Chapter 1

Preliminaries

“If economy fluctuates for whatever reason, then it is almost always possible
to neutralize these cyclic motions by means of anti-cyclic demand policies”

1
Chapter 2

Goodwin Model

2.1 The model

In the present section we follow (Costa Lima, 2013).
The Goodwin model is a two-dimensional dynamical system based on
Lotka-Volterra model in which wages and employment work as predator and
prey respectively. We assume that production follows a Leontief behavior
determined by capital and labor.
 
K(t)
Y (t) = min , a(t)L(t)
ν

The capital-to-output ratio which indicates productivity of capital invest-

ment is represented by ν. We consider the growth of labor productivity and
total labor force constant, this is

ȧ(t)
= α, (2.1)
a(t)
Ṅ (t)
=β (2.2)
N (t)

therefore the functions evolve according to

a(t) = a0 eαt , (2.3)

βt
N (t) = N0 e (2.4)

for some real numbers a0 and N0 . Employment rate is denoted by

L(t)
λ(t) = (2.5)
N (t)

2
In order not to move away from the optimal allocation of resources we take
on full capital utilization
K(t)
Y (t) = = a(t)L(t) (2.6)
ν
Next, we assume that the real-wage rate is increasing around full employ-
ment. If we denote w(t) as the real-wage function the latter can be expressed
by
ẇ(t)
= Φ(λ(t)) (2.7)
w(t)
for an increasing Phillips curve, Φ(λ(t)).
The model also embraces Say’s law which states that real investment is
equal to real profits and real consumption is equal to real earnings from labor.
This is

U (t) = Y (t) − C(t) = I(t)

C(t) = w(t)L(t)

This implies that capital grows according to the following:

K̇(t) = (Y (t) − w(t)L(t)) − δK(t) (2.8)

here, δ is depreciation rate . We define ω as the wage share or labor share of

the economy.
w(t)L(t)
ω(t) :=
Y (t)
From equation 2.6, the following relationships hold:
w(t)L(t) w(t)
ω(t) = = , (2.9)
a(t)L(t) a(t)

U (t) = Y (t) − w(t)L(t) = Y (t) − ω(t)a(t)L(t) = (1 − ω(t))Y (t)

and

Y (t)ν = K(t)

replacing the latter in equation 2.8 we have

ν Ẏ (t) = (1 − ω(t))Y (t) − δνY (t)

= (1 − ω(t) − δν)Y (t)

3
then
Ẏ (t) (1 − ω(t))
= −δ
Y (t) ν
we define µ as the rate of growth of the product for later analysis

Ẏ (t)
µ(ω) = .
Y (t)
Taking logarithm to equation 2.9, deriving and taking equation 2.7 into ac-
count we have
ω̇(t) ẇ(t) ȧ(t)
= − = Φ(λ(t)) − α
ω(t) w(t) a(t)
From the definition of employment rate in equation 2.5 and full capital uti-
lization we get

λ̇(t) Ẏ (t) Ṅ (t) ȧ(t) (1 − ω(t))

= − − = −δ−α−β
λ(t) Y (t) N (t) a(t) ν
Finally, we have the following differential equation system:

ω̇(t) = ω(t) [Φ(λ(t)) − α]

(2.10)
 
1 − ω(t)
λ̇(t) = λ(t) −α−β−δ
ν

2.1.1 Equilibrium
The non-trivial equilibriums of the system 2.10 are equal to

(ω̄, λ̄) = (1 − ν(α + β + δ), Φ−1 (α)) (2.11)

The jacobian matrix at the equilibrium points is the following:

0 Φ0 (λ̄)
 
J(ω̄, λ̄) =
− λ̄ν 0

the trace of the matrix is zero and the determinant is

λ̄ω̄ 0
Φ (λ̄)
ν
assuming Φ0 (λ̄) > 0 as we did in the previous section the solutions around
the equilibrium are orbits.

4
2.1.2 Interpretation
We can write the previous system 2.10 in terms of rate of growth as follows

ω̇(t)
= Φ(λ(t)) − α
ω(t)
λ̇(t) 1 − ω(t)
= −α−β−δ
λ(t) ν

If product growth rate rises faster than the constant rates of labor produc-
tivity and total labor force, equations 2.3 and 2.4, then employment growth
rate rises and labor share shrinks. Assuming an increasing behavior of em-
ployment this yields a raise in labor, because of the assumed increasing form
of the total labor force, meaning that firms hire more workers so that em-
ployment grows faster than unemployment. By the full capital utilization
assumption, as labor and labor productivity rises so does product thus by
constant capital-to-output assumption capital increases so showing a trade-
off between capital and labor. The demand for employers pushes the growth
rate of wages up. This is a result of considering an increasing Phillips curve.
Labor share growth will only increase if wage growth is greater than that
of labor productivity. As employment continues rising the rate of growth of
wages star receding and so labor growth. In this way product and capital
decay and labor share appreciates.

2.2 Stylized facts

We compare the assumptions of the model with real data.
In order to see if the growth of productivity and total labor force are
constant, we perform a linear regression based on equations 2.1 and 2.2.
table A.1 in the appendix shows the results. According to the fit and the
plots in figures 2.1 and 2.2 we observe that the slopes are significant therefore
the assumption doesn’t match the data.

5
 0.06
0.02 a⋅ (t)/a(t): Productivity growth
a⋅ (t)/a(t)

−0.06 −0.02

1990 1995 2000 2005 2010 2015

Figure 2.1
Source: WDI, own elaboration.

⋅
N(t)/N(t): Total labor force growth
0.06
N(t)/N(t)

0.02
⋅

−0.02

1990 1995 2000 2005 2010 2015

Figure 2.2
Source: WDI, own elaboration.

We check the assumption of constant capital-to-ouput ratio. From the

plot and the linear regression with results on table A.2 we assure that the

6
ratio is not constant and has a decreasing trend.

ν
8
capital−to−output

7
6
5
4

1990 1995 2000 2005 2010 2015

Figure 2.3
Source: WDI, own elaboration.

We test the fit of two types of Phillips curve:

Φ1 = φ1 λ + φ2
φ1
Φ2 = + φ2
(1 − λ)2

table A.3 provides the results. We see that the linear model explains more
than the cuadratic inverse model.

7
 Linear Phillips curve

●
0.00 0.05

●
● ●
● ●
●
● ●
● ●
w⋅ (t)/w(t)

●
● ●●
● ●
● ●
● ●
●
−0.10

0.940 0.945 0.950 0.955 0.960 0.965 0.970

λ(t)

Figure 2.4
Source: WDI, own elaboration.

Cuadratic inverse Phillips curve

●
0.00 0.05

●
● ●
● ●
●
● ●
● ●
w⋅ (t)/w(t)

●
● ●●
● ●
●
●● ●
●
−0.10

400 600 800 1000

1/(1−λ)2

Figure 2.5
Source: WDI and PWT, own elaboration.

The following plot let us notice the trade-off between lambda and omega,
when lambda increases omega decreases and vice versa.

8
 0.970
0.45 ω(t)
λ(t)

0.960
0.40
ω(t)

λ(t)
0.950
0.35

0.940
0.30

1990 1995 2000 2005 2010 2015

Figure 2.6
Source: WDI and PWT, own elaboration.

The following omega and lambda parametric plot unveils what seems to
be a clockwise cycle behavior

(ω(t),λ(t))
0.970

2014
2012
2013
2011
2010
0.960

2009
2008
2007
2006
λ(t)

0.950

2005 2003
2004 1994
1995
1997
1993
0.940

2000
2002 1996 1991
1999
2001
1998 1992

0.30 0.32 0.34 0.36 0.38 0.40 0.42

ω(t)

Figure 2.7
Source: WDI, own elaboration.

9
2.3 Calibration for Peru
The purpose of this section is to provide details on how we handled data to
estimate the parameters of the model and how we validate it.
We worked with the World Development Indicator (WDI) database from
the World Bank and the Penn World Table (PWT) which are more complete
for our purposes compared to BCRP and INEI databases. There is data
available from 1950 to 2019, but not for all variables (e.g employment rate).
The parameters of the Goodwin model we need to estimate are the fol-
lowing: growth rate of productivity and labor, the capital-to-output ratio
and the deprecation rate of capital.
The details of the estimation are found in table 3.4. We obtain α̂ = 0.0129
and β̂ = 0.0315.
−13.1
ln(a(t))

−13.3
−13.5

1990 1995 2000 2005 2010 2015

t
25.8
ln(N(t))

25.4
25.0

1990 1995 2000 2005 2010 2015

Figure 2.8
Logarithm of productivity and labor respectively

10
 For capital-to-output constant ν we notice that the variation is significant
before 1990. From 1990 the average is 7 and from 2004 on is 5 and the trend
is downwards. Taking into account (Carranza, Fernandez-Baca, & Morón,
2005) which states that the capital-to-output constant is going up during the
period 1968-1990 we have the set of possibilities ν̂ ∈ {3, · · · , 7}.

ν
8
capital−to−output

7
6
5
4

1990 1995 2000 2005 2010 2015

Figure 2.9
Capital-to-output rate

From PWT we get δ̂ = 0.051 which is a value closed to the one used in
(Céspedes & Ramı́rez-Rondán, 2014).

δ
0.050
0.045
delta

0.040
0.035

1990 1995 2000 2005 2010 2015

Figure 2.10
Capital depreciation

We have the following table:

11
 Parameter value
α̂ 0.0129
β̂ 0.0315
ν̂ {3, · · · , 7}
δ̂ 0.051

We replace the values of the parameters in the set of equations 2.10 and
obtain the following model for Peru:

ω̇(t) = ω(t) [Φ(λ(t)) − 0.0129]

 
1 − ω(t)
λ̇(t) = λ(t) − 0.0776
ν̂
we verify that
1
− α̂ − β̂ − δ̂ > 0
ν̂
1
ν ν
−α−β−δ
3 0.256
4 0.172
5 0.122
6 0.089
7 0.065
in order to perform the simulation of the theoretical model we need to assume
a functional form of the Phillips curve. Table A.3 provides details of the
estimations of the curves.

2.3.1 Model 1: linear Phillips curve

we take the following phillip curve:

Φ1 = 2.4935λ(t) − 2.37

Then we have the following model

ω̇(t) = ω(t) [2.4935λ(t) − 2.3829]

λ̇(t) = λ(t) [−0.333ω(t) + 0.256]

with non trivial equilibrium equal to

(ω̄, λ̄) = (0.768, 0.953) (2.12)

12
we solve the system for the initial values ω(0) = 0.34 and λ(0) = 0.95 and
obtain the following the following plots:

nu=3, w_0=0.34, lambda_0=0.95 nu=4, w_0=0.34, lambda_0=0.95

1.5

1.5
1.0

1.0
Lambda

Lambda
0.5

0.5
0.0

0.0
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5

Omega Omega

nu=5, w_0=0.34, lambda_0=0.95 nu=6, w_0=0.34, lambda_0=0.95

1.5

1.5
1.0

1.0
Lambda

Lambda
0.5

0.5
0.0

0.0

0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5

Omega Omega

nu=7, w_0=0.34, lambda_0=0.95

1.5
1.0
Lambda

0.5
0.0

0.0 0.5 1.0 1.5

Omega

13
 nu=3, w_0=0.34, lambda_0=0.95 nu=4, w_0=0.34, lambda_0=0.95
1.5

1.5
Omega Omega
Lambda Lambda
Omega, Lambda

Omega, Lambda
1.0

1.0
0.5

0.5
0.0

0.0
0 20 40 60 80 100 0 20 40 60 80 100

Time Time

nu=5, w_0=0.34, lambda_0=0.95 nu=6, w_0=0.34, lambda_0=0.95

1.5

1.5
Omega Omega
Lambda Lambda
Omega, Lambda

Omega, Lambda
1.0

1.0
0.5

0.5
0.0

0.0

0 20 40 60 80 100 0 20 40 60 80 100

Time Time

nu=7, w_0=0.34, lambda_0=0.95

1.5

Omega
Lambda
Omega, Lambda

1.0
0.5
0.0

0 20 40 60 80 100

Time

we appreciate that there is a predator-prey behavior, but the variables

reach values that surpass the meaningful limits. The cycle limit is

T = 7.864 ≈ 8 years

2.3.2 Model 2: inverse quadratic Phillips curve

we take the following phillip curve:
φ1
Φ2 = + φ2
(1 − λ(t))2

14
Regressing from the data we have that the model is the following

9 ∗ 10−5
Φ2 = − 0.04
(1 − λ(t))2

Then we have the following model

ω̇(t) = ω(t) 9 ∗ 10−5 (1 − λ(t))−2 − 0.04

 

λ̇(t) = λ(t) [−0.333ω(t) + 0.256]

with non trivial equilibrium equal to

(ω̄, λ̄) = (0.768, 0.953) (2.13)

we solve the system for the initial values

15
 nu=3, w_0=0.34, lambda_0=0.95 nu=4, w_0=0.34, lambda_0=0.95
1.5

1.5
1.0

1.0
Lambda

Lambda
0.5

0.5
0.0

0.0
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5

Omega Omega

nu=5, w_0=0.34, lambda_0=0.95 nu=6, w_0=0.34, lambda_0=0.95

1.5

1.5
1.0

1.0
Lambda

Lambda
0.5

0.5
0.0

0.0

0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5

Omega Omega

nu=7, w_0=0.34, lambda_0=0.95

1.5
1.0
Lambda

0.5
0.0

0.0 0.5 1.0 1.5

Omega

16
 nu=3, w_0=0.34, lambda_0=0.95 nu=4, w_0=0.34, lambda_0=0.95
1.5

1.5
Omega Omega
Lambda Lambda
Omega, Lambda

Omega, Lambda
1.0

1.0
0.5

0.5
0.0

0.0
0 20 40 60 80 100 0 20 40 60 80 100

Time Time

nu=5, w_0=0.34, lambda_0=0.95 nu=6, w_0=0.34, lambda_0=0.95

1.5

1.5
Omega Omega
Lambda Lambda
Omega, Lambda

Omega, Lambda
1.0

1.0
0.5

0.5
0.0

0.0

0 20 40 60 80 100 0 20 40 60 80 100

Time Time

nu=7, w_0=0.34, lambda_0=0.95

1.5

Omega
Lambda
Omega, Lambda

1.0
0.5
0.0

0 20 40 60 80 100

Time

17
 nu=6, w_0=0.34, lambda_0=0.95
1.5
1.0
Lambda

0.5
0.0

0.0 0.5 1.0 1.5

Omega

Figure 2.11
Theoretical and real ω and λ

18
Chapter 3

Keen Model:

We assume that capitalists do not invest the total of the profits this way we
introduce the finance sector. We have a new variable D which is the amount
of debt in real terms
U = (1 − ω − rd)Y
where d = D/Y is the debt ratio in the economy. Reinvesting all the profits
lead to constant debt levels which differ from reality according to Minsky 82.
The real behavior is that high net profits lead to more borrowing whereas
low net profits lead to deleveraging of the economy. Accumulation of capital
is
K̇ = κ(1 − ω − rd)Y − δK
where the rate of new investment is a nonlinear increasing function of the
net profit.] As before we have
ω̇ = ω [Φ(λ) − α]
 
κ(1 − ω − rd)
λ̇ = λ −α−β−δ
ν
we assume the following behavior of debt:
Ḋ = κ(1 − ω − rd)Y − (1 − ω − rd)Y
so we have the
ω̇ = ω [Φ(λ) − α]
 
κ(1 − ω − rd)
λ̇ = λ −α−β−δ
ν
 
κ(1 − ω − rd)
d˙ = d r − + δ + κ(1 − ω − rd) − (1 − ω)
ν

19
Appendices

20
Appendix A

Estimations

Table A.1
Results

Dependent variable:
growth.prodty growth.lf
(1) (2)
YEAR 0.002∗∗ −0.001∗∗∗
(0.001) (0.0003)

Constant −4.754∗∗ 2.966∗∗∗

(1.975) (0.651)

Observations 23 27
R2 0.217 0.448
Adjusted R2 0.180 0.426
Residual Std. Error 0.031 (df = 21) 0.013 (df = 25)
F Statistic 5.832∗∗ (df = 1; 21) 20.328∗∗∗ (df = 1; 25)
∗ ∗∗ ∗∗∗
Note: p<0.1; p<0.05; p<0.01

21
Table A.2
Results

Dependent variable:
nu
YEAR −0.457∗∗∗
(0.036)

Constant 920.719∗∗∗
(71.744)

Observations 55
R2 0.751
Adjusted R2 0.746
Residual Std. Error 4.251 (df = 53)
F Statistic 160.018∗∗∗ (df = 1; 53)
∗ ∗∗ ∗∗∗
Note: p<0.1; p<0.05; p<0.01

22
Table A.3
Results

Dependent variable:
growth.w
(1) (2)
lambda 2.494∗∗∗
(0.705)

invcuadl 0.0001∗∗
(0.00003)

Constant −2.372∗∗∗ −0.042∗∗

(0.672) (0.017)

Observations 23 23
R2 0.373 0.275
Adjusted R2 0.343 0.240
Residual Std. Error (df = 21) 0.035 0.038
F Statistic (df = 1; 21) 12.499∗∗∗ 7.957∗∗
∗ ∗∗ ∗∗∗
Note: p<0.1; p<0.05; p<0.01

We compute α and β by performing a linear regression to the logarithm of

productivity and logarithm of labor to time from equations 2.3, 2.4.

ln a = ln a0 + αt
ln N = ln N0 + βt

23
Table 3.4
Results

Dependent variable:
lnprodty lnlabor DELTA
(1) (2) (3)
YEAR 0.013∗∗∗ 0.032∗∗∗ −0.0003∗∗∗
(0.002) (0.001) (0.0001)

Constant −39.200∗∗∗ −37.607∗∗∗ 0.564∗∗∗

(4.307) (2.354) (0.122)

Observations 24 28 65
R2 0.621 0.965 0.219
Adjusted R2 0.604 0.964 0.206
Residual Std. Error 0.073 (df = 22) 0.050 (df = 26) 0.009 (df = 63)
F Statistic 36.022∗∗∗ (df = 1; 22) 720.535∗∗∗ (df = 1; 26) 17.646∗∗∗ (df = 1; 63)
Note: ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01

24
References
Carranza, E., Fernandez-Baca, J., & Morón, E. (2005). Markets, government,
and the sources of growth in peru. Sources of Growth in Latin America.
What is Missing.
Céspedes, N., & Ramı́rez-Rondán, N. (2014). Total factor productivity
estimation in peru: Primal and dual approaches. Peruvian Economic
Association.
Costa Lima, B. R. (2013). The dynamical system approach to macroeco-
nomics (Published doctoral dissertation). McMaster University, Hamil-
ton, Ontario.
Gabish, G., & Lorenz, H. (1987). Bussines cycle theory: A survey of methods
and concepts (Vol. 283). Berlin-Germany: Springer-Verlag.
Goodwin, R. (1967). Socialism, capitalism and economic growth: Essays
presented to Maurice Dobb. In C. Feinstein (Ed.), (p. 54-58). Cambrige:
Cambridge University Press.
Harvie, D. (2000, May). Testing Goodwin: Growth cycles in the oecd coun-
tries. Cambridge Journal of Economics.
Herrera-Medina, E., & Garcia-Molina, M. (2010). Are there Good-
win employment-distribution cycles? international empirical evidence.
Cuadernos de economı́a.

25