Advanced Macroeconomics

Chapter 1: Exogenous Growth and accumulation of

Knowledge
Christian Ghiglino

October 8, 2009

Contents
1 Introduction 3
1.1 Fundamental economic issues . . . . . . . . . . . . . . . . . . 3
1.2 The Kaldor stylized facts and more . . . . . . . . . . . . . . . 3
1.3 What have been found (Sala–i-Martin (2002)) . . . . . . . . . 3

2 The Solow Model 4

2.1 Main assumptions . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 The equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 The dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.5 Comparison with the facts . . . . . . . . . . . . . . . . . . . . 5
2.6 The Solow model with exogenous technological progress . . . 6
2.7 The Solow model in discrete time . . . . . . . . . . . . . . . . 6

2.7.1 Stability of the steady state . . . . . . . . . . . . . . . 7

2.7.2 Global properties . . . . . . . . . . . . . . . . . . . . . 7

3 The Ramsey Model 7

3.1 Main equations . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.2 Phase diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 8

4 Neoclassical two-sector model 9

4.1 Main equations . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4.2 The firms’ problem . . . . . . . . . . . . . . . . . . . . . . . . 9
4.3 The consumer’s problem . . . . . . . . . . . . . . . . . . . . . 10
4.4 Specific formulation of the model . . . . . . . . . . . . . . . . 11

1
5 Diamond Model 12
5.1 The consumer’s problem . . . . . . . . . . . . . . . . . . . . . 12
5.2 The firm’s problem . . . . . . . . . . . . . . . . . . . . . . . . 13
5.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

6 Knowledge accumulation without capital 14

6.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
6.2 Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
6.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
6.3.1 Case 1: θ < 1 . . . . . . . . . . . . . . . . . . . . . . . 16
6.3.2 Case 2: θ = 1 . . . . . . . . . . . . . . . . . . . . . . . 17
6.3.3 Case 3: θ > 1 . . . . . . . . . . . . . . . . . . . . . . . 18

7 Knowledge accumulation with capital 19

7.1 Main equations . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.2 decreasing returns: β + θ < 1 . . . . . . . . . . . . . . . . . . 19
7.3 constant returns: β + θ = 1 . . . . . . . . . . . . . . . . . . . 19
7.4 increasing returns: β + θ > 1 . . . . . . . . . . . . . . . . . . 19

8 Learning by Doing 20
8.1 Main equations . . . . . . . . . . . . . . . . . . . . . . . . . . 20
8.2 Case 1: θ < 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
8.3 Case 2: θ = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
8.3.1 n=0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
8.3.2 n>0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
8.4 Case 3: θ > 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

9 Human Capital 24
9.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
9.2 Main equations . . . . . . . . . . . . . . . . . . . . . . . . . . 24
9.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
9.3.1 The steady state . . . . . . . . . . . . . . . . . . . . . 25
9.3.2 The phase diagram . . . . . . . . . . . . . . . . . . . . 26

9.3.3 Quantitative analysis . . . . . . . . . . . . . . . . . . . 27

9.3.4 An application . . . . . . . . . . . . . . . . . . . . . . 28

10 Malthus 30

2
1 Introduction
1.1 Fundamental economic issues
• What is the engine of long-run growth? Why is there a constant growth
rate? How to stimulate growth?
• Why is there structural change? Why the importance of services rise
so sharply?
• Why are there inequalities across countries? How to reduce them?
• Why are there short-run fluctuations? Are these bad? How to reduce
them?
• Why is there unemployment? How to eliminate it?

1.2 The Kaldor stylized facts and more

• Continuing growth in output per worker (balanced growth, constant
growth rate)
• Continuing growth in capital per worker
• The capital-output ratio is constant
• The rate of return to capital is constant
• The shares of labor and capital in national income are constant
• Systematic changes in the relative importance of various sectors (agri-
culture, manufacturing, services)
• Labor force in R & D follows a balanced growth path
• Total number of Patents in the R & D sector is constant
• Large differences in productivity growth rates across countries

1.3 What have been found (Sala–i-Martin (2002))

There is no good theory!!!
• There is no simple determinant for growth
• The initial level of income is the most determinant and robust variable
• The size of the government does not matter but its quality yes
• Relation between human capital and growth is weak
• Institutions are important for growth
• More open economies growth faster

3
2 The Solow Model
This is the standard model of capital accumulation. Without assuming exo-
geneous technological progress, this model predicts convergence to a steady
state in per capita output. The driving force behind this result is the fact
that there are diminsihung returns in capital, the only accumulable input.

2.1 Main assumptions

• There is one aggregate good which can be consumed or invested to
accrue capital.

• There are two inputs, labor and capital.

• There is constant returns to scale (possibility of replication).

• Capital depreciates at the constant rate δ.

• Total population and total workforce (assumed to be equal) have a

constant growth rate of n.

• The saving rate is exogenous, s.

2.2 The equations

Let K(t) and L(t) be the total capital and labor and Y (t) be total output
in period t. Let C(t) be total consumption. Let F be the constant returns
production function. The main equations of the Solow model are:
 
Y (t) = F K(t), L(t) (1)

Y (t) = C(t) + K̇(t) + δK(t) (2)

 
sF K(t), L(t) = K̇(t) + δK(t) (3)

L̇(t) = nL(t) (4)

Equation (3) can be expressed in per capita terms:
 
k̇(t) = sf k(t) − (δ + n)k(t) (5)

At a steady state k̇(t) = 0, so k ∗ is implicitly given by

sf (k ∗ ) = (n + δ)k ∗ (6)

4
2.3 The dynamics
To focus the analysis on a steady state is relevant only if this is dynamically
stable so that this equilibrium is robust to shocks. The diagrammatic anal-
ysis of the dynamic equation shows that k ∗ is globally dynamically stable.
The economy converges to k ∗ independently of the initial value k(0).

2.4 Results
1. y(t), k(t) and c(t) converge to their steady states values.

2. Y (t), K(t) and C(t) grow with the population growth rate n.

3. The speed of convergence is ....

2.5 Comparison with the facts

1. The Solow model do not explain well the differences across countries.
Indeed, in order to explain the differences income level the differences
in capital per capita and in the interest rate across countries need to
very large. This is not compatible with the observed pattern.

2. The model do not allow for fluctuations without exogenous shocks to

the production function f(k)

5
2.6 The Solow model with exogenous technological progress

Let A(t) be the state of knowledge. It is assumed to be labour augmenting,

so Lef f (t) = A(t)L(t). Intensive variables are defined through normalizing
by A(t)L(t). Let F be the constant returns production function. The main
equations of the model are:

Y (t)  
y(t) = = F k(t), 1) = f (k(t)) (7)
A(t)L(t)

We assume
•
A(t) = gA(t) (8)
leading to
A(t) = A(0)egt
The dynamics in intensive form is:
 
k̇(t) = sf k(t) − (g + δ + n)k(t) (9)

At a steady state k̇(t) = 0, so k ∗ is implicitly given by

sf (k ∗ ) = (g + n + δ)k ∗ (10)

2.7 The Solow model in discrete time

Most economic models can be formulated in both continuous and discrete

time. The choice is usually made considering the solvability of the model.
Economically, discrete time makes more sense in real macro models because
macro data is usually provided by discrete time series with large time peri-
ods. On the other hand, financial data is usually so dense that a continuous
time formulation is more appropriated.

Let Kt and Lt be the total capital and labor and Yt be total output
in period t. Let Ct be total consumption. Let F be the constant returns
production function. The main equations of the Solow model are:
 
Yt = F Kt , Lt (11)

Yt = Ct + Kt+1 − Kt + δKt (12)

 
sF Kt , Lt = Kt+1 − Kt + δKt (13)

Lt+1 = (1 + n)Lt (14)

6
Equation (3) can be expressed in per capita terms:
 
kt+1 = sf k(t) + (1 − δ − n)kt (15)

At a steady state kt+1 = kt = k ∗ , so k ∗ is implicitly given by

sf (k ∗ ) = (n + δ)k ∗ (16)

The value of the steady state (and the other properties) depends on the
properties of the function f. What do we know about f ?
The function f inherits some properties of the production function F. In
particular, F is assumed to be constant returns to scale, increasing in both
arguments and with infinite marginal productivity at 0. A good example is
F (K, L) = K α L1−α with 0 < α < 1. The per capita form is f (k) = k α which
is concave with infinite marginal productivity at zero and asymptotically
zero.

2.7.1 Stability of the steady state

A steady state is economically relevant only if it is robust to perturbations,
i.e. locally stable (The Samuelson principle). There are several ways to
analyze the local stability properties of a steady state: using the phase
diagram, computing the eigenvalues of the Jacobian matrix at the steady
state, and by hand kt+1 < kt ).

2.7.2 Global properties

Globally stable.

3 The Ramsey Model

The structure of the Ramsey model is as for the Solow model but the saving
rate is endogenous. This means that the behavior of the consumer should
be explicitly modeled as an optimization.

3.1 Main equations

Yt = F (Kt , Lt ) (17)
It = sYt = Kt+1 − Kt + δKt (18)
Yt = It + Ct (19)

7
 Kt+1
Substitute (??) into (??), divide by Lt , use the fact that kt+1 = Lt+1 =
Kt+1 Lt Kt+1 1
Lt Lt+1 = Lt 1+µ , rearrange and you get capital per capita:

sf (kt ) + (1 − δ)kt f (kt ) − ct + (1 − δ)kt

kt+1 = = (20)
1+µ 1+µ

Rearrange (??) and you get consumption per capita:

ct = f (kt ) − (1 + µ)kt+1 + (1 − δ)kt (21)

The maximization problem of the representative consumer is given by:

∞
X
M ax σ t U (ct ) s.t. (??) (22)
t=0

Substitute (??) into the utility function of (??), take the derivations with
respect to kt and you get the Euler equations:

U 0 (ct ) 1+µ
0
=   ∀t ≥ 1 (23)
U (ct−1 ) σ f 0 (kt ) + (1 − δ)

3.2 Phase diagram

The dynamics of capital per capita (kt ) and consumption per capita (ct ) is
described by equations (??) and (??)

∆k=0: If kt = kt+1 := kt , then we have with (??):

ct = f (kt ) − (δ + µ)kt (24)

∆c=0: If ct = ct+1 , then we have with (??):

1+µ
f 0 (kt ) = − (1 − δ) (25)
σ
Equation (25) determines the level of capital per capita, where consumption
is constant. We denote this level by k∆c=0

ct > f (kt ) − (δ + µ)kt : ct > f (kt ) − (δ + µ)kt . Substitute (??) for ct and
you get: f (kt ) − (1 + µ)kt+1 + (1 − δ)kt > f (kt ) − (δ + µ)kt . Rearrange and
you finally get: kt+1 − kt < 0.
We therefore have the following dynamics:

ct > f (kt ) − (δ + µ)kt ⇔ kt+1 − kt < 0 (26)

8
kt > k∆c=0 : If kt > k∆c=0 , then we have with (25) and because f 00 (kt ) < 0:
f 0 (kt ) < 1+µ
σ − (1 − δ). Rearrange and use (??) to get:
0
0<  1+µ  = U0 (ct ) . Rearrange and use that U 00 (ct ) < 0 and you
U (ct−1 )
σ f 0 (kt )+(1−δ)
finally get: ct − ct−1 < 0.
We therefore have the following dynamics:
1+µ
kt > k∆c=0 ⇔ f 0 (kt ) < − (1 − δ) ⇔ ct − ct−1 < 0 (27)
σ

4 Neoclassical two-sector model

This is the standard two-sectors model in which the two sectors are fully
specialized. The model is important as it shows that endogenous cycles
without shocks may persist.

4.1 Main equations

In this model, there are two goods, a consumption good ct and an investment
good kt . There is an investment sector and a consumption sector.
The main equations are given by:

ct = F 1 (kt1 , lt1 ) (28)

kt+1 = F (kt2 , lt2 )
2
(29)
kt = kt1 + kt2 (30)
l : = lt = lt1 + lt2 (31)

4.2 The firms’ problem

The profit maximization problem of the firm in the consumption sector is
given by:
M ax p1t F 1 (kt1 , lt1 ) − p2t−1 kt1 − wt1 lt1 (32)
The first-order conditions to problem (32) are given by:
∂F 1 (kt1 , lt1 )
p1t = p2t−1 (33)
∂kt1
∂F 1 (kt1 , lt1 )
p1t = wt1 (34)
∂lt1
The maximization problem of the firm in the investment sector is given by:

M ax p2t F 2 (kt2 , lt2 ) − p2t−1 kt2 − wt2 lt2 (35)

9
The first-order conditions to problem (35) are given by:
∂F 2 (kt2 , lt2
p2t = p2t−1 (36)
∂kt2
∂F (kt2 , lt2
2
p2t = wt2 (37)
∂lt2
Assuming that labor can move from one sector to the other, we have fur-
thermore:
wt1 = wt2 (38)
The profit maximization gives the optimal value of inputs (kt1 , kt2 , lt1 , lt2 ) as
a function of prices.

4.3 The consumer’s problem

The consumer’s utility maximization problem is given by:
∞
X
M ax β t U (ct ) s.t. (28) − (31), k0 given (39)
t=0

Define the transformation function T : R2 → R that gives the maximal

consumption for next period, given initial capital and investment:
T (kt , kt+1 ) := M ax F 1 (kt1 , lt1 ) s.t. (29) − (31) (40)
Then, problem (39) is equivalent to the following problem:
X∞  
M ax β t U T (kt , kt + 1) s.t. k0 given
t=0
∞
X
: = M ax β t V (kt , kt + 1) s.t. k0 given (41)
t=0

The first-order conditions to problem (41) are given by the Euler equations:
   
U 0 T (kt , kt+1 ) T2 (kt , kt+1 ) + βU 0 T (kt+1 , kt+2 ) T1 (kt+1 , kt+2 ) = 0
V2 (kt , kt+1 ) + βV1 (kt+1 , kt+2 ) = (42)
0
The Transversality condition is given by:
lim β t kt V1 (kt , kt+1 ) = 0 (43)
t→∞

Given prices, (42) and (43) determine the path of investment kt+1 and there-
fore also the path of consumption ct . Because we consider only one homo-
geneous and representative consumer, prices could be excluded from the
maximization problem (the equilibria are Pareto-efficient in this case). At
the steady state, we have kt+1 = kt := k∗ and the Euler equation is given
by:
T2 (k∗, k∗) + βT1 (k∗, k∗) = 0 (44)

10
4.4 Specific formulation of the model
Consider the following specification of the neoclassical two-sector model:
F 1 (kt1 , lt1 ) = (kt1 )1−α (lt1 )α (45)
k2
F 2 (kt2 , lt2 ) = min(lt2 , t ) (46)
γ
c1−µ
U 0 (ct ) = (47)
1−µ
Because we have a Leontief production function in the investment sector,
we must have at the optimum:
kt2
lt2 = (48)
γ
Together with (29) and (46) we must have furthermore:
kt 2
kt+1 = lt2 = (49)
γ
With (40) and (45), the transformation function is given by:
T (kt , kt+1 ) = M ax (kt1 )1−α (lt1 )α s.t. : (50)
k2
1
kt+1 2
+ kt+1 = kt+1 = lt2 = t , lt1 + lt2 = 1
γ
Substitute the conditions in (50) into the target function and you get:
T (kt , kt+1 ) = (kt − γkt+1 )1−α (1 − kt+1 )α (51)
T1 (kt+1 , kt+2 ) and T2 (kt , kt+1 ) are given by:
T1 (kt+1 , kt+2 ) = (1 − α)(1 − kt+2 )α (kt+1 − γkt+2 )−α (52)

T2 (kt , kt+1 ) = −α(1 − kt+1 )α−1 (kt − γkt+1 )1−α −

γ(1 − α)(1 − kt+1 )α (kt − γkt+1 )−α (53)
Substitute (52) and (53) into (44) and solve for k*:
(β − γ)(1 − α)
k∗ = (54)
(β − γ)(1 − α) + α(1 − γ)
If µ = 0 we have with (47): U (ct ) = ct and the Euler equations (42) become:
−α(1 − kt+1 )α−1 (kt − γkt+1 )1−α − γ(1 − α)(1 − kt+1 )α (kt − γkt+1 )−α
+β(1 − α)(1 − kt+2 )α (kt+1 − γkt+2 )−α (55)
= 0 (56)
(??) describes a one-dimensional, second-order dynamically system. This
can be rewritten into a two-dimensional, first-order dynamically system,
which can be studied with respect to cycles.

11
Theorem 1 Consider the following linear, homogeneous, 2-dimensional,
first-order dynamically system:
zt = Azt−1 , zt ∈ R2 , A ∈ M at(2 ∗ 2, R). Denote with λ1 , λ2 the eigenvalues
of A.
λ1 = −1 or λ2 = −1 ⇒ a cycle of period two exists. Assume α < (1−γ) 2 .
γ α+γ
Then there is a cycle of period 2 for β ∈ 1−2α , 1−α , e.g. for β in this
range: λ1 = −1 or λ2 = −1

5 Diamond Model
In the models investigated so far, agents were infinitly lived. Here we focus
the analysis on finitely lived agents. The standard model is the overlapping
generations model in which generations overlap. In the infinitly lived agents
models, agents can be considered as dynasties. These may be obtained by
assuming that the successive generations are linked through bequests. In
the overlapping generation models these links are broken.

5.1 The consumer’s problem

c1−θ
We assume that the instantaneous utility function is V (c) = 1−θ The con-
sumer’s maximization problem is given by:
M ax U (ct , ct+1 ) = M ax V (ct ) + δV (ct+1 )
c1−θ c1−θ 1
= M ax t
+ δ t+1 s.t. ct + ct+1 =(57)
ŵt
1−θ 1−θ 1 + rt+1
The first-order conditions to (57) are given by:
∂L
= c−θ
t +λ=0 (58)
∂ct
∂L λ
= δc−θ
t+1 + =0 (59)
∂ct+1 1 + rt+1
∂L 1
= ct + ct+1 − ŵt = 0 (60)
∂λ 1 + rt+1
Divide (58) by (59), solve for ct+1 , substitute the result into (60) and you
finally get:
1 1 −1
 
ct = ŵt 1 + [δ(1 + rt+1 )] θ (61)
1 + rt+1
1
ct+1 = ct [δ(1 + rt+1 )] θ (62)
If we define the saving rate implicitly by ct = (1 − st )ŵt , we get:
1−θ
(1 + rt+1 ) θ
st = −1 1−θ (63)
δ θ + (1 − rt+1 ) θ

12
5.2 The firm’s problem
We have exogenous growth of technology At :
At+1 = At (1 + g) (64)
Effective labor, L
e t , is given by:

L
e t = At L t (65)
Furthermore, wage is given by:
ŵt = At wt (66)
The firm’s profit maximization problem is then given by:
M ax F (Kt , At Lt ) − rt Kt − ŵt Lt (67)
With (66) problem (67) is equivalent to:
M ax F (Kt , At Lt ) − rt Kt − At wt Lt (68)
Kt
Divide (68) by At Lt , define capital per effective labor by kt = At Lt to get:
M ax f (kt ) − rt kt − wt (69)
The first-order conditions to problem (69) are given by:
f 0 (kt ) = rt (70)
wt = f (kt ) − kt f 0 (kt ) (71)

5.3 Analysis
Equilibrium on the capital market demands that investments are equal to
total savings for all t. In our model, the only way to save is to own capital.
Therefore equilibrium condition is given by:
Kt+1 = st rt+1 ŵt Lt = st rt+1 At wt Lt (72)
Divide (72) by At+1 Lt+1 to get this condition in per effective labor terms:
1
kt+1 = st rt+1 wt (73)
(1 + n)(1 + g)
Substitute (70) and (71) into (73) to get the following first-order dynamically
system:
1
kt+1 = st f 0 (kt )[f (kt ) − kt f 0 (kt )] (74)
(1 + n)(1 + g)
st in (74) is given by (63)
Theorem 2 When preferences are log and production is Cobb-Douglas, then
the steady state is unique and globally stable.

13
6 Knowledge accumulation without capital
In Section 2.6, the Solow model was extended introducing exogenous techo-
logical progress. The process was sponteneous without the need of any
investment and didn’t use any resources. It is now time to model techno-
logical progres. A first possibility is to assume that knowledge is produced
in the economy and specify the required technology to produce ideas. In
this Section it is assumed that only labour is needed while in the next both
capital and labour wil be used as input to produce knowledge.

6.1 Assumptions

The assumptions of the model are driven by two issues:

• What are the properties of knowledge?

Knowledge is a non-rival good, i.e. its use by a person does not make
its use by another person more difficult. Knowledge, may be excludable
(encoded TV program) or non-excludable (the Pythagoras theorem).

• Is the production function for knowledge constant returns in the phys-

ical (rival) inputs?

At least CRS is assumed in physical goods production because of the

possibility to replicate individual firms even when these have an optimal
size, i.e. returns become eventually decreasing. On the other hand, in-
creasing returns to scale are possible whenever externalities exist between
firms. Increasing returns within a firm are usually excluded. In the case of
knowledge, constant returns cannot really be assumed.

The structure of the model is:

• One physical output which is fully consumed

• One physical input, labour. No need of capital.

• A non-rival good, A, interpreted as knowledge.

• The production function for output is CRS in labour. The production

function for knowledge is arbitrary.

14
6.2 Equations
The main equations of the model are:
 
Y (t) = F AY (t), KY (t), LY (t) = (1 − aL )L(t)A(t) (75)
 γ
Ȧ(t) = B aL L(t) A(t)θ (76)

L̇(t) = nL(t) (77)

(77) is a first-order, linear and homogenous dynamical system. The solution
is given by:
L(t) = L(0)ent (78)
In order to analyze the dynamics, define the growth rate of knowledge with
equation (76):
Ȧ(t)  γ
gA (t) = = B aL L(t) A(t)θ−1 (79)
A(t)
Take the derivation of gA (t) with respect to time and you get

" • #
• Ȧ(t) •
gA = = γB[aL L(t)]γ−1 L(t)A(t)θ−1 + (80)
A(t)
•
+(θ − 1)B[aL L(t)]γ A(t)θ−2 A(t) (81)

so that
" • #
• Ȧ(t) •
gA = = γB[aL L(t)]γ−1 L(t)[L(t)/L(t)]A(t)θ−1 + (82)
A(t)
•
+(θ − 1)B[aL L(t)]γ A(t)θ−1 [A(t)/A(t)] (83)

and finally
• 2
gA (t) = γngA (t) + gA (t)(θ − 1) (84)
The steady state is obtained for
2
0 = γngA (t) + gA (t)(θ − 1) (85)

giving
γn ∗
gA = = gA (86)
1−θ

15
6.3 Analysis
6.3.1 Case 1: θ < 1
∗ where
If θ < 1 it follows from (84) that gA (t) converges to gA
∗ γn
gA = (87)
1−θ
•
∗ we have g (t) < 0 so that g (t) decreases while
Indeed, for gA (t) > gA A A
•
∗ we have g (t) > 0 so that g (t) increases. Equation (87) de-
gA (t) < gA A A
scribes a one-dimension, first-order, linear and homogeneous dynamically
system. In other words, the equation
Ȧ(t) ∗ γn
= gA =
A(t) 1−θ
has as the solution: γn
A(t) = A(0)e 1−θ t (88)
γn
Consequently, both A(t) and Y (t)/L(t) grow at the rate 1−θ

n=0 From equation (88), knowledge A(t) is constant and the growth rate
of knowledge is 0:
A(t) = A(0) (89)
From (89), (78) and (75) output is constant and the growth rate of output
is 0:
Y (t) = (1 − aL )L(0)A(0) (90)
1. Conclusion: The growth rates of output, output per capita and knowl-
edge are all 0

n>0 With (88), (75) and (78), output is given by:

 
γn
t 1−θ
+n
Y (t) = A(0)L(0)(1 − aL )e (91)
With (91) and (78), output per capita is given by:
Y (t) γn
y(t) = = A(0)(1 − aL )et 1−θ (92)
L(t)
From (91), the growth rate of output is given by:
Ẏ (t) γn
= +n (93)
Y (t) 1−θ
From (92), the growth rate of output per capita is given by:
ẏ(t) γn
= (94)
y(t) 1−θ

16
Observations:

1. The growth rates of output and output per capita depend on popula-
tion growth n. This is called the scale effect. It is not observed in
the empirical comparisons across countries.

2. The growth rates of output and output per capita do not depend on
aL : the size of the Research & Development sector is irrelevant for
growth.

6.3.2 Case 2: θ = 1
From (76), we have:  γ
Ȧ(t) = B aL L(t) A(t) (95)

Substitute (78) into (95), rearrange and you get the growth rate of knowl-
edge:
Ȧ(t)
gA (t) = = BaγL L(t)γ (96)
A(t)
Furthermore, from (84) you get:

ġA (t) = γngA (t) (97)

n=0 From (96) we get:

 γ
Ȧ(t) = B aL L(0) A(t) (98)

and
ġA (t) = 0 (99)
(98) is a first-order, linear, homogeneous dynamical system. The solution is
given by: γ γ
A(t) = A(0)eBaL L(0) t (100)
With (100), (75) and (78), output is given by:
γ γt
Y (t) = (1 − aL )L(0)A(0)eBaL L(0) (101)

With (101) and (78), output per capita is given by:

Y (t) γ γ
y(t) = = (1 − aL )A(0)eBaL L(0) t (102)
L(t)
Therefore, the growth rate of output and the growth rate of output per
capita are given by:
ẏ(t)  γ
= B aL L(0) (103)
y(t)

17
Observations:

1. The growth rates of output, output per capita and knowledge are
constant and the same: B(aL L(0))γ

2. The growth rates of output, output per capita and knowledge depend
on aL : The amount of labor in the Research & Development sector
matters for the long run behavior.

n>0 From (96) we have the growth rate of knowledge

Ȧ(t)  γ
gA (t) = = Beγnt aL L(0) (104)
A(t)

Furthermore, from (97) we have the evolution of the growth rate:

ġA (t) = γngA (t) (105)

Observations:

1. Because of (104), the growth rate of knowledge and therefore the

growth rates of output and output per capita are > 0

2. Because of (105), the growth rate of knowledge and therefore the

growth rates of output and output per capita are even increasing.
This is not consistent with empirical data that shows nearly constant
growth rates.

6.3.3 Case 3: θ > 1

From (84), we have the evolution of the growth rate of knowledge
2
ġA (t) = γngA (t) + gA (t)(θ − 1) (106)

Observations:

1. Because of (106) we have ġA (t) > 0 even if n = 0 the growth rate of
knowledge and therefore the growth rates of output and output per
capita are increasing forever.

18
7 Knowledge accumulation with capital
7.1 Main equations
The main equations of this model are:
 1−α
K̇(t) = sY (t) = (1 − ∂K )α (1 − ∂L )1−α K(t)α A(t)L(t) (107)

 β  γ
Ȧ(t) = B ∂K K(t) ∂L L(t) A(t)θ (108)

7.2 decreasing returns: β + θ < 1

The main results are the same as in the model of section (6.3.1):

1. The growth rate of knowledge converges: limt→∞ gA (t) = ḡA (t)

2. n=0: the growth rates of knowledge, output and output per capita are
all 0

3. n>0: the growth rates of knowledge, output and output per capita are
constant and depend on the growth rate of population

4. ∂L and ∂K have no impact on the growth rates of knowledge, output

and output per capita

7.3 constant returns: β + θ = 1

The main results are the same as in the model of section (6.3.2):

1. n=0: the growth rates of knowledge, output and output per capita are
constant

2. n>0: the growth rates of knowledge, output and output per capita are
increasing

3. ∂L , ∂K matter

7.4 increasing returns: β + θ > 1

The main results are the same as in the model of section (6.3.3):

1. The growth rates of knowledge, output and output per capita are
positive and ever increasing even if n=0

2. n, ∂K , ∂L matter

19
8 Learning by Doing
Another way to obtain growth is to assume that the process of production
becomes more effective as production occurs. This is different than assuming
exogeneous technological progress because progress here depends on how
much it is produced. However, no specific investment is required.

8.1 Main equations

The main equations of this model are:
 1−α
Y (t) = K(t)α A(t)L(t) (109)

A(t) = BK(t)θ (110)

sY (t) = K̇(t) (111)
L(t) = L(0)ent (112)
Substitute (110) and (109) into (111) and you get:

K̇(t) = sB 1−α L(t)1−α K(t)α+θ(1−α) (113)

In order to analyze the dynamics, define the growth rate of capital:

K̇(t)
gK (t) = = sB 1−α L(t)1−α K(t)α+θ(1−α)−1 (114)
K(t)

Take the derivation of gK (t) with respect to time and you get:

ġK (t) = n(1 − α)gK (t) + (1 − α)(θ − 1)gk (t)2 (115)

8.2 Case 1: θ < 1

If θ < 1, then it follows from (115) that the growth rate of capital gK (t)
converges to ḡK , where ġK (t) = 0. The growth rate of capital is given by:

K̇(t) n
ḡK = = (116)
K(t) 1−θ

(116) describes a first-order, linear, homogeneous dynamical system whose

solution is given by: n
K(t) = et 1−θ K(0) (117)
With (112), capital per capita is given by:

K(t) nθ
k(t) = = k(0)et 1−θ (118)
L(t)

20
Take the derivation of k(t) with respect to t, and you get the growth rate
of capital per capita:
k̇(t) nθ
= (119)
k(t) 1−θ
Substitute (110), (112) and (117) into (109), and you get the output:
nt
Y (t) = K(0)α+θ(1−α) L(0)1−α B 1−α e 1−θ (120)
From (120) you get the growth rate of output:

Ẏ (t) n
= (121)
Y (t) 1−θ
Divide (120) by L(t) and use (112) to get output per capita:
Y (t) nθ
y(t) = = k(0)K(0)θ(1−α) et 1−θ (122)
L(t)
From equation (122), the growth rate of output per capita is straightforward:
ẏ(t) nθ
= (123)
y(t) 1−θ

Observations:
1. The growth rates depend on the population growth rate and are inde-
pendent of the saving rate s
2. If n=0, then all the growth rates are equal to 0
3. If n>0, then the growth rates are positive and constant. They depend
on the value of n.

8.3 Case 2: θ = 1
If θ = 1, we have from (115):

K̇(t)
ġK (t) = = n(1 − α)gK (t) (124)
K(t)
Equation (124) describes a first-order, linear, homogeneous dynamical sys-
tem. The solution is given by:
K̇(t) K̇(0)
gK (t) = = etn(1−α) = etn(1−α) gK (0) (125)
K(t) K(0)
If θ = 1, then gK (0) is given from (114):

K̇(0)
gK (0) = = sB 1−α L(t)1−α = sB 1−α L(0)1−α ent(1−α) (126)
K(0)

21
8.3.1 n=0
If n=0, we have from (125) and (126):

K̇(t) = K(t)gK (0) = sB 1−α L(0)1−α K(t) (127)

The solution to (127) is given by:

1−α L(0)1−α
K(t) = K(0)etsB (128)

Therefore, the growth rate of capital is given by:

K̇(t)
= sB 1−α L(0)1−α (129)
K(t)

From (128) you get immediately capital per capita and the growth rate of
capital per capita:

K(t) 1−α 1−α

k(t) = = k(0)etsB L(0) (130)
L(t)

k̇(t)
= sB 1−α L(0)1−α (131)
k(t)
Substitute (110), (112) and (128) into (109) and you get output and therefore
the growth rate of output:
1−α L(0)1−α
Y (t) = K(0)L(0)1−α B 1−α etsB (132)

Ẏ (t)
= sB 1−α L(0)1−α (133)
Y (t)
Output per capita and the growth rate of output per capita are given by:
1−α L(0)1−α
y(t) = k(0)L(0)1−α B 1−α etsB (134)

ẏ(t)
= sB 1−α L(0)1−α (135)
y(t)

Observations:

1. Even if n=0, the growth rates are positive and constant

2. All the growth rates are the same

3. The growth rates depend on the saving rate s

22
8.3.2 n>0
Substitute (126) into (125) and you get the growth rate of capital:

K̇(t)
gK (t) = = sB 1−α L(0)1−α e2nt(1−α) (136)
K(t)

Observations:

1. Because of (136), the growth rate of capital is positive

2. Because of (124), the growth rate of capital and therefore all the other
growth rates are ever increasing

3. The growth rates depend on the populations growth n and the saving
rate s

8.4 Case 3: θ > 1

Observations:

1. Because of (114), the growth rate of capital and therefore all the other
growth rates are positive

2. Because of (115), the growth rates are ever increasing

23
9 Human Capital
Theories based on knowledge accumulation cannot explain cross-country
differences in incomes because knowledge is not rival and most of the time
non-excludable. In fact even when it is excludable it is likely that all coun-
tries could buy the use of the best technology.
However, the use of knowledge requires workers with appropriated skills
or abilities. This specific knowledge is called human capital.

• Human capital is excludable and rival as usual physical capital. In-

deed, human capital is attached to a worker. When a worker is used to
produce output his human capital cannot be used by another worker.

• The accumulation of human capital is similar to the accumulation of

physical capital. Devoting more resources to its accumulation increases
the output.

• Some worker’s earning reflect these acquired skills. The estimation of

the part of national income devoted to accumulable inputs is not only
reflected by the capital share. Indeed, the labour share is the sum of
the pure raw labour share and the rewards to human capital.

9.1 Assumptions
• There is one physical output.

• There are four inputs: Physical capital, Raw labour, Human capital
and non-rival Knowledge.

• Physical and Human capital accumulate according to the correspond-

ing investment in the physical good, noted sK and sH. These could
be endogeneised but are here exogenous.

• The accumulation of Knowledge could be endogeneised as in the past

lectures. It is here assumed to grow at the rate g.

• There are CRS in respect to both sorts of capital and labour.

9.2 Main equations

The main equations of the model are the following:
 1−α−β
α β
Y (t) = K(t) H(t) A(t)L(t) (137)

K̇(t) = sK Y (t) (138)

Ḣ(t) = sH Y (t) (139)

24
 L(t) = L(0)ent ⇔ L̇(t) = nL(t) (140)
A(t) = A(0)egt ⇔ Ȧ(t) = gA(t) (141)
From (137) and (138) we get:
 1−α−β
K̇(t) = sK K(t)α H(t)β A(t)L(t) (142)

From (137) and (139) we get:

 1−α−β
Ḣ(t) = sH K(t)α H(t)β A(t)L(t) (143)

(142) and (143) define a two-dimensional, first-order differential equation.

9.3 Analysis
Define capital per effective labor and human capital per effective labor:
K(t)
k(t) = (144)
A(t)L(t)
H(t)
h(t) = (145)
A(t)L(t)
Take the derivations of k(t) and h(t) with respect to time and use (142) and
(143) to get:
•
 \  • • •
K(t) KAL − K(AL + AL)
k̇(t) = = (146)
A(t)L(t) A2 L2
• • •
 
K K A L
= − + (147)
AL AL A L

sK K α H β (AL)1−α−β K
= − (g + n) (148)
AL AL
= sK k(t)α h(t)β − (n + g)k(t) (149)

Similarly
ḣ(t) = sH k(t)α h(t)β − (n + g)h(t) (150)

9.3.1 The steady state

(146) and (??) define a two-dimensional, first-order differential equation that
can be studied within a phase diagram. At the steady state, k̇(t) = 0 and
ḣ(t) = 0. Use this together with (146) and (??) to calculate the steady
state: 1
∗

1−β β 1  1−α−β
k = sK sH (151)
n+g

25
 1
 1  1−α−β
h∗ = sαK sH
1−α
(152)
n+g
An important relation is that
sK k
=
sH h

At the steady state (k*, h*), we have:

k̇(t) h  i0 h  K(t) i0

0= = ln k(t) = ln (153)
k(t) A(t)L(t)

This means that

K(t)
= constant
A(t)L(t)
Manipulate this equation and rearrange to get:

K̇(t)∗
=g+n (154)
K(t)∗

Analogously, the growth rate of human capital is given by:

•
H(t)∗
=g+n (155)
H(t)∗

Observations: The growth rate of output do not depend on the saving

rates sK and sH .It only depends on g and n. The per capita output only
depends on g. As the steady state will be shown to be stable, the long run
growth is independent of the initial conditions. However, the level of output
per capital and capital per capita do depend on the saving rates.

9.3.2 The phase diagram

•
Situation k(t) ≥ 0
From
•
k = sK k α hβ − (n + g)k ≥ 0
we get
k(n + g) ≤ sK k α hβ

k 1−α ≤ sK hβ (n + g)−1

26
   1
sK 1−α β
k≤ h 1−α
n+g
with
β 1−α
< =1
1−α 1−α
•
So the curve is concave. Above k(t) = 0 the derivative is negative so k is
•
decreasing with time. The arrow goes down. Below k(t) = 0 the situation
is opposite.
•
Situation y(t) ≥ 0

From
•
h = sH k α hβ − (n + g)h ≥ 0
we get
h(n + g) ≤ sH k α hβ

1 1−β
kα ≥ h (n + g)
sH

 1
n+g α 1−β
k≥ h α
sH
with
1−β 1−β
> =1
α 1−β
•
So the curve is convex. To the left of h(t) = 0 the derivative is positive so h
•
is increasing with time. The arrow goes to the right. To the right of h(t) = 0
the situation is opposite. The phase diagram shows that the steady state is
stable.

9.3.3 Quantitative analysis

In order to calculate the elasticity of output per effective labor with respect
to sK and to sH at the steady state, first take the logarithm on both side of
equations (151) and (152) and manipulate them to get:
1  
ln(k ∗ ) = (1 − β)ln(sK ) + βln(sH ) − ln(n + g) (156)
1−α−β
1  
ln(h∗ ) = αln(sK ) + (1 − α)ln(sH ) − ln(n + g) (157)
1−α−β

27
Divide (137) by A(t)L(t) to get output per effective labor:

y(t) = k(t)α h(t)β (158)

Take the logarithm to both side of (158), manipulate and evaluate at the
steady state to get:
ln(y ∗ ) = αln(k ∗ ) + βln(h∗ ) (159)
Substitute (156) and (157) into (159) and rearrange:
α β α+β
ln(y ∗ ) = ln(sK ) + ln(sH ) − ln(n + g) (160)
1−α−β 1−α−β 1−α−β
Take the derivative with of (160) with respect to sK , sH respectively and
you get:
ẏ ∗ α 1
= [ln(y∗)]0 = (161)
y∗ 1 − α − β sK
ẏ ∗ β 1
= [ln(y∗)]0 = (162)
y∗ 1 − α − β sH
The elasticities of output per effective labor with respect to sK and sH are
then given by:
ẏ∗
y α
e sK = ṡK
= (163)
sK
1−α−β
ẏ∗
y β
e sH = ṡH
= (164)
sH
1−α−β

9.3.4 An application
Some part of the worker’s earning is due to human capital. To find out
how much we assume that the minimum wage reflects the inherent skills.
For example if we assume then that 14 the average wage w is associated to
non-accumulated skills while 34 w is due to human capital. As usual let α
be the physical capital share of national income. Then (1 − α) is the share
remaining for both acquired and non-aquired labour abilities. Assuming
that 43 comes from human capital we get β = 34 (1 − α).
In agreement with this order of magnitude let α = 0.35 and β = 0.4 so
that 1 − α − β = 0.25. With these values we get
0.35 0.4 0.75
ln(y ∗ ) = ln(sK ) + ln(sH ) − ln(n + g) (165)
0.25 0.25 0.25

or
ln(y ∗ ) = 1.4 ln(sK ) + 1.6 ln(sH ) − 3 ln(n + g) (166)

28
while with Solow we would get

ln(y ∗ ) = 0.54 ln(sK ) − 0.54 ln(n + g) (167)

The difference between the two model;s is large. Indeed, assume that the
saving rates are twice in country B compared to country A and that (n + g)
is 20% smaller in country B. Then

log yA − log yB = 1.4(log sA A A A

K − log sK ) + 1.6(log sH − log sH ) − 3 log(0.8)

giving

log yA − log yB = 1.4 log 2 + 1.6 log 2 − 3 log(0.8) = 2.75

so that
yB
= e2.75 = 15.6
yA
while with Solow we would have
yB
= e0.4 = 1.6
yA
Because of large elasticities of output respect the underlying determinants
of the model, the model has the potential to explain cross country income
differences.

29
10 Malthus
Historical data shows that before the industrial revolution, increase in to-
tal output was “used” to increase population rather than income per capita.
This is because the minimal subsistance level constraint was binding. Malthu-
sian models try to explain the observed rise in population during this pe-
riod. They endogeneise n. The models of knowledge accumulation with and
without capital can we used as they predict a relationship between growth
in total output and population growth n.
The main assumptions of the model we consider here are the following:

1. There is an input in fixed supply: Land (:=R). Capital is ignored. We

assume  1−α
Y (t) = Rα A(t)L(t) (168)

2. Population adjusts so that the output per capita is fixed at the sub-
sistance level ȳ
Y (t)
= ȳ ⇔ Y (t) = ȳL(t) (169)
L(t)

3. Growth rate of knowledge is proportional to L(t)

Ȧ(t)
gA (t) = = BL(t) (170)
A(t)
The model can be solved. Indeed, first put (eq: 64) and (169) together,
multiply both side of this equation by L(t)α−1
 1−α
L(t)α−1 Rα A(t)L(t) = ȳL(t)L(t)α−1 (171)

giving
 1−α
Rα A(t) = ȳL(t)α (172)

Solving for L(t) you get:

1−α
A(t) α
L(t) = R 1 (173)
ȳ α
The growth rate of population is then given by:

L̇(t)  1 − α  Ȧ(t)  1 − α 
n(t) = = = BL(t) (174)
L(t) α A(t) α
Equation (174) can be tested empirically by an econometric regression. The
result is
n(t) = −0.0023 + 0.524L(t)

30
and the fit is very good.
A second test of the theory is to consider economies that were isolated
for long periods of time. They were four blocs that were separated at the
end of the ice age with no communication until 1500. As we can assume
thay the density was identical when they were separated, the population
density in 1500 reflects the poplutation growth rate since the end of the ice
age. Larger blocs had a larger initial population so that they are expected
to grow faster with a resulting higher population density in 1500. The data
agree with the prediction

• Eurasia-Africa: Surface=84 millions km2. Density in 1500: 4.9 h/km

• Americas: Surface=38 millions km2. Density in 1500: 0.45 h/km

• Australia: Surface=8 millions km2. Density in 1500: 0.03 h/km

• Tasmania: Surface=0.1 millions km2. Density in 1500: 0.03 h/km

31