357.
Kaldor, Nicholas (1963) “Capital Accumulation and Economic Growth,” in
Proceedings of a Conference Held by the International Economic Association,
F.A. Lutz, D.C. Hague, eds. MacMillan, London.
Jones, Larry and Rodolpho Manuelli (1990) “A Convex Model of Equilibrium
Growth: Theory and Policy Implications,” Journal of Political Economy, 98,
1008-1038.
Jorgenson, Dale and Barbara Fraumeni (1989) “The Accumulation of Human and
Non-human Capital,” in The Measurement of Saving, Investment, and Wealth,
R. Lipsey, H. Tice, eds. University of Chicago Press, Chicago.
Judd, Kenneth L. (1997) “Computational Economics and Economic Theory:
Substitutes or Complements?” Journal of Economic Dynamics and Control
21(6), 907-942.
Koopmans, Tjalling (1968) “On the Concept of Economic Development,” in The
Economic Approach to Development Planning. Elsevier, Amsterdam.
Kurz, Mordechai (1968) “The General Instability of a Class of Competitive
Growth Models,” Review of Economic Studies 35, 155-174.
Kydland, Fynn and Edward Prescott (1982) “Time to Build and Aggregate
Fluctuations,” Econometrica, 50, 1345-1370.
Lebedev, Nikolai Nikolaevich (1972) Special Functions and Their Applications,
translated and edited by R. Silverman. Dover Publications, New York.
Lettau, Martin and Stanley Ludvigson (2004) “Understanding Trend and Cycle in
Asset Values: Reevaluating the Wealth Effect in Consumption,” American
Economic Review 94 276-299.
Long, John B. and Charles Plosser (1983) “Real Business Cycles,” Journal of
Political Economy, 91(1), 36–69.
Lucas, Robert E. (1988) “On the Mechanics of Economic Development,” Journal
of Monetary Economics, 22, 3-42.
Maddison, Angus (1982) Phases of Capitalist Development. Oxford University
Press, Oxford.
Mankiw, N. Gregory, David Romer, and David Weil (1992) “A Contribution to
the Empirics of Economic Growth,” Quarterly Journal of Economics 107,
407-437.
Mehlum, Halvor (2005) “A Closed Form Ramsey Saddle Path,” BE Contributions
to Macroeconomics 5 (1), Article 2:
http://www.bepress.com/bejm/contributions/vol15/iss1/art2l.
Merton, Robert (1975) “An Asymptotic Theory of Growth Under Uncertainty,”
Review of Economic Studies 42, 375-393.
Mulligan, Casey B. and Xavier Sala-i-Martin (1991) “A Note on the TimeElimination Method for Solving Recursive Economic Models.” Technical
Working Paper no. 116, NBER.
Ramsey, Frank P. (1928) “A Mathematical Theory of Saving,” Economic Journal
38 543-559.
Romer, David (2005) Advanced Macroeconomics, 3rd edition, McGraw-Hill,
Boston.
Romer, Paul (1986) “Increasing Returns and Long Run Growth,” Journal of
Political Economy 1002-1103.
Simon, Julian L (1990) “Great and Almost Great Magnitudes in Economics,”
Journal of Economic Perspectives 4, 149-156.
Solow, Robert M (1956) “A Contribution to the Theory of Economic Growth,”
Quarterly Journal of Economics 70, 65-94.
Smith, William T. (1996) “Taxes, Uncertainty, and Long Run Growth,” European
Economic Review 40 (8), 1647-1664.
Smith, William T. (2005) “Inspecting the Mechanism Exactly: A Closed Form
Solution to the Stochastic Growth Model,” Working Paper.
Turnovsky, Stephen J (1999) International Macroeconomic Dynamics, 2nd
edition, M.I.T. Press, Cambridge.
Turnovsky, Stephen J. (2000) Methods of Macroeconomic Dynamics, 2nd edition,
M.I.T. Press, Cambridge.
Vissing-Joregensen, Annette and Orazio Attanasio (2003) “Stock-Market
Participation, Intertemporal Substitution, and Risk Aversion,” American
Economic Review Papers and Proceedings 93, 383-391.
Wälde, Klaus (2005) “Endogenous Growth Cycles,” forthcoming International
Economic Review.
Weisstein, Eric (2005) “Hypergeometric Functions,” From MathWorld – a
Wolfram Web Resource.
http://Mathworld.wolfram.com/HypergeometricFunctions.html.
Xie, Danyang (1991), “Increasing Returns and Increasing Rates of Growth,”
Journal of Political Economy 99, 429-435.
Xie, Danyang (1994), “Divergence in Economic Performance: Transitional
Dynamics with Multiple Equilibria,” Journal of Economic Theory 63, 97-112.
Zhang, Qiang (2004) “Accounting for Human Capital and Weak Identification in
Evaluating the Epstein-Zin-Weil Non-Expected Utility Model of Asset
Pricing,” Working Paper, University of Memphis.
View publication stats
