Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Project Muse 723895

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/333132532

How Does Dependency Ratio Affect Economic Growth In the Long Run? Evidence
from Selected Asian Countries

Article  in  The Journal of Developing Areas · May 2019


DOI: 10.1353/jda.2020.0015

CITATIONS READS

2 3,238

4 authors:

Sayema Haque Bidisha S M Abdullah


University of Dhaka University of Dhaka
37 PUBLICATIONS   186 CITATIONS    31 PUBLICATIONS   78 CITATIONS   

SEE PROFILE SEE PROFILE

Salina Siddiqua Mohammad Mainul Islam


University of Dhaka University of Dhaka
10 PUBLICATIONS   66 CITATIONS    53 PUBLICATIONS   206 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

SANEM_MJF Work on Recognition of Women's Unaccounted Work View project

Tobacco Control Capacity Programme (TCCP) View project

All content following this page was uploaded by Mohammad Mainul Islam on 31 March 2020.

The user has requested enhancement of the downloaded file.


How Does Dependency Ratio Affect Economic Growth In the Long
Run? Evidence from Selected Asian Countries

Sayema Haque Bidisha, S M Abdullah, Salina Siddiqua, Mohammad Mainul


Islam

The Journal of Developing Areas, Volume 54, Number 2, Spring 2020, (Article)

Published by Tennessee State University College of Business


DOI: https://doi.org/10.1353/jda.2020.0015

For additional information about this article


https://muse.jhu.edu/article/723895

Access provided at 31 Mar 2020 03:21 GMT from University of Dhaka


The Journal of Developing Areas
Volume 54 No. 2 Spring 2020

HOW DOES DEPENDENCY RATIO AFFECT


ECONOMIC GROWTH IN THE LONG RUN?
EVIDENCE FROM SELECTED ASIAN
COUNTRIES
Sayema Haque Bidisha
S M Abdullah
Salina Siddiqua
Mohammad Mainul Islam
University of Dhaka, Bangladesh

ABSTRACT

It has been widely accepted that the structure of population of a country can have significant
contribution towards its growth effort. With the proportional increase of relatively older group of
people in developed countries alongside the proportional rise of the youth population in the
developing part of the world, it is of paramount importance to understand the long term economic
consequences of this changed structure of population. Despite having a favorable demographic
composition, depending on public investment and policy strategy adopted, economies differ in terms
of their ability to attain demographic dividend. The issue is therefore an empirical one which requires
appropriate quantitative analysis. Against this backdrop, with the help of a 37-year long panel of 15
Asian countries, this study has made an attempt to understand the long run effect of changing
demographic pattern on economic growth of Asian countries. Despite increased importance of
demographic composition on labor supply, asset accumulation, savings, hence on the overall growth
process, there is no literature focusing on the relationship from a cross-country perspective and while
applying suitable methodological tools. Thus, this paper utilized econometric methods of panel data
constructed using World Development Indicators (WDI) and Health Nutrition and Population
Statistics (HNPS) from World Bank database to understand the linkage between dependency ratio
and per capita GDP growth. Following cross section dependence test stationarity property of the
variables has been diagnosed employing appropriate panel unit root tests. The short run and long run
parameters have estimated with the help of Mean Group (MG), Pooled Mean Group (PMG) and
Dynamic Fixed Effect (DFE) estimators. PMG estimates have been preferred over the DFE and MG.
It revealed the fact that for the Asian countries Dependency Ratio (DR) does not have any short run
impact on Per Capita GDP Growth (PCGDPG) but it has significant long run impact therefore with
an increase in working age population in comparison to the non-working group, there is a resulting
increase in per capita GDP growth of Asian countries. Since changed demographic composition can
only result in higher economic growth, if complemented by effective public policies, this paper
provides empirical evidences in favor of such policies of human resource development. Besides, in
order to reap the maximum benefit from working age population while increasing investment, it is
important to engage them in productive, growth enhancing activities and that requires investment in
employment generation.

JEL Classifications: J11, C10, C13, O0


Keywords: Dependency Ratio, Demographic Dividend, Panel Data, Panel Unit Root
Test, Dynamic Fixed Effect.
Corresponding Author’s Email Address: abdullahsonnet@gmail.com
48

INTRODUCTION

It has been widely accepted that the structure of population of a country can have
significant contribution towards its growth effort. With the proportional increase of
relatively older group of people in developed countries alongside the proportional rise of
the youth population in the developing part of the world, it is of paramount importance to
understand the long term economic consequences of this changed structure of population.
Any such change in the population structure not only is expected to bring about changes in
the composition of labor force, but is also likely to have important implications towards
investment, savings and above all economic growth of the countries. Globally the
experience of such demographic change is rather mixed as different countries can argue to
be at different stages of this transition. Besides, given the health and population policies
along with public investment in the crucial areas of human resource development, the
resulting effect of such transition on economic growth has been quite different too. The
issue of demographic dividend or the positive consequence of demographic transition on
growth is however a long run issue and can even occur over the course of several
generations. According to Fang and Wang (2005) over time demographic structure of a
country can change through several phases where at the first stage, high dependency ratio
of children is expected to be matched by an increase in the proportion of working people
in the second stage. Finally, a prevalence of high dependency ratio of the aged will rise in
the third phase. This change in demographic composition can have a number of effects; the
first is the positive effect of changed supply of labor due to change in working age
population. Secondly, it can help capital accumulation and hence GDP growth due to its
implication for personal savings and consumption (Kelly, 1973). On the other hand,
greater proportion of older people in the economy are assumed to increase public
expenditure due to greater spending in pension and medical expenses, which consequently
is expected to affect spending in productive sector in a negative manner (Fang and Wang,
2005).
As discussed, despite having a favorable demographic composition, depending on
public investment and policy strategy adopted, economies differ in terms of their ability to
attain demographic dividend. The issue is therefore an empirical one which requires
appropriate quantitative analysis. In this regard in context of Asia, on one hand there are
East Asian countries like South Korea, Thailand, Malaysia which have already experienced
beneficial effect of proportional increase of working age population on economic growth,
on the other hand there are South Asian countries like Bangladesh or Nepal or Pakistan
which are at the moment argued to go through a demographic transition, it is interesting to
understand the long run consequences of changed demographic structure on the growth
prospects of these countries. Regardless of the crucial importance of the linkage between
changed demographic structure and economic growth, to the extent of knowledge, the long
run relationship between demographic structure and economic growth has not been studied
properly in the literature. In particular, the issue has not been examined with necessary
empirical rigor and with suitable database. Against this backdrop, with the help of a 37-
year long panel of 15 Asian countries, this study has made an attempt to understand the
long run effect of changing demographic pattern on economic growth of Asian countries.
49

LITERATURE REVIEW

Studies relating the economic growth and population are plenty. Nevertheless, impact of a
variable reflecting the demographic transition such as dependency ratio on economic
growth has been little explored empirically. Vijayakumer (2013) have tried to examine the
association among the variables such as poverty, economic growth, agricultural and
industrial employment and dependency ratio. By using cross country data of 41 countries
selected from Asia, Latin America and Sub Saharan Africa region dependency ratio was
found to contain a significant positive impact on incidence of poverty. However, the
findings can be questioned because the regression results do not present the most important
diagnosis test result for cross country analysis which is heteroscedasticity. Should there be
any such problem the efficiency of the parameters will be distorted and the use of findings
for policy analysis will be misleading. Following relatively better estimation technique and
using only the countries from Sub – Saharan Africa, Fayissa and Gutema (2010) have tried
to explain the growth puzzle in light of demographic factors especially the dynamics of
dependency ratio. Based on Solow’s growth equation a panel data model has been
estimated using one way Fixed Effect and Random Effect as well as both way Fixed Effect
and Random Effect method. It was found that an increase in dependency ratio contains a
severe adverse effect on the per capita GDP growth rate. It was argued that due to the high
dependency ratio of the sample countries, they were constrained to have a stable
equilibrium at an income level lower than other countries around the world. Nevertheless,
the estimated growth model in the study to a large extent is inappropriate as it misses some
fundamental variables which accounted for growth now a days for example, social capital.
Since the model ignored those variables there could potentially be endogeneity problem in
the estimation results which was never discussed. In contrast using the similar
methodology, Basu et al. (2013) have tried to find out the effect of decreasing dependency
ratio on the BRICS (Brazil, Russia, India, China and South Africa) countries growth in
relation to few developed countries. It was found that the effect of initial level of working
age population and growth rate of that population have significant impact on growth.
Bloom et al. (2010) have given effort to measure the growth opportunity for
Nigeria offered by the country’s demographic transition by exploiting a 5-year country
level panel data from 1965 to 2005 and applying Ordinary Least Squares (OLS) and
Instrumental Variable (IV) method. They have estimated that Nigeria’s potential
demographic dividend has the ability to boost the country’s per capita income 30 per cent
or even more by the year 2030. However, to harness this potential benefit the study strongly
prescribed policy formulation regarding spending on health and education and creating
enough job opportunities for the potential skilled workers. Shifting the focus from world
to Asia, Iqbal et al. (2015) have used ARDL Bound Testing approach along with Error
Correction Model (ECM) and cointegration technique by using time series data from 1974
to 2011 for Pakistan to study the impact of demographic transition on economic growth.
They have found that the working age population has positive effect in the long run while
has negative effect in the short run on growth of Pakistan. On the other hand, Mody and
Aiyar (2011) have used a panel of 22 Indian States with a time period of ten-year intervals
from 1961 to 2001 to identify the impact of working age ratio on growth. Earlier, Bloom
and Finlay (2009) have examined the importance of demographic transition in explaining
cross country difference in economic growth with a special focus on East Asia. They have
used a 5 year panel from 1960 to 2005 with 14 Asian countries (East Asia, South East Asia
50

and South Asia). By applying OLS and IV method of estimation they found that
demographic factors are a significant contributor for economic growth in East Asia.
The literature review therefore highlights the necessity of the current research as
the crucial relationship between demographic structure of a country and economic growth
has not been examined in detail. For example, the existing researches are mostly descriptive
by nature, without considering the analytical requirement required for conducting such
analysis. Besides, to the extent of authors’ knowledge, no study has incorporated such a
vast pool of Asian countries as in this analysis.

DATA AND METHODOLOGICAL FRAMEWORK

DATA AND STATISTICAL SOFTWARE

The exercise has employed two secondary data sources to construct the complete panel,
namely World Development Indicators (WDI) and Health Nutrition and Population
Statistics (HNPS) from World Bank database. A total of 15 Asian countries have been
considered for the time period of 37 years (1980 to 2016). The selection of the countries is
primarily that of empirical and those Asian countries have been considered for which long
time series for all the relevant variables observed to be available. The list of countries
included Bangladesh, China, Indonesia, Iran, Israel, Japan, Jordan, Korean Republic,
Kuwait, Malaysia, Pakistan, Philippines, Sri Lanka, Syria and Thailand. With regards to
the selection of countries, as the purpose was to examine the impact from a long run
perspective the study targeted to select a reasonable number of countries while maximizing
the time period. Along with some other countries it had to sacrifice for example India from
sample because of having few variables on which the series for India started from late 90’s.

METHODOLOGICAL FRAMEWORK

The study is concerned with the impact of demographic dividend on economic growth. In
order to examine the linkage, particular focus has been given on the issue of how
dependency ratio contributes towards economic growth. Demographic transition results
fall in the dependency ratio, which opens the window of resource mobilization from
consumption to savings and hence investment that stimulates growth and eventually pave
the way towards achieving demographic dividend. In this connection, this study has
attempted to estimate the long run relationship between economic growth and dependency
ratio while controlling for both physical and social capital. The model that this paper has
estimated can be outlined in the following manner:

PCGDPGit = β0 + β1 DR it + β2 GDPSGFCFit + β3 GDPSPSEit + β4 GDPSTRADEit +


β5 SGER it + εit (1)

where, i = 1, 2, ------ N denotes the cross-sectional units and t = 1, 2, ------ T denotes time
period. Here, PCGDPGit is per capita GDP growth rate in country i at time t which is used
as a proxy of economic growth. DR it is dependency ratio measured in percentage,
GDPSGFCFit is share of gross fixed capital formation in GDP, GDPSPSEit is share of public
spending on education in GDP, GDPSTRADEit denotes Trade-GDP ratio and SGER it is
51

secondary gross enrollment ratio. Finally, εit denotes error term containing unobserved
factors.
In the above model, the coefficient of interest is β1 , magnitude of which measures
the impact of dependency ratio (demographic dividend) on per capita GDP growth. It is
expected that due to demographic transition, the ratio of economically active people to
inactive people increases and as a consequence there will be a scope of saving which is
expected be transmit into growth through the channel of investment. The coefficient hence
is expected to carry a negative sign.

ESTIMATION CHALLENGES AND PROPOSED REMEDIES

CROSS SECTION DEPENDENCE TEST

One of the plausible problems of estimating the model as proposed in equation (1) is the
possibility of cross-sectional dependence. In this connection, the significance of pair-wise
covariance among residuals in the standard panel regression required to be tested. The
hypothesis can be stated as follows:

𝐻0 : 𝜌𝑖𝑗 = 𝜌𝑗𝑖 = 𝐶𝑜𝑣(𝜀𝑖𝑡 , 𝜀𝑗𝑡 ) = 0, 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑡, 𝑖 ≠ 𝑗 (2𝑎)

𝐻1 : 𝜌𝑖𝑗 = 𝜌𝑗𝑖 = 𝐶𝑜𝑣(𝜀𝑖𝑡 , 𝜀𝑗𝑡 ) ≠ 0, 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑡, 𝑖 ≠ 𝑗 (2𝑏)

In this paper, four different tests namely Breuch-Pagan Lagrange Multiplier (LM) (1980),
Pesaran Cross Sectional Dependence (CD) (2004), Pesaran Scaled LM (2004) and Baltagi,
Feng and Kao Bias Corrected Scaled LM (2012) have been carried out to diagnose the
significance of pair-wise covariance among residuals.

PANEL UNIT ROOT TEST

In case of estimation of macro panel, increase in data frequency improves the power
property of stationarity tests. Therefore, panel based unit root tests are superior to their
pure time series counterparts. Im, Pesaran and Shin (1995) proposed demeaning procedure
(subtracting group mean from the data) to denounce the pair-wise covariance among
residuals i.e. contemporaneous correlation of the data. Hence, the study applied Im, Pesaran
and Shin (IPS) (Im et al.2003) and Levin, Lin and Chu (LLC) (Levin et al. 2002) panel unit
root test on the demeaned data for different variables. Both the tests at first estimate an
autoregressive (AR) process of the concerned variable as shown below:
𝑃𝑖

𝑦𝑖𝑡 = 𝜌𝑖 ∑ 𝑦𝑖𝑡−𝑗 + +𝑥𝑖𝑡′ 𝛿 + 𝜈𝑖𝑡 (3)


𝑗=1

Here, 𝑦𝑖𝑡 denotes the variable of concern for which stationary would be tested and 𝑥𝑖𝑡
stands for other control variables including individual trends and fixed effect. Then for
each variable the following Augmented Dickey Fuller (ADF) test regression has been fitted
to perform the test:
52

𝑝
Δ𝑦𝑖𝑡 = 𝛼𝑦𝑖𝑡−1 + ∑𝑗=1
𝑖
𝛽𝑖𝑗 Δ𝑦𝑖𝑡−𝑗 + 𝑥𝑖𝑡′ 𝛿 + 𝜀𝑖𝑡 (4)

The IPS panel unit root test use the assumption that the persistence parameter 𝜌𝑖 varies
across the countries. Thus, the unit root process is not identical and it could be different for
each of the cross sectional units. The null hypothesis here is stated as 𝐻0 : 𝛼𝑖 = 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑖.
The IPS test statistic to test this null is expressed as follows:
−1 ∑𝑁 𝐸(𝑡̅ (𝑝 )))
√𝑁(𝑡̅̅̅̅̅̅
𝑁𝑇 −𝑁 𝑖=1 𝑖𝑇 𝑖
𝑊̅̅̅̅̅̅
𝑡𝑁𝑇 = → 𝑁(0, 1) (5)
√𝑁−1 ∑𝑁 ̅
𝑖=1 𝑉𝑎𝑟(𝑡𝑖𝑇 (𝑝𝑖 ))

𝑡𝑁𝑇 denotes the average of the t – statistics for 𝛼𝑖 from the country specific ADF
Here, ̅̅̅̅
regressions, 𝑡𝑖𝑇𝑖 (𝑝𝑖 ). Thus,

(∑𝑁
𝑖=1 𝑡𝑖𝑇 (𝑝𝑖 ))
̅̅̅̅
𝑡𝑁𝑇 = 𝑖
(6)
𝑁

̅ (𝑝𝑖 )) and 𝑉𝑎𝑟(𝑡𝑖𝑇


𝐸(𝑡𝑖𝑇 ̅ (𝑝𝑖 )) are the expected value and variance of the ADF regressions t
– statistics respectively.

In contrast to IPS, LLC panel unit root test assumes that the persistence parameter is
identical for all countries i.e. 𝜌𝑖 = 𝜌. Therefore, in the above ADF test regression it
assumes 𝛼 = 𝜌 − 1 to test the null hypothesis, 𝐻0 : 𝛼 = 0. By using standardized and
autocorrelation and deterministic component free proxies for ∆𝑦𝑖𝑡 and 𝑦𝑖𝑡−1 the test derives
the estimates of 𝛼.

MODEL ESTIMATION

For estimating the dynamic heterogeneous panel with both large cross section (N) and time
span (T), one extreme option is to follow Dynamic Fixed Effect (DFE) estimation approach
and the other extreme would be Mean Group (MG) estimation approach developed by
Pesaran and Smith (1995). Pesaran, Shin and Smith (1997, 1999) combined both pooling
and averaging and have developed an intermediate option namely Pooled Mean Group
Estimation (PME). In comparison to the earlier two alternatives, this intermediate estimator
would allow different intercept, short – run coefficients and error variance for each country
while binding the long – run coefficients to be homogeneous across the countries. This
intermediate estimator is expected to capture the true dynamics in data as it is quite
reasonable to assume that though the short – run dynamics might be heterogeneous, the
long – run equilibrium relationship may be homogeneous across the countries (Tan, 2009).
The coefficients in all the three cases have been checked and the long – run slope
homogeneity was tested following Hausman test (Hausman, 1978).
Assuming that the panel is non stationary and the order of integration of variables
is not clear, an ARDL(1, 1, 1, 1, 1, 1) dynamic panel specification has been estimated to
53

derive the estimate of long – run coefficients. The error correction equilibrium
reparameterization of the related specification can be expressed as follows:

Δ𝑃𝐶𝐺𝐷𝑃𝐺𝑖𝑡 = 𝜙𝑖 (𝑃C𝐺𝐷𝑃𝐺𝑖,𝑡−1 − 𝜃0𝑖 − 𝜃1𝑖 𝐷𝑅𝑖𝑡 − 𝜃2𝑖 𝐺𝐷𝑃𝑆𝐺𝐹𝐶𝐹𝑖𝑡 −


𝜃3𝑖 𝐺𝐷𝑃𝑆𝑃𝑆𝐸𝑖𝑡 − 𝜃4𝑖 𝐺𝐷𝑃𝑆𝑇𝑅𝐴𝐷𝐸𝑖𝑡 − 𝜃5𝑖 𝑆𝐺𝐸𝑅𝑖𝑡 ) − 𝛿11𝑖 Δ𝐷𝑅𝑖𝑡 − 𝛿21𝑖 Δ𝐺𝐷𝑃𝑆𝐺𝐹𝐶𝐹𝑖𝑡 −
𝛿31𝑖 Δ𝐺𝐷𝑃𝑆𝑃𝑆𝐸𝑖𝑡 − 𝛿41𝑖 Δ𝐺𝐷𝑃𝑆𝑇𝑅𝐴𝐷𝐸𝑖𝑡 − 𝛿51𝑖 Δ𝑆𝐺𝐸𝑅𝑖𝑡 + 𝜀i𝑡 (7)

𝜇𝑖 𝛼10𝑖 +𝛼11𝑖
In the above error correction specification, 𝜙𝑖 = (1 − 𝜆𝑖 ), 𝜃0𝑖 = , 𝜃1𝑖 = ,
(1−𝜆𝑖 ) (1−𝜆𝑖 )
𝛼20𝑖 +𝛼21𝑖 𝛼30𝑖 +𝛼31𝑖 𝛼40𝑖 +𝛼41𝑖 𝛼50𝑖 +𝛼51𝑖
𝜃2𝑖 = , 𝜃3𝑖 = , 𝜃4𝑖 = and 𝜃5𝑖 = . Here, error correction
(1−𝜆𝑖 ) (1−𝜆𝑖 ) (1−𝜆𝑖 ) (1−𝜆𝑖 )
parameter is 𝜙𝑖 . It is expected that the speed of adjustment parameter would carry a
negative sign should the variable get back to the long run equilibrium. The long run
coefficients 𝜃0𝑖 , 𝜃1𝑖 , 𝜃2𝑖 , 𝜃3𝑖 , 𝜃4𝑖 and 𝜃5𝑖 are of main interest. Among all 𝜃1𝑖 is the measure
of long run elasticity of per capita GDP growth with respect to dependency ratio.

EMPIRICAL RESULTS

DESCRIPTIVE ANALYSIS

As discussed, the analysis has utilized data of 15 Asian countries which are quite
heterogeneous in terms of socio-economic structure as well as demographic profiles. Table
1 contains summary statistics of two key variables of the analysis, namely dependency ratio
and per capita GDP growth for the sample countries during the time period under analysis.
It can be observed that, among the economies highest average per capita GDP growth was
experienced by China with a value of 8.75 per cent which was followed by Korea, Thailand,
Sri Lanka, Indonesia and Malaysia. All these countries are found to have a low average
dependency ratio compared to other countries considered in the data set and it turns out
that the countries that usually tend to have a high average per capita GDP growth also
maintained low dependency ratio. The relationship between these two variables, with
certain exceptions therefore should be negative and that was also found in the correlation
among them in different countries.
54

TABLE 1: SUMMARY STATISTICS OF KEY VARIABLES

PCGDPG DR Corr.(DR,
Country
Mean Std. Dev. Mean Std. Dev. PCGDPG)
Bangladesh 2.75 1.87 73.68 12.37 -0.80
China 8.75 2.72 47.51 9.69 -0.17
Indonesia 3.78 3.68 61.36 10.11 0.11
Iran 0.45 7.79 70.24 22.85 -0.20
Israel 2.40 3.46 64.61 4.11 -0.03
Japan 1.76 2.33 48.85 5.55 -0.26
Jordan 1.42 5.25 83.02 15.31 -0.15
Korea Rep. 5.34 3.82 43.35 6.91 0.24
Kuwait 0.26 10.63 48.07 11.99 -0.16
Malaysia 3.56 3.58 61.55 9.38 0.06
Pakistan 2.21 1.90 81.04 8.33 0.15
Philippines 1.29 3.54 73.05 8.37 -0.52
Sri Lanka 4.01 2.24 55.20 6.89 -0.35
Syria 1.57 4.58 86.15 15.35 -0.12
Thailand 4.21 3.85 50.11 10.90 0.15
Source: Authors’ own calculation.

ECONOMETRIC ANALYSIS

CROSS SECTION DEPENDENCE AND PANEL UNIT ROOT TEST

As discussed in methodology, Table 2 contains results of different cross section


dependence tests of the variables used in the analysis. It is evident from the results that all
the concerned variables have contemporaneous correlation since the null hypothesis of no
cross section dependence can possibly be rejected for all variables in all four tests.

TABLE 2: TEST RESULTS FOR CROSS SECTIONAL DEPENDENCE OF THE


VARIABLES

Breusch - Pesaran - Scaled Bias Corrected Pesarn


Variables
Pagan LM LM Scaled LM CD
H0: No Cross - Section Dependence
Statistic 254.304* 10.302* 10.082* 6.408*
PCGDPG
Prob. 0.000 0.000 0.000 0.000
Statistic 2903.649* 193.125* 192.904* 42.879*
DR
Prob. 0.000 0.000 0.000 0.000
Statistic 586.444* 33.222* 33.002* 7.902*
GDPSGFCF
Prob. 0.000 0.000 0.000 0.000
Statistic 571.675* 32.203* 31.983* -2.197**
GDPSPSE
Prob. 0.000 0.000 0.000 0.028
Statistic 792.929* 47.471* 47.251* 9.164*
GDPSTRADE
Prob. 0.000 0.000 0.000 0.000
Statistic 2227.119* 146.440* 146.219* 45.734*
SGER
Prob. 0.000 0.000 0.000 0.000
Source: Authors’ own calculation; Note: * and ** Indicates 1 per cent and 5 per cent level of
significance respectively.
55

Thus, in order to take care of the cross-section dependence feature of the variables, the
exercise demeaned the variables by subtracting the cross-sectional average from the
original data and as shown in Table 3 and Table 4, two types of panel unit root tests namely
IPS and LLC have been applied. The IPS panel unit root test examines the existence of unit
root for each cross section units while LLC does so for the panel as a whole. As results in
Table 3 indicates, IPS panel unit root confirms that PCGDPG, DR and GDPSGFCF can be
treated as stationary and integrated of order zero i.e. I(0) variable. On the other hand,
GDPSPSE is nonstationary at trend specification but stationary at no trend specification.
However, first difference of GDPSPSE is stationary in both specification which indicates
that this variable can be treated as nonstationary but integrated of order one i.e. I(1).
Similarly as first difference of GDPSTRADE and SGER is stationary, these variables can
also be considered as nonstationary and I(1). The findings revealed by LLC test have found
to similar to a large extent. According to LLC panel unit root test PCGDPG and DR are
stationary at level and thus I(0) while all other variables are stationary at first difference
and hence could be characterized as I(1).

TABLE 3: PANEL UNIT ROOT TEST RESULTS OF THE VARIABLES

Im – Pesaran – Shin (IPS) Test for Panel Unit Root


Null: Panels Contain Unit Roots (Individual)
Variables Comments
Intercept Intercept and Trend
IPS W - Stat Prob. IPS W - Stat Prob.
PCGDPG -11.116* 0.000 13.424* 0.000 I(0)
DR -2.699* 0.000 -2.030** 0.021 I(0)
GDPSGFCF -2.953* 0.001 -4.186* 0.000 I(0)
GDPSPSE -2.351* 0.009 -1.049 0.146 D(GDPSPSE) is
Stationary in Both
D(GDPSPSE) -15.074 * 0.000 -14.298 * 0.000 Specification. So it
may be I(1)
GDPSTRADE -0.640 0.261 -1.381*** 0.083 D(GDPSTRADE) is
Stationary in Both
D(GDPSTRADE) -19.348* 0.000 -17.449* 0.000 Specification. So it
may be I(1)
SGER 0.725 0.765 -0.318 0.375 D(SGER) is
Stationary in Both
D(SGER) -9.925* 0.000 -7.421* 0.000 Specification. So it
may be I(1)
Source: Authors’ own calculation; Note: *, ** and *** Indicates significance at 1 per cent, 5 per
cent and 10 per cent level respectively.
56

TABLE 4: PANEL UNIT ROOT TEST RESULTS OF THE VARIABLES

Levin, Lin & Chu Test for Panel Unit Root


Null: Panels Contain Unit Roots (Common)
Variables Comments
Intercept Intercept and Trend
LLC t – Stat Prob. LLC t – Stat Prob.
PCGDPG I(0) -11.116* 0.000 -13.021* 0.000
DR I(0) -5.742* 0.000 -4.108* 0.000
GDPSGFCF D(GDPSGFCF) is
-0.589 0.277 -2.823* 0.002
Stationary in Both
D(GDPSGFCF) -12.281* 0.000 -10.726* 0.000 Specification. So it may
be I(1)
GDPSPSE -1.518** 0.064 -0.778 0.218 D(GDPSPSE) is
Stationary in Both
D(GDPSPSE) -14.437* 0.000 -13.771* 0.000 Specification. So it may
be I(1)
GDPSTRADE -0.332 0.369 -0.386 0.349 D(GDPSTRADE) is
Stationary in Both
D(GDPSTRADE) -18.542 * 0.000 -16.079 * 0.000 Specification. So it may
be I(1)
SGER -2.530* 0.005 -0.339 0.367 D(SGER) is Stationary in
Both Specification. So it
D(SGER) 8.552* 0.000 -6.654* 0.000 may be I(1)
Source: Authors’ own calculation; Note: *, ** and *** Indicates significance at 1 per cent, 5 per
cent and 10 per cent level respectively.

FIGURE 1: MODEL SELECTION SUMMARY


Schwarz Criteria
8.0

7.6

7.2

6.8

6.4

6.0

5.6
ARDL(1, 1, 1, 1, 1, 1)

ARDL(2, 1, 1, 1, 1, 1)

ARDL(3, 1, 1, 1, 1, 1)

ARDL(4, 1, 1, 1, 1, 1)

ARDL(1, 2, 2, 2, 2, 2)

ARDL(2, 2, 2, 2, 2, 2)

ARDL(3, 2, 2, 2, 2, 2)

ARDL(4, 2, 2, 2, 2, 2)

ARDL(1, 3, 3, 3, 3, 3)

ARDL(2, 3, 3, 3, 3, 3)

ARDL(3, 3, 3, 3, 3, 3)

ARDL(4, 3, 3, 3, 3, 3)

ARDL(2, 4, 4, 4, 4, 4)

ARDL(1, 4, 4, 4, 4, 4)

ARDL(3, 4, 4, 4, 4, 4)

ARDL(4, 4, 4, 4, 4, 4)
57

From Table 3 and Table 4, it can be deduced that the integration order of all the variables
are not unique but none of the variables are integrated of order two i.e. I(2). Given such
characteristics of variables, the long run coefficients of the variables can be estimated with
the help of an ARDL dynamic panel specification. The order of the model has been selected
by minimizing Schwarz Bayesian Information Criterion (SIC) as it is established in the
literature that whenever the sample size is large, SIC provides more parsimonious and
consistent result than any other information criterion. For each of the variables four years
have been chosen as the optimum maximum lag length. Figure 1 shows that the ARDL
model with first lag for all variables as the order is the one which have minimum
information criterion and hence considered as the most appropriate one.
Table 5 contains the estimation results where ARDL (1, 1, 1, 1, 1, 1) model has
been estimated by suing three different methods; Mean Group (MG), Pooled Mean Group
(PMG) and Dynamic Fixed Effect (DFE). The two natural concerns with large cross section
and large time dimension dynamic panels are nonstationarity and the slope homogeneity
(Blackburne and Frank, 2007). For estimating nonstationary panels, mean group and
pooled mean group are widely applied econometric techniques (Blackburne and Frank,
2007; Martinez-Zarzoso and Bengochea-Morancho, 2004; Frank, 2005; Tan, 2009; Genc
and Artar, 2014; Njoupouognigni and Ndambendia, 2010). The PMG estimators although
allow the short run dynamics (short run coefficients) to be heterogeneous across the
countries, the long run coefficients could remain homogeneous for all countries. In such
cases, the long run coefficients are generally considered as the most crucial estimates to
consider. As the results reveal the estimated long run elasticity of PCGPG with respect to
DR is properly signed and statistically significant in PMG and DFE estimation method.
The magnitude of the coefficient however has been found to be different in different
methods. In particular according to PMG estimator for 1 per cent decrease in DR would
lead to on an average 0.093 percent increase in PCGDPG in the long run holding all other
variables constant. As per the results of DFE this impact should be 0.076 percent.

TABLE 5. ESTIMATION RESULTS

Pooled Mean Group


Mean Group Estimation Dynamic Fixed Effect
Estimation
Variables
Std. Std.
Coeff. Coeff. Std. Error Coeff.
Error Error
Long Run Estimates with D(PCGDPG) as Dependent Variable
DR -0.066 0.085 -0.093* 0.018 -0.076* 0.022
GDPSGFCF 0.019 0.099 0.167* 0.021 0.050 0.057
GDPSTRADE 0.023 0.052 0.013** 0.006 -0.013 0.019
SGER 0.037 0.097 -0.062* 0.017 -0.049*** 0.026
GDPSPSE 1.534* 0.563 0.429** 0.180 1.333** 0.549
Short Run Coefficients with D(PCGDPG) as Dependent Variable
Constant -7.152 16.228 5.792* 0.568 4.500*** 2.298
EC -1.104* 0.062 -0.925* 0.062 -0.800* 0.046
D (DR) 0.377 0.477 -0.136 0.324 -0.299 0.322
D (GDPSGFCF) 0.451* 0.113 0.466* 0.113 0.366* 0.109
D (GDPSTRADE) 0.058 0.048 0.059 0.054 -0.026 0.039
D (SGER) -0.199 0.161 0.004 0.105 -0.162** 0.065
D (GDPSPSE) -1.994* 0.472 -1.671* 0.531 -1.627* 0.582
Source: Authors’ own calculation; Note: *, ** and *** Indicates significance at 1 per cent, 5 per
cent and 10 per cent level respectively.
58

The speed of adjustment estimates in all three models has found to be also properly signed
and statistically significant. The different values of this error correction parameter indicate
the significant different short run dynamics. The PMG estimator constrains the long run
parameters to be homogeneous across the panels. The difference in these models can be
examined by the help of Hausman test (Table 6). The calculated value of Hausman test
statistic is measured to be 6.96 with probability 0.223. Thus, there was not enough evidence
to reject the null arguing that the difference in MG and PMG estimates are not systematic
and the restriction of long run homogeneous coefficients is valid. Consequently, the PMG
estimators which are considered as efficient under null would be preferred. The fact is also
evident as lower standard error and slower speed of adjustment coefficient has been
obtained while PMG is implemented in relation to MG. Because of the endogeneity
between error term and lagged dependent variable simultaneity bias could be present in
DFE. In order to examine the extent of endogeneity problem the similar Hausman test as
before has been performed between MG and DFE estimates. As the results show here also
the null hypothesis that the difference in MG and DFE estimates are not systematic could
not be rejected. Hence the endogeneity problem would be minimum and DFE is preferred
over MG.

TABLE 6. TEST OF DIFFERENCE BETWEEN MODELS

Hausman Test: H0: Difference in Mean Group and Pooled Mean Group Estimates are not
Systematic
Chi –Square Statistic Prob.
6.96 0.223
H0: Difference in Mean Group and Dynamic Fixed Effect Estimates are not Systematic
8.61 0.125
Source: Authors’ own calculation

So, both PMG and DFE estimates would be preferred over MG and from a theoretical point
of view the most appropriate estimation strategy would be to use PMG estimates. The
rationale is that, although in the short run it is rational to assume that the effect of DR on
PCGDPG may differ from country to country (and most likely immediate impact would be
absent), due to increase in saving, women empowerment, increase in physical as well as
social capital in the long run decrease in DR should enhance PCGPG in all countries. The
PMG estimates presented in Table 5 is in line with the theory revealing the fact that for the
Asian countries DR does not have any short run impact on PCGDPG but it has significant
long run impact. While looking at other coefficients of PMG estimation, it is evident that
GDPSGFCF, GDPSTRADE, SGER and GPDSPSE have significant long run impact on
PCGDPG, where the sign of SGER is in contrast to the prior expectation.

CONCLUSIONS

This particular analysis reveals that there exists a statistically significant negative
association between dependency ratio and per capita GDP growth-with a reduction in
dependency ratio or an increase in working age population in comparison to the non-
working group (those who are below 15 years and above 65 years), there is a resulting
increase in per capita GDP growth of the selected Asian countries. This paper expects to
fill a gap in the literature while addressing this crucial relationship with the help of
59

advanced econometric estimation techniques. This finding therefore emphasizes on


policies paving the way towards achieving demographic dividend. It is needless to say that,
changed demographic composition can only result in higher economic growth, if
complemented by effective public policies. On one hand, greater spending in health and
family planning programs is of prime importance for reducing birth rate as well as death
rate and also to combat different illnesses affecting productivity of the labor force. On the
other hand, the role of education as well as skill training programs should emphasize in
attaining demographic dividend. This paper provides empirical evidences in favor of such
policies of human resource development. Besides, in order to reap the maximum benefit
from working age population while increasing investment, it is important to engage them
in productive, growth enhancing activities and that requires investment in employment
generation. Given the current contribution of the study, it is worth mentioning that there
are areas where further contribution can be made. From methodological point of view as
cross section dependence was detected it would have been optimal to apply panel unit root
tests which are robust against this problem. Incorporating this would enhance the strength
of the exercise. However, demeaning procedure would at least minimize the problem. Also,
from practical point of view as long as demographic variables are concerned leaving India
behind could limit the generalization of the findings. Nevertheless, incorporating all these
issues could lead to a further research.

REFERENCES

Baltagi, B.H., Feng, Q. and Kao, C., 2012. A Lagrange Multiplier test for cross-
sectional dependence in a fixed effects panel data model. Journal of Econometrics, 170(1),
pp.164-177.
Basu, T., Barik, D. and Arokiasamy, P., 2013. Demographic determinants of
economic growth in BRICS and selected developed countries. In XXVII IUSSP
International Population Conference (IUSSP 2013).
Blackburne III, E.F. and Frank, M.W., 2007. Estimation of nonstationary
heterogeneous panels. The Stata Journal, 7(2), pp.197-208.
Bloom, D., Finlay, J., Humair, S., Mason, A., Olaniyan, O. and Soyibo, A., 2010,
June. Prospects for economic growth in Nigeria: a demographic perspective. In IUSSP
Seminar on Demographics and Macroeconomic Performance, Paris, France.
Bloom, D.E. and Finlay, J.E., 2009. Demographic change and economic growth
in Asia. Asian Economic Policy Review, 4(1), pp.45-64.
Breusch, T.S. and Pagan, A.R., 1980. The Lagrange multiplier test and its
applications to model specification in econometrics. The Review of Economic Studies,
47(1), pp.239-253.
Fayissa, B. and Gutema, P., 2010. Dependency Ratio and the Economic Growth
Puzzle in Sub-Saharan Africa (No. 201010). Middle Tennessee State University,
Department of Economics and Finance.
Fang, C. and Wang, D., 2005. Demographic transition: implications for growth.
The China boom and its discontents, p.34.
Frank, M.W., 2005. Income inequality and economic growth in the US: A panel
cointegration approach. Sam Houston State University Working Paper, pp.05-03.
Genc, E.G. and Artar, O.K., 2014. The effect of exchange rates on exports and
imports of emerging countries. European Scientific Journal, ESJ, 10(13).
60

Hausman, J.A., 1978. Specification tests in econometrics. Econometrica, Journal


of the Econometric Society, pp.1251-127.
Pasaran, M.H., Im, K.S. and Shin, Y., 1995. Testing for Unit Roots in
Heterogeneous Panels (No. 9526). Faculty of Economics, University of Cambridge.
Im, K.S., Pesaran, M.H. and Shin, Y., 2003. Testing for unit roots in
heterogeneous panels. Journal of econometrics, 115(1), pp.53-74.
Iqbal, K., Yasmin, N. and Yaseen, M.R., 2015. Impact of demographic transition
on economic growth of Pakistan. Journal of Finance and Economics, 3(2), pp.44-50.
Kelley, A.C., 1973. Population growth, the dependency rate, and the pace of
economic development. Population Studies, 27(3), pp.405-414.
Levin, A., Lin, C.F. and Chu, C.S.J., 2002. Unit root tests in panel data:
asymptotic and finite-sample properties. Journal of econometrics, 108(1), pp.1-24.
Martı́nez-Zarzoso, I. and Bengochea-Morancho, A., 2004. Pooled mean group
estimation of an environmental Kuznets curve for CO2. Economics Letters, 82(1), pp.121-
126.
Mody, M.A. and Aiyar, M.S., 2011. The demographic dividend: Evidence from
the Indian states (No. 11-38). International Monetary Fund.
Njoupouognigni, M., 2010. Foreign aid, foreign direct investment and economic
growth in Sub-Saharan Africa: Evidence from pooled mean group estimator (PMG).
International journal of economics and finance, 2(3), p.39.
Pesaran, M.H. and Smith, R., 1995. Estimating long-run relationships from
dynamic heterogeneous panels. Journal of econometrics, 68(1), pp.79-113.
Im, K.S., Pesaran, M.H. and Shin, Y., 1997. Testing for Unit Roots in
Heterogeneous Panels', University of Cambridge. Working Paper, 9526.
Pesaran, M.H., Shin, Y. and Smith, R.P., 1999. Pooled mean group estimation of
dynamic heterogeneous panels. Journal of the American Statistical Association, 94(446),
pp.621-634.
Pesaran, M.H., 2004. General diagnostic tests for cross section dependence in
panels (No. 1229). CESifo Working Paper.
Tan, K.Y., 2009. A pooled mean group analysis on aid and growth. Applied
Economics Letters, 16(16), pp.1597-1601.
Vijayakumar, S., 2013. An Empirical Study on the Nexus of Poverty, GDP
Growth, Dependency Ratio and Employ-ment in Developing Countries. Journal of
Competitiveness, 5(2), pp.67-82.
The World Bank, 2017. World Development Indicators (WDI), Washington DC,
USA
The World Bank, 2017. Health Nutrition and Population Statistics (HNPS),
Washington DC, USA

View publication stats

You might also like