Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

DIFFERENTIAL-DIFFERENCE EQUATIONS

IN ECONOMICS: ON THE NUMERICAL


SOLUTION OF VINTAGE CAPITAL
GROWTH MODELS
Aryaman Bansal (2015CH10089)
Abhiprai Misra (2015CH10073)
Mukul Mittal (2015CH70168)
14th November 2016

Abstract
This paper examines techniques for the analytical and numerical solu-
tion of Solow’s vintage capital growth model equations. Delay di↵erential
type equations occur in the continuous time modelling of vintage capi-
tal growth models, which form a particularly important class of models in
modern economic growth theory. In general, though, the state-dependence
of a model prevents its analytical solution in all but the simplest of cases.
This paper reviews numerical method for solving state-dependent mod-
els and analyses the Solow vintage capital growth model and plots, some
important curves according to Solow’s model explaining the underlying
economic sense beneath them.

1
1 Introduction
Often in various economic problems, we encounter the problems related to the
time heterogeneity where we have to use the value of various goods and assets at
a common time, despite the di↵erence in their ages to decide the optimum time
of their replacement. In most cases, this becomes the key decision to decide
the optimum time of replacement as newer goods and assets are always backed
up by the technological advancement, so replacing older goods with newer ones
brings technological advancement, but it also comes with the loss of value of old
goods, hence a need to decide optimum time. To illustrate this point, we have
used the capital replacement decision as our issue. The basic structural model
used is Solow’s vintage capital growth model that serves as the basis of all the
economical equations used in this paper. From an economic point of view, we
have addressed one of the important issues of machine replacement i.e. Deciding
the optimum time of replacement of old machines in an economy.
While solving the issue addressed above of machine replacement as per Solow
model, we encounter the system of di↵erential equation of the form y 0 (t) =
F [t, y(t), y(t T (t, y(t)))] where t is the time index, y(t) is the vector of en-
dogenous variables, F (·) is an appropriately dimensioned vector function, and
T (·) is a real-valued function. This type of di↵erential equation is known as
di↵erential di↵erence equations and system of this type of equations is known
as delay di↵erential equation (DDE) system. The lag term T (t, y(t)) is state
dependent as it depends on the endogeneous variable y(t). Due to lag term
being state dependent, Laplace transform approach is not very useful in solving
such DDEs. So to solve such DDEs, we have used the method of steps.

2 Problem formulation
The structural equations of the Solow model considered as an economic appli-
cation to algorithm of DDE are:
Z t
y(t) = i(z)dz, (1)
t T (t)
Z t
l(t) = i(z)exp{ (z)z}dz, (2)
t T (t)

i(t) = sy(t), (3)


where y(t) is production, l(t) is labor demand, i(t) is investment and is spec-
ified for t < 0, and T (t) is the age of the oldest machine still in use at time t.
The parameter s 2 (0, 1) and represents the saving rate.
Equations 1 and 2 represents the age structure of the technology. While in
use, each machine of generation t is assumed to require exp{ (t)t} workers
to operate it - its labor requirement. Furthermore, we assume l(t) = 1 for all
time i.e. Labor supply is inelastic and constant over the time. Equation 3 as-
sumes that a fixed proportion of income is saved and used in the investment

2
in the new machines. The time-dependent function (t) represents the tech-
nological progress. We have safely assumed that (t)t is increasing with time
as new machines are always more productive than older machines, in the sense
that older machines require more workers to produce the same quantity of goods.

Now, by di↵erentiating the equations 1-3, we obtain (for t 0),

y 0 (t) = sy(t)(1 R(t)), (4)

i(t)
T 0 (t) = 1 ⇥ R(t), (5)
i(t T (t))
i0 (t) = sy 0 (t), (6)
where,
exp{ (t)t}
R(t) = (7)
exp{ (t T (t))(t T (t))}
The initial conditions y(0), T (0) and i(0) are obtained from 1-3 by putting t = 0.
Since it is assumed that (t)t is increasing, the production y(t) and investment
i(t) have a trend component. Note that the rate of growth depends on the re-
quirement ratio R(t), i.e., The ratio of the labor requirement of new machines
to the labor requirement of the machines that they are replacing.

3 Numerical analysis
In this section, we will discuss the problem solving approach, indicating the
methods to be employed. So, in order to understand the approach clearly, we
will start with some basic DDEs and then further increase the complexities.

3.1 Scalar constant lag DDE-I


In this subsection, we will solve the equation

y 0 (t) = y(t ⌧) t 0, (8)

where ⌧ > 0. Here, to solve DDE, we require initial function y0 (t) specified over
time [ ⌧, 0] to be able to solve equation 8.
Let us take an example, y0 (t) = a where a is a constant. Now, as mentioned in
the paper by Raouf Boucekkine, Ornar Licandro and Christopher Paul (1995),
we used the technique, method of steps, in computational mathematics. Further,
we varied a and ⌧ to study the functions that we obtain. In Figures 3 and 4,
we can see the variation by a and ⌧ respectively. In Figure 5, we further varied
a of the initial function y0 (t) = a with general functions.

3
3.2 Scalar constant lag DDE-II
Now in this subsection, let us apply the method of steps to a more general
equation that cannot be solved analytically. Consider the scalar equation

y 0 (t) = G[y(t), y(t ⌧ )] t 0, (9)

where both ⌧ > 0 and y(t) = y0 (t) on [ ⌧, 0] are given, and G(·) is any real
valued function. Here, to observe the variation with di↵erent functions of G(·),
we took G(·) as sin(·), cos(·), log(·) and 1(·).
Now, by applying the MSA technique, we get the Figures 6, 7 and 8.
In the Figure 6, we varied the initial function, in the Figure 7, we varied ⌧ and
in the Figure 8, we varied the function G(·) to observe their changes on the
function.

3.3 Time-varying State Dependent DDE


Now in this subsection, we are considering a time-varying state dependent DDE.
In this case, the MSA technique that we use has problems as the meshpoints are
not directly available to us. So, we modify the MSA technique in a way that we
first obtain the unknown meshpoint and then applying MSA. This way of solv-
ing this case has been extensively explained in the paper by Raouf Boucekkine,
Ornar Licandro and Christopher Paul (1995). Further, we numerically ana-
lyzed this case by taking examples for functions. We varied the initial(history)
function y0 (t) and the general lag function ⌧ (t) in the equation 10.

y 0 (t) = y(⌧ (t)) t 0, (10)

We can see the results obtained in Figures 9 and 10. In the Figure 9 we took
initial(history) function y0 (t) = sin(t), cos(t) and log(t) and in the Figure 10
we took general lag function ⌧ (t) = exp(j ⇤ (1 y(t)) with j = {1, 2, 3, 4}

4 Results and Discussion


4.1 Solow Vintage Capital Growth Model
As Solow et al showed, if (t) is constant ( (t) = > 0) and s < , then the
economy converges to a balanced growth path.
Now, Figures 11 and 12 gives T (t) and detrended production when the initial
investment profile is exponential: i0 (t) = kexp{ t} for k > 0. In this case it is
easy to show that T (0) = 1/k from the structural equations.

When k = 1/T ⇤ , the solution is trivial, since limt!0 i(t) = i(0) and the
past investment i0 (t) corresponds to the balanced growth path. When k < 1/T ⇤
(resp. k > 1/T ⇤ ), the economy starts with a low (resp. high) stock of machines
relative to the balanced growth path. We can see from Figure 11 that the initial
lifetime of capital must be greater (resp. smaller) than T ⇤ to allow the labor

4
market to clear . After a jump in the investment at t = 0, since limt!0 i(t)
is smaller (resp. greater) than i(0), the model converges monotonically to the
balanced growth path. The above results have been explained in the graph
obtained in figure 11 by plotting the solution of the DDEs obtained through
structural equations.

4.2 Extensive Solow Vintage Capital Growth Model


To investigate the part of convergence with technological factor varying and
fluctuating, we consider a time-dependent technological progress with a peri-
odic component that vanishes asymptotically, so that we recover the Solow et
al specification of the technological progress at the limit. We specify the tech-
nological progress as follows:

sin(!t)
(t) = +↵ ↵ > 0, ! > 0, (11)
t
with > ↵! to ensure that (t)t is an increasing function.
But, despite the fact that we have used specification of the technological progress,
such that it quickly converges to the SVCM, we obtain two completely di↵erent
types of dynamics in the figures obtained of T (t) depending on the periodicity
parameter !. In both the cases, we have assumed i0 (t) = kexp{ (t)t}. In the
13, when ! = 2⇡/T ⇤ , so that sin(wt) has periodicity T ⇤, we can see that T (t)
plot converges to the long term solution T ⇤ and similarly y(t) plot converges to
the long term solution y⇤ whereas in the 14, when ! = ⇡/5⇤ i.e. not a multiple
of T ⇤, we can see that both of the graph doesn’t converges. The economic reason
behind this is that in the first case, periodic component of the requirement ratio
vanishes after a few periods, whereas in the second case, periodic component of
requirement ratio never vanishes, hence both graph remains periodic with the
same periodicity of sin(wt).

5 Level 2 - Introducing the Labor function


Throughout the whole model, that we have been analyzing according to this
term paper, we have taken a basic assumption that labor supply is fixed and
inelastic over the time i.e. l(t) = 1, which can be easily questioned over in a
functional economy as we know that labor supply is not completely fixed over
the time. Instead in a general economy, labor supply can be more conveniently
assumed to be increasing with time. Also, it shouldn’t be an exponential increase
or a straight line increase that just increases over the course of time signifying
unusual labor supply in a time interval, so we think that labor supply function
should be approximated by a function which increases but the rate of
increase of the function should be decreasing. Therefore, for further
research, we assumed labor to be l(t) = 1 + exp{ t}, replacing l(t) = 1.
After changing the labor function in the original model and keeping the rest
of the model as it is, we again formulated the new di↵erential equations and

5
plotted T (t) again and interpreted the plot obtained further in the report. The
di↵erential equations produced in this case are present in the Matlab code
submitted along with the report.

Figure 1: T (t) t plot after including the Labor function as 1 + exp{ t}

Figure 2: T (t) t plot after including the Labor function as 1 + exp{ 5t}

6
Interpretation of graphs

The labor supply in the starting is high owing to the low value of time (t).
The graph tends to rise in the starting because as the labor supply is high,
the economies won’t feel the need to invest in new more productive machines
owing to the readily available cheap labor. As the time increases, the labor
supply will tend to dry and the economies will feel the pressure to invest in
new and more productive machines to tackle the problem of shortage of labor,
this explains the sudden dip in our graph. As the time becomes very large
the graph starts to flatten out as the labor demands becomes constant (= 1)
and the graph tends to converge, in line with what was expected as in Figure 11.

On changing the labor function to l(t) = 1 + exp{ 5t} we get the Figure 2.
This graph is quite similar to the above, just the di↵erence is that in this
graph the labor function is decreasing more rapidly and decreases 5 times faster
than the one in Figure 1. In the starting T (t) doesn’t increase owing to the
rapidly decreasing labor supply as the economies are forced to buy more ma-
chines owing to rapidly decreasing labor supply. As time increases both the
curves start to converge as the labor functions tend to become same for large
values of t.

6 Conclusion
In this paper, we first learnt what DDEs are and the e↵ect of changing varying
parameters on the solution and this knowledge helped us in analysis of the Solow
model. Then we analysed the Solow model for long term economic performance
on the basis of some structural equation that results in the formation of DDEs
with constant and varying time lags. These are then solved and plotted using
Matlab to analyse the long term performance and explain the underlying logics
based on economics for the obtained figures.

7 Self Assessment
In this term paper, we basically focused on solving Solow’s growth model for
analyzing the production y(t) and the parameter T (t) i.e. the age of the old-
est machine still in use in the economy at any given time t which includes the
formation and solution of system of Delayed Di↵erential Equation (DDE). In
doing so, we firstly learnt about solving DDEs through Matlab and plotting
them, and plots of solving DDEs with simple examples have been put already
in this term paper. We also studied the e↵ects of changing various parameters
of Delay Di↵erential equation on their solution and the plots of the solutions
with the changing parameters are included in the report. Then we came to
our objective i.e. solving the Solow’s model and tried to get all the results and
plots presented in the journal by understanding and solving the system of delay

7
di↵erential equations underlying Solow’s model. In the midterm report of this
paper, we already got 40% of the plots presented in the journal after forming
and solving the required delay di↵erential equations which were of the DDE
form.

Two types of DDEs were formed in the process:

• Lag is time dependent.

• Lag is time independent.

There aren’t many easy analytical methods present to solve the second type
of DDE but we could solve them using Matlab and have inserted them in
the term paper. These plots were in good agreement with those in the journal
and economic significance attached with all the graphs have been highlighted in
the Results and Discussion section. The Matlab codes used to get all the
plots have been attached in the codes file submitted with the term paper. After
the mid-term report, we got all the plots present there in the journal with
Matlab codes. After we were finished with verifying and plotting all the plots
present in the journal, we tried to have a deeper look at the model and tried to
alter and think of more equations that can make the model more realistic. One
of the significant change, we could think of was the change in Labor function,
thus extending the project to Level 2. The labor function used in the model
assumed labor to be constant and inelastic and that assumption can easily be
shed o↵.
In the paper the authors have assumed that the labor supply is constant and is
equal to one, i.e. l(t) = 1. We went forward to change that assumption because
in actual economies labor supply may change with time. We assumed that the
labor supply is a function of time and varies as 1 + exp( t). The reason of
making such an assumption was, that in mature economies over the period of
time the labor supply will saturate as the population will tend to saturate. The
thought behind such an assumption was that as the resources in any country
are constant, they can only support a maximum level of population.
On putting the above mentioned labor function in the DDE’s and plotting T (t)
i.e the age of the oldest machine in use with time and taking all other parameters
same, we got the Figure 1. We have explained the figure thus obtained with the
most suitable explanation according to us above in the paper.

References
[1] Raouf Boucekkine, Ornar Licandro and Christopher Paul, 1995
DIFFERENTIAL-DIFFERENCE EQUATIONS IN ECONOMICS: ON
THE NUMERICAL SOLUTION OF VINTAGE CAPITAL GROWTH
MODELS.Working Paper 95-59,Economic Series 29

8
Figure 3: Scalar constant lag DDE-I

Figure 4: Scalar constant lag DDE-I

9
Figure 5: Scalar constant lag DDE-I

Figure 6: Scalar constant lag DDE-II

10
Figure 7: Scalar constant lag DDE-II

Figure 8: Scalar constant lag DDE-II

11
Figure 9: Time-varying State Dependent DDE

Figure 10: Time-varying State Dependent DDE

12
Figure 11: Solow Model with i0 (t) = kexp{ t}

Figure 12: Solow Model with i0 (t) = kexp{ t}

13
Figure 13: Extended Solow Model with i0 (t) = kexp{ (t)t} and ! = 2⇡/T ⇤

Figure 14: Extended Solow Model with i0 (t) = kexp{ (t)t} and ! = ⇡/5

14
Figure 15: Extended Solow Model with i0 (t) = kexp{ (t)t} and ! = ⇡/5

Figure 16: Extended Solow Model with i0 (t) = kexp{ t} and ! = 2⇡/T ⇤

15
Figure 17: Extended Solow Model with i0 (t) = kexp{ t} and ! = 2⇡/T ⇤

Figure 18: Extended Solow Model with i0 (t) = kexp{ (t)t} and ! = ⇡/5

16
Figure 19: Extended Solow Model with i0 (t) = kexp{ (t)t} and ! = ⇡/5

17

You might also like