Bachelor Thesis Economics and Business Specialisation: Economics and Finance
Faculty of Economics and Business Academic year: 2017 – 2018
The effect of population ageing on long run
economic growth of developed countries
Student Name: Richard Yongjoon Youn Student Number: 10714332
Abstract
Population ageing sprung from longevity and low total fertility rates has been present in the developed OECD member countries since the 1970s and it has been accelerating. Although multitudinous empirical and theoretical studies have been done with respect to the effects of population ageing on economic growth, the arguments are conflict and diffuse. This impedes seeking for a solution and policy making with an integrated vision. This being the case, this paper both theoretically and empirically investigated the main mechanisms of the conflict arguments to generalise the effects of population ageing on economic growth for the policy implication. To analyse, panel data regression models were employed with the control of autocorrelation and cross sectional dependence. The empirical results demonstrate that an increase in life expectancy has negative effects on economic growth whereas low fertility rates have positive effects on economic growth.
Statement of Originality
This document is written by Richard Yongjoon Youn who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of contents
1. Introduction p. 4
2. Population ageing: facts and trends p. 6
3. Literature review p. 8
3.1. The positive effects of population ageing on economic growth p. 8
3.2. The negative effects of population ageing on economic growth p. 10
3.3. Conclusion on the literature p. 11
3.4. Hypotheses p. 12 4. Data description p. 14 4.1. Data sources p. 14 4.2. Data variables p. 15 4.2.1. Dependent variable p. 15 4.2.2. Independent variables p. 16 4.2.3. Control variables p. 16 4.2.4. Dummy variables p. 18
4.2.5. Balanced panel data p. 19
4.3. Variable statistics p. 19
5. Methodology p. 22
5.1. Pretests p. 23
5.1.1. Cross sectional independence p. 23
5.1.2. Stationarity p. 26
5.1.3. Autocorrelation p. 27
5.1.4. Hausman test p. 28
5.2. Empirical models p. 30
6. Empirical results and interpretation p. 30
7. Limitations and discussion p. 34
8. Conclusion p. 35
Reference p. 36
1. Introduction
The global age structure has been rapidly metamorphosing for a recent few decades and the world is entering an unprecedented territory of population ageing (Phang, 2011). According to the world population projection of the United Nations (UN), the number of the world population above age 60 is expected to be doubled by 2050 (United Nations, 2017). The UN (2017) also stated that an increase in life expectancy and a decrease in total fertility rates are the two main causes of the global ageing population. In general, this ongoing demographic transition to an aged society is expected to affect economic growth negatively in the long run, and stands in need of an urgent global solution to minimize the effects (Auerbach, Hagemann, Kotlikoff, &
Nicolette, 1989; Borsch-supan, 2002; Bloom, Canning & Fink, 2010; Phang, 2011; and World Bank, 1994).
Among them, the most developed of the OECD member countries are at the centre of the heated debates regarding the effects of population ageing for two reasons (Bloom et al, 2010). First, these countries have been far below the level of the global average fertility rates of 2.5 births per woman since the 1970s and the level of the generation replacement rates, for which a population replaces itself from the current generation to the next, of 2.1 births per woman since the 1980s (United Nations, 2017). Along with globally increasing life expectancy, the low fertility bespeaks population ageing of these countries has been faster than other countries (United Nations, 2017). Secondly, as the population ageing has been present for a recent few decades in these countries, numerous studies have been done with respect to economic consequences (Phang, 2011). Howbeit, the results are mainly bifurcated into positive and negative effects of the population ageing on economic growth sprung from different economic theories (Bloom et al, 2010).
The arguments of negative effects of ageing population on economic growth have been deemed as orthodoxy (Nagarajan, Teixeira, & Silva, 2016). Most of the literature argues on the negative effects focusing on the circular relationship between macroeconomic variables (Phang, 2011). This perspective shares a common framework that declining fertility rates and increasing longevity have negative effects on economic growth through a set of long run mechanism: decreasing labour input due to the low fertility rates; decreasing savings, consumption, and
labour productivity influenced by increasing life expectancy; increasing social welfare cost; increasing old age dependency ratio; and stagnated economic growth (Phang, 2011). Auerbach et al (1989); World Bank (1994); OECD (1998); Borsch-supan (2002); Bloom et al (2010); and Phang (2011) all addressed that the low fertility rates and longevity have negative impacts on economic growth of developed countries by employing the long run mechanism.
On the other hand, the arguments of positive effects on economic growth focalise mostly on the ‘endogenous’ growth theory developed by Romer (1986) and Lucas (1988). In this framework, investments in knowledge and human capital accompany a higher productivity of output, and population ageing induce these investments to supplement decreasing labour force (Phang, 2011). Moreover, women in developed countries tend to have fewer children than developing countries to actively participate in the labour market, and contribute to positive economic growth by investing in knowledge and skills instead of in children (Fougere & Merette, 1999, p.412). Fougere and Merette (1999); Futagami and Nakajima (2001); Elgin and Tumen (2012); and Guo, Liu and Liu (2016) all concluded that the low fertility rates have positive effects on economic growth for developed countries by employing the endogenous growth model.
Despite the fact of the acceleration of population ageing, it lacks an integrated vision of the policy implication (Nagarajan et al, 2016). Most importantly, Nagarajan, Teixeira, and Silva (2016) analysed 54 papers on the topic of the effects of population ageing on economic growth, and 31 out of 54 papers were theoretical studies. Furthermore, they concluded that the number of the high quality empirical literature is scarce to reach on the consensus between scholars and policymakers (Nagarajan et al, 2016). It is therefore salient to delve into the extent of which the concurrence is empirically arrived at the long run effects of the population ageing on economic growth for developed OECD member countries for the policy implication making the best use of the past data of these countries. Thus, this paper aims at investigating following research
question: “To what extent population ageing in most developed OECD countries has affected on economic growth during the period from 1961 to 2015?”.
So as to answer the research question, annual balanced panel data of selected twenty countries are used. All the datasets of variables are manually constructed from World Bank and
OECD databases for the period from 1961 to 2015. Most of the literature with respect to the effects of population ageing on economic growth paid attention on life expectancy and total fertility rates as independent variables. However, as it is mentioned, the dissensus is sharpley made due to different theoretical assumptions. Thus this paper mainly focuses on the empirical test of the general arguments and additionally sheds light on the conflicting arguments by employing control variables for the period from 1978 to 2013 to investigate to what extent the debates are empirically verified for the policy implication.
This paper is structured as follows: section two will briefly visualise the facts and trends of causes of population ageing: life expectancy and total fertility rates. With this section two, it is confirmed that developed OECD member countries are apposite subjects of an empirical test. In
section three, the literature reviewwill provide a sufficient background information of debates
regarding the effects of population ageing on economic growth. After this, the hypotheses will be derived on the basis of the literature. The data parameterisation and the methodology will be followed in sections four and five respectively. The test results and the interpretation will be explained in section six and limitations and discussion will be followed in section seven. Finally the conclusion will be in section eight.
2. Population Ageing: facts and trends
This section depicts the facts and trends of main causes of population ageing: total fertility rates and life expectancy. The structure of the global population has been changing since the 1960s, most developed OECD member countries have been below the level of the global average fertility rates of 2.5 births per woman since the 1970s and the generation replacement fertility
rates of 2.1 births per woman since the 1980s (UN, 2017).Figure 2.1 depicts the trend of fertility
rates of the 20 selected most developed OECD member countries between 1961 and 2015. In this figure 2.1, the total fertility rates of selected countries and the average of OECD member
countries except for Chile and South Korea have been lower than the world average for all periods. This is in line with the aforementioned debate that population ageing of selected countries has been faster than developing countries.
Figure 2.1 Total fertility rates between 1961 and 2015
Source: Graphs constructed by author based on data from World Bank (2018)
Figure 2.2 visualises the global trend of increasing life expectancy. The life expectancy of the world in average increased from 52 to 72 years whereas the growth in the OECD member countries in average is from 67 to 80 years. In addition, except for Chile and South Korea, the life expectancies of selected countries have been higher than the world average for all periods. This indicates that the natural mortality in the selected countries has been being delayed and the proportion of the old populace has been increasing. Combined with decreasing total fertility rates, population ageing in these countries have been accelerating since the 1970s as the UN (2017) stated. Hence, to analyse empirical effects of the population ageing on economic growth, the most developed OECD countries are apposite.
Figure 2.2Life expectancy between 1961 and 2015
Source: Graphs constructed by author based on data from World Bank (2018)
3. Literature review
In this section, the crux of the arguments concerning the effects of population ageing on economic growth are discussed. First, inasmuch as the literature and arguments are generally bifurcated into positive and negative economic consequences of population ageing, sections 3.1 and 3.2 comprehensively examine the debates of positive and negative effects in literature respectively. Afterwards, the conclusion of the literature review is followed.
3.1. The positive effects of population ageing on economic growth
The foundation of optimistic view on the relationship between population ageing and economic growth is based on the ‘endogenous’ growth theory of Romer (1986) and Lucas (1988). The main idea of this theory assumes the knowledge and human capital are the prime movers of economic growth and are accumulated as an economy develops (Fougere & Merette, 1999). Thus a return to investment in knowledge and human capital is increasing and this assumption
creates a material difference with the pessimistic view of neoclassical assumptions of diminishing return to investment in knowledge and capital (Fougere & Merette, 1999).
Population ageing within this framework induces investments into human capital and technology as labour supply becomes a scarce production factor, and hence productivity growth (Phang, 2011). Moreover, in developed countries, active labour participation of women curtails birth rates further and promotes an investment in human capital for themselves (Fougere & Merette, 1999). As a consequence, ageing population increases the investment, productivity, labour supply of women and is salutary to economic growth in the long run with the optimistic view (Fougere & Merette, 1999).
Laying emphasis on the endogenous growth theory of Romer (1986) and Lucas (1988), Fougere and Merrete (1999) theoretically analysed overlapping generation (OLG) models in order to examine the effects of low fertility rates on economic growth for seven industrialised OECD countries. The authors compared OLG models with and without featuring endogenous growth theory, and the result suggested that the model with the endogenous growth theory induced investments in human capital and economic growth (Fougere & Merrete, 1999). Moreover, Futagami and Nakajima (2001) unearthed theoretical evidence of the positive economic growth when population ages, using general equilibrium model of life cycle savings with the endogenous growth theory. More recently, Elgin and Tumen (2012) conducted a panel regression analysis with the framework of the endogenous growth model of 50 countries for 60 years. The authors addressed that the low fertility rates lead to investments in human capital and a betterment of economic growth as a consequence (Elgin & Tumen, 2012).
Notwithstanding theoretical and empirical findings, there have been criticisms regarding endogenous growth framework. Firstly, a number of empirical studies resulted in collective failure in explaining convergence of countries of similar economic situations (Sachs, 1997). Secondly, Parente (2001) argued the endogenous growth theory is an empirically weaker model than exogenous growth theory concerning income divergence between countries. Thirdly, Krugman (2013) asserted that the endogenous growth theory makes immoderately many assumptions for unmeasurable variables, which precludes finding empirical evidence. Last but not least, Prettener (2013) found that the negative effects of the low fertility rates on economic
growth in the long run whereas the life expectancy had positive effects with the endogenous growth framework.
3.2. The negative effects of population ageing on economic growth
Contrary to the optimistic view, most of the literature advocates the negative relationship between population ageing and economic growth (Nagarajan et al, 2016). This viewpoint assumes low fertility rates and increasing life expectancy as the two principal sources of population ageing, and has a set of interrelated mechanism: decreasing labour force due to declining birth rates; decreasing savings, consumption, and labour productivity affected by longevity; increasing public expenditure; and stagnated economic growth (Phang, 2011). The slow economic growth overburdens the care for older people as the expenditure on the old increases whereas the income decreases for both private and public sectors (Phang, 2011). Furthermore, a return to investment in physical and human capital within this framework is diminishing as economies develop in contrast to the optimistic view (Fougere & Merrete, 1999). Therefore, population ageing with the pessimistic view exacerbates economic growth triggering a vicious cycle between variables (Auerbach et al, 1989; Borsch-supan, 2002; Bloom et al, 2010; Nagarajan et al, 2016; OECD, 1998; Phang, 2011; and World Bank, 1994).
In conjunction with the pessimistic view, Walder and Doring (2012) and Velarde and Hermann (2014) stressed that population ageing dwindles income per capita of all generations due to decreasing number of workers in a family, and affects households’ consumption and saving patterns to decline with lessened disposable income. Nardi, French, and Johns (2010), Aguila, Attanasio, and Meghir (2011), Ewijk and Volkerink Rohwedder (2013) and Velarde and Hermann (2014) empirically demonstrated that the consumption and the saving rates of
households substantially decreased after the retirement due to the fall in disposable income for developed countries. Diaz-Gimenez and Diaz-Saavedra (2009) subsumed public expenditure on retirement to consumption and saving patterns. For these authors, more educated workers generally pay higher payroll taxes during their working lives and receive the greater pensions after the retirement (Nagarajan et al, 2016). Thus an upsurge in retirement of educated workers due to ageing population augments the government’s expenditure on retirement (Nagarajan et al,
2016). Withal, Dostie (2011); Göbel and Zwick (2012); Lisenkova, Merette, and Wright (2013); and Mahlberg, Freund, Crespo, and Prskawetz (2013) contended that the ageing of workers lowers the level of productivity of any worker considering diminishing physical ability and it further slows down the economic growth. In aggregate, the arguments of consumption, saving, and public expenditure converge on an increase in old age dependency ratio, meaning that the smaller number of workers are obliged to care for the old (Nagarajan et al, 2016).
Nonetheless, there exist counterviews on the negative effects of population ageing from the optimistic view. Alders and Broer (2004) and Hock and Weil (2012) argue that the decline in the labour supply raises the wages of all generations as the labour becomes scarce and the
working age group is willing to abjure having children to maintain the consumption and saving rates with higher wages. According to these authors, population ageing will not affect
households’ consumption and saving patterns (Nagarajan et al, 2016). Furthermore, Hurd and Rohwedder (2011) and Okumura and Usui (2014) asserted that retirement ages as well as public expenditure on retirement differ by the country specific policies. Lastly, Dostie (2011); Gobel and Zwick (2012) and Mahlber et al (2013) empirically demonstrated that the productivity level of old aged workers was significantly positive in the number of sectors of developed countries and it indicated that no adamant empirical evidence of a negative impact of population ageing on labour productivity.
3.3. Conclusion of literature review
In this section, the literature was reviewed with respect to the positive and negative effects of population ageing on economic growth. The optimistic view assumes that the endogenous growth theory that argues investments in human capital and knowledge are the key factors
invigorating an economy and returns to accumulated investments are positive. Within this theory, the population ageing induces the investments in the key factors to supplement decreasing labour supply and thus the higher productivity propels economic growth. Furthermore, consumption and saving patterns are not affected as the working age group is willing to abjure having children to maintain their standard of living. Lastly, the public expenditure is contingent on a country
specific policy and is not necessarily increasing. Hence with the optimistic view, ageing population is a beneficiary to economic growth.
On the other hand, the pessimistic view focus on a set of interrelated mechanism between factors. Most of the literature argued that consumption, saving, labour force, labour productivity, and disposable income decline and public expenditures on the elderly increase as population ages. The arguments of negative effects of population ageing converged on the increase in old age dependency ratio. Thus, changes in variables overburden the working age groups with a vicious cycle of old age dependency ratio and deteriorate economic growth. Figure 3.1
summarises the literature review. In this figure 3.1, the changes in consumption, saving, labour supply, labour productivity, returns to investment, old age dependency and public expenditure are contradicting between the optimistic and pessimistic view.
Figure 3.1 Summary of literature review
Source: Figure constructed by author based on literature review
3.4. Hypotheses
The juxtaposition of the optimistic perspective with the pessimistic perspective in the literature review obscures drawing hypotheses. Both theoretical and empirical studies are conflicting and the direction of the effects are ambiguous. Nonetheless, the preponderance is generally
established in the pessimistic view. The null hypotheses in this paper are thus constructed selectively in accordance with the optimistic view aiming at rejecting the null hypotheses. Consequently, the alternative hypotheses advocate the pessimistic view.
The hypotheses for the first group is motivated by the general literature review. Most literatures expect negative effects of population ageing on economic growth and focus on life expectancy and total fertility rates as independent variables. Hence, using the past data of developed OECD member countries where the ageing population has been accelerating, empirical regressions of the first group will be conducted following the literature during the period from 1961 to 2015. Furthermore, an exogenous effect of global financial crisis in 2008 is expected to affect the economic growth. The motivation of the global crisis will be followed in section 4.2.2.
1st group
1st hypothesis: a significant positive correlation between life expectancy and economic growth is expected.
2nd hypothesis: a significant positive correlation between low total fertility rates and economic growth is expected.
The hypotheses for the second group also follow the literature. Consumption and saving of households, health expenditures, labor force, labor productivity and old age dependency ratio have been debated but the opinions are in discord. These variables are expected to have effects on the independent and dependent variables according to the literature. Hence to clarify the relationship between dependent and independent variables, control variables are created and controlled for the second group. The time period and countries are lessened from 1978 to 2013 and to 16 countries respectively due to the data availability.
2nd group
1st hypothesis: a significant positive correlation between life expectancy and economic growth is expected without control variables.
2nd hypothesis: a significant positive correlation between low total fertility rates and economic growth is expected without control variables.
3rd hypothesis: a significant positive correlation between life expectancy and economic growth is expected with control variables.
4th hypothesis: a significant positive correlation between low total fertility rates and economic growth is expected with control variables.
4. Data description
This section provides descriptions of the obtained data. First, the source of the data from where the time series and cross sectional variables are collected is specified. Second, all different variables of the empirical models are defined. Third, statistical descriptions are provided for the empirical analysis.
4.1. Data sources
To answer the research question, annual panel data of 35 OECD member countries which are also the most developed countries for the time period from 1961 to 2015 were employed. The empirical trends figured in section two and literature review elucidated appropriacy of these countries as the subject of an empirical test that the population ageing has been present in these countries for a recent few decades. Therefore, past data of these countries facilitate investigating the effects of population ageing on economic growth and inferencing of a policy implication. A broader timespan is preferred to empirically test the effects of population ageing as it takes a long time for an influence. The selected time period of 55 years was the possible longest period concerning the data availability and is predicted to provide meaningful information. All the data were available from the World Bank and OECD databases and are manually constructed.
Firstly, from the data of 35 OECD member countries obtained from the World Bank and OECD datasets, 20 countries were included in the balanced dataset. The reason for exclusion of
the 15 countries is missing data for at least one decade andfrequently omitted data within the
database. The 20 included countries are both on the lists of IMF advanced economies and high income OECD member countries which are considered as the most developed countries in general. Hence, the balanced panel data set is N= 20 (Countries) and T=55 (Annual time period).
This dataset contains dependent and independent variables (Hereafter referred to as group 1). Secondly, to obtain the data for control variables, time periods and the number of the countries of the data were lessened from 1978 to 2013 and to 16 countries respectively due to data
availability. As a consequence, the dataset of control variables is constructed with N=16 and T=36 (Hereafter referred to as group 2). Both datasets are employed in this paper for different
tests.Table 4.1 in Appendix provides the overall information of the two datasets.
4.2. Data variables
For all variables in this paper, the subscript “i” refers to a specific country and “t” represents the time period of a specific year. The variables of interest in this study are population ageing and economic growth. The variables of population ageing and economic growth follow the literature review. All authors in the literature review assented increasing life expectancy and declining total fertility rates are the main causes of population ageing. Hence, the main variables of interest in this paper are: life expectancy, total fertility rates, and economic growth. For group 1, the data of 20 countries for the period from 1961 to 2015 are applied. In the literature, the underlying arguments as well as measurement methods for other variables such as consumption and saving patterns, labour productivity, human capital, and others, were highly varied. Thus, control variables are created for the contradicting variables. The data of consumption, saving, old age dependency ratio, and health expenditure were available from the World Bank and OECD databases for 16 countries. For group 2, time period from 1978 to 2013 are applied for all variables.
4.2.1. Dependent variable
The dependent variable in all regression models of selected countries is real GDP per capita
measured by the World Bank.The real GDP per capita is expressed in annual percentage based
on the constant local currency and is converted to the constant 2010 U.S dollars. The calculation of GDP per capita is performed as: a country’s gross domestic products divided by their
and taxes and subsidies which are not included in the gross value. Hence the dependent variable
represents a country’s economic output by its total population at time t. The current year’s
growth rate of annual percentage is calculated as a division of difference between the GDP per capita in time t and the GDP per capita in time t-1 by the GDP per capita in time t-1. Hence the dependent variable represents economic growth of a country i at time t. The dependent variable in this paper will be parameterized as: YGi,t
4.2.2. Independent variables
The independent variables of this paper: Life expectancy and total fertility rates, are
parameterised as LEi,t and TFRi,t respectively. All the data measurement rules follow the World
Bank databases.
LEi,t indicates the number of years a newborn’s expected lifetime on the basis of patterns
of mortality at the time of its birth. LEi,t is expressed in years and it indicates the average
expected lifetime of both the male and female. Thus the increases in life expectancy imply a downturn of mortality rates and contribute to the ageing population. The World Bank
measurement is based on the data source of the UN’s World population prospects and statistics report, statistical publications of national statistical offices, Eurostat, U.S Census bureau, and Secretariat of the Pacific Community.
TFRi,t denotes the number of children that would be born if the maternity was to live and
bore the children in a specific year. This covers all age groups of maternity which were
registered in a registration system. Therefore the TFRi,t indicates how many newborns were to be
born in a country i in a specific year t. The measurement puts its basis on the data sources: the
UN’s world population prospects, statistical publications of national statistical offices, Eurostat,
U.S Census bureau. The unit of TFRi,t is expressed in births per woman.
4.2.3. Control variables
The four control variables are created in accordance with the literature: old age dependency ratio,
private consumption, health expenditure and private saving, and are parameterised as: ODi,t,
ODi,t represents the ratio of people older than 64 to the working groups of ages between 15 and 64. The ratio is percentage of working age population. The meaning of the ratio is that the dependency burden that working age population bears and the ratio increases in either case: higher number of older people or less working age population. Thus, population ageing will overburden the accountability to the working age groups economically and it might affect patterns of economic decisions. The World Bank measurement is from the data source of the UN’s world population prospects.
LOGPCi,t indicates the amount of final consumption expenditure of households in a
country i in a specific year t. This contains overall consumptions of households such as food,
clothing, housing, energy, health costs, durable goods and services. Notably, the health costs within the private consumption variable is not overlapped with the health expenditure variable. In addition, the household spending includes the government transfers and non-profit institutions serving households (NPISHs) which directly benefit households. Hence this implies the actual individual consumption. The data are from the OECD database and the measurement is in million U.S dollars. From the obtained data, it is scaled with logarithm to control the unit size.
Therefore, LOGPCi,t implies the total consumption of households including government transfers
and NPISHs in logarithmic scale.
LOGHEi,t is measured in U.S dollars per capita and it is scaled with logarithm. The data
are obtained from the OECD dataset. LOGHEi,t represents the final consumption of healthcare
goods and services for both the personal and public. These healthcare is financed by the government spending on health, voluntary health insurance, private funds such as households’ out-of-pocket payments, private corporations and NGOs. Hence this expresses the amount of money spent on health excluding health care costs of private consumption in logarithmic scale.
SAVi,t refers to the difference between disposable income plus changes in households’ net
equity in pension funds and final consumption expenditure. The data obtained from the OECD dataset and the measure is in percentage of GDP. From the obtained data, the unit is manipulated with a division of million U.S dollars after the multiplication of the same GDP used in the consumption ratio to the ratio of private consumption to GDP to isolate the savings from the different GDP measurement between the World Bank and OECD. As the measurement is based
on the net value, it implies the possible negative values. Therefore a division of million U.S dollars is operated for the unit control scaling instead of logarithm scale to avoid any calculation
error. Thus SAVi,t reflects the net residual income of households invested in financial and
non-financial assets in million U.S dollars.
4.2.4. Dummy variables
The dummy variable of global financial crisis in 2008 is made to investigate possible temporal exogenous effects on the saving and economic growth. The motivation of this expectation is
from the visualization of obtained data. After visualization of the all variables, only YGi,t in and
SAVi,t exhibited kinked graphs thereabout 2008. To avoid the scaling mistakes as the two
variables contain the negative values, the growth rate of all variables was examined and the results were indifferent. The global financial crisis in 2008 is widely known as a severe
economic crisis negatively affected to economic growth and hence the financial crisis in 2008 is predicted as a possible exogenous effect. The aftermath of the crisis is set for 3 years in 2008, 2009 and 2010 following the figure 4.2.4. In this figure 4.2.4, sharp downturns started to recover from 2009 for selected 15 countries. The saving decreased to -0.2 million U.S dollars and
economic growth declined to -8% in maximum in 2009. The dummy variable is parameterised as
CR having value 1 if the years are 2008, 2009 and 2010 and having value 0 otherwise. The CR is
also applied to interaction term with SAVi,t as CRSAVi,t= CR*SAVi,t.
Figure 4.2.4 visualisation of group 2’s economic growth and saving
4.2.5. Balanced panel data
Panel data is a combination of time series and cross-sectional data (Baltagi, 2001). The time series data is a sequence of data points collected in time order (Baltagi, 2001). The cross
sectional data is a collection of data observing many subjects at the same point of time (Baltagi, 2001). It is called balanced panel when the dataset contains all observation values in all time frame for all cross sectional entities and the dataset used in this paper is the balanced panel data having no time gap in the time series.
4.3. Variable statistics
Before the empirical test, statistical descriptions of group 1 and group 2 are created to depict the properties of obtained data. Table 4.3.1 displays the summary statistics of group 1 and 2. As can be seen, there are no significant differences in mean and standard deviation concerning the dependent and independent variables between groups. Notably, negative minimum values of saving and economic growth predicted are in line with the arguments of dummy variable above. The possible interpretation of negative saving is that saving measure is the net saving and it is possible not to save and to spend the previous savings during the economic crisis to maintain the consumption and standard of living. The minimum and maximum values of economic growth were arbitrary and temporary for group 1: -14% growth of Chile in 1975 and 13.6% growth of Portugal in 1970. For group 2, -8.7% growth was Finland in 2009 and 11.58% growth was South Korea in 1983 and this possibly can be supported by financial crisis in 2008 and rapid economic growth of East Asia in the 1980s respectively.
Table 4.3.1Statistical descriptions of group 1 and group 2
Group 1: 1961- 2015
Variables Mean Std.dev. Minimum Maximum N
YG 2.5323 3.0330 -14.3181 13.6151 1,100 LE 75.4181 4.6238 53.7367 83.8437 1,100 TFR 2.0067 0.6636 1.076 5.929 1,100 Group 2: 1978- 2013 YG 1.9905 2.5670 -8.7070 11.5849 576 LE 77.3505 2.9162 64.9761 83.3320 576 TFR 1.7088 0.2625 1.076 2.9 576 OD 20.9465 4.9376 6.6122 39.5834 576 SAV 0.0537 0.9787 -0.3004 0.6221 576 LOGPC 12.0314 1.6662 7.0732 16.2185 576 LOGHE 7.3401 0.8025 3.8568 9.0614 576 CR 0.1111 0.3145 0 1 576 CRSAV 0.0035 0.0275 -0.3004 0.2480 576
Source: the table is constructed by author based on the complete dataset
Table 4.3.2 exhibits the correlation matrix of group 1 and 2. For group 1, a correlation between life expectancy and total fertility rates is high so they are considered to have a strong correlation (-0.7576). It is generally accepted as a strong correlation when the correlation between independent variables is higher than 0.7 and it might lead a multicollinearity problem (Baltagi, 2001). Nevertheless it is still considered as giving meaningful information in so far as the correlation does not exceed 0.8 or 0.85 (Baltagi, 2001). The strong correlation between independent and dependent variables are not deemed as a serious problem on the other hand (Baltagi, 2001). For group 1, there are only two independent variables, and consequently no action is taken. Figure 4.3.2 visualises the strong correlation between independent variables of group 1 with red boxes.
Table 4.3.2 Correlation matrix
Group 1 YG LE TFR YG 1.0000
LE 0.2525 1.000
TFR -0.3940 -0.7576 1.000
Group 2 YG LE TFR OD SAV LOGPC LOGHE CR CRSAV
YG 1.0000 LE -0.3467 1.0000 TFR -0.0612 -0.2151 1.0000 OD -0.3546 0.5150 -0.2437 1.0000 SAV 0.1221 0.0819 -0.0664 -0.1760 1.0000 LOGPC -0.0673 0.2219 -0.2793 0.0918 0.5914 1.0000 LOGHE -0.3774 0.8301 -0.0844 0.5181 0.1366 0.2977 1.0000 CR -0.3376 0.3208 0.0289 0.1582 -0.0769 0.1230 0.3196 1.0000 CRSAV -0.0400 0.2345 -0.1106 0.0304 0.3122 0.1310 0.1797 0.1684 1.0000
Source: the table is constructed by author based on the complete dataset
Figure 4.3.2 Correlation matrix graph: group 1
Source: the figure is constructed by author based on the complete dataset
For group 2 in the table 4.3.2, the only detected strong correlation between independent variables
is between LOGHEi,t and LEi,t with 0.8301. In this case, generally one of the variables is omitted
to avoid the multicollinearity problem. However, as health expenditure is one of the control variables which has a strong correlation with only life expectancy, it is still employed for group 2
by controlling the effects. Figure 4.3.2 displays the strong correlation between LOGHEi,t and LEi,t with red boxes.
Figure 4.3.2 Correlation matrix graph: group 2
Source: the figure is constructed by author based on the complete dataset
5. Methodology
To investigate the effects of population ageing on economic growth, this section delineates the
empirical models as well as the the important methodologies regarding the empirical test.First,
pretests issues with respect to the characteristics of the panel data are discussed. The results of the pretests proved that simply running panel regressions without the pretests leads to selection
model has the most statistical benefits. Third, the empirical models in this paper are defined pursuant to the results of the pretests and the Hausman test.
5.1. Pretests
The characteristic of panel data is a combination of time series and cross-sectional data as mentioned in section 4.2.5. Therefore, it is always argued that possible problems should be checked prior to simply running empirical tests using panel data by scholars and practitioners (Min & Choi, 2014). More importantly, the selection of empirical models highly depends on the pretest results (Min & Choi, 2014). Therefore, in this paper, the pretest regarding the panel data analysis is conducted before the selection of regression models. The main issues of the pretests in panel data analysis are: cross sectional dependence, stationary, and autocorrelation (Min & Choi, 2012). The following subsections intensively examine the each issue.
5.1.1. Cross sectional independence
One of the problematic assumptions in the panel data analysis is the cross sectional independence (Quah, 1994). This assumes all cross-sectional entities are independent each other which is rejected in many cases (Costantini, 2010). Furthermore, this assumption is applied to a number of pretests and it becomes a more serious problem in the unit root test (Costantini, 2010). Under this assumption of independence, the central limit theorem is applied to obtain the asymptotic normality of panel statistics and the values of the regression tests are biased when the assumption is rejected (Costantini, 2010). Hence, the value of the regression tests will not be meaningful. According to Baltagi (2001), the problem of cross sectional dependence is present in macro panels with more than 20 years of time period. Accordingly, it is necessary to test in this paper as the time series of both groups 1 and 2 is T=55 and T=36 respectively.
There are two widely known tests for the cross sectional independence of panel data: Breusch-Pagan Lagrange Multiplier (BPLM) and Pesaran’s cross dependence test
(Torres-Reyna, 2007). Both tests are analogous in employing the null hypothesis that residuals across entities are not correlated each other, and are different in displaying the results
exhibits each correlation of residuals between countries (Torres-Reyna, 2007). In this paper, the BPLM test is used to exhibit the cross dependence between countries. The null and alternative hypotheses for the BPLM model are as follow:
H0: Cov(ei,t, ej,t) = 0
H1: Cov(ei,t, ej,t) ≠ 0 (i ≠ j, where 𝜀i,t and 𝜀j,t stand for residuals of the country i and j). The table 5.1.1 exhibits the results of the BPLM test for both groups. The null hypothesis
of no correlation of residuals across countries is rejected for both groups (p<0.01) and hence the
alternative hypothesis indicates the existence of cross sectional dependence within the data. The
variables employed in the BPLM test for both groups are: YG, LE, and TFR for group 1; YG, LE,
TFR, OD, LOGHE, LOGPS, and SAV for group 2. In this table 5.1.1, the strong residual
correlations (for instance, higher than 0.7 with bold letters) between countries are most likely present in where the borders are close to each other. In group 1, the correlations between Austria, Belgium and France; USA and Canada; and Sweden and Iceland are high and the borders are close to each other. In group 2, Belgium and the Netherlands; Finland and Sweden; and Portugal and Spain are presented as having strong correlations. The table provides the intuitive insight in the real world that the countries close to each other are more likely to be correlated by the common monetary and fiscal policies, culture, trade and others.The comparison of the two groups imply changes in the correlation with time evolution. The result of BPLM test of presence of cross sectional dependence thus will be considered for all the following steps of the model selection.
Table 5.1.1 Breusch-Pagan Lagrange Multiplier (BPLM) test for group 1 and 2
GROUP 1
AUS AUT BEL CAN CHL DNK FIN FRA GRC ISL ITA JAP NLD NOR PRT KOR ESP SWE GBR USA AUS 1.00 AUT 0.28 1.00 BEL 0.33 0.65 1.00 CAN 0.33 0.33 0.49 1.00 CHL 0.29 0.18 0.17 0.23 1.00 DNK 0.16 0.51 0.54 0.47 -0.03 1.00 FIN 0.41 0.54 0.56 0.52 0.02 0.43 1.00 FRA 0.27 0.75 0.84 0.51 0.20 0.55 0.55 1.00 GRC 0.13 0.32 0.38 0.23 -0.16 0.36 0.35 0.35 1.00 ISL 0.15 0.29 0.30 0.41 0.15 0.22 0.32 0.37 0.22 1.00 ITA 0.05 0.54 0.69 0.39 0.15 0.45 0.48 0.68 0.31 0.12 1.00 JAP -0.01 0.26 0.31 0.26 -0.01 0.46 0.20 0.37 0.33 -0.12 0.44 1.00 NLD 0.35 0.55 0.68 0.56 0.08 0.56 0.41 0.71 0.37 0.30 0.48 0.31 1.00 NOR -0.03 0.17 0.19 0.28 0.02 0.49 0.22 0.04 0.36 0.28 0.27 0.18 0.25 1.00 PRT 0.18 0.53 0.62 0.19 0.12 0.28 0.29 0.66 0.39 0.23 0.58 0.32 0.56 -0.05 1.00 KOR 0.00 0.21 0.12 0.09 0.04 0.13 0.00 0.14 0.01 -0.11 0.19 0.17 0.10 -0.02 0.13 1.00 ESP -0.01 0.49 0.64 0.36 -0.10 0.38 0.44 0.64 0.52 0.23 0.65 0.42 0.51 0.16 0.51 -0.01 1.00 SWE 0.43 0.45 0.61 0.57 0.01 0.53 0.75 0.57 0.27 0.25 0.49 0.21 0.52 0.18 0.25 0.03 0.48 1.00 GBR 0.38 0.41 0.40 0.53 0.16 0.47 0.51 0.49 0.34 0.29 0.31 0.25 0.41 0.07 0.48 0.37 0.25 0.51 1.00 USA 0.21 0.25 0.36 0.81 0.32 0.52 0.33 0.42 0.27 0.33 0.32 0.34 0.54 0.30 0.26 0.19 0.28 0.42 0.65 1.00 Chi2 (190) = 1544.273***
(*=significant at 0.10 level, **=significant at 0.05 level, ***=significant at 0.01 level)
GROUP 2
AUS AUT BEL CAN DNK FIN ISL JAP NLD NOR PRT KOR ESP SWE GBR USA AUS 1.00 AUT 0.32 1.00 BEL 0.43 0.70 1.00 CAN 0.43 0.22 0.38 1.00 DNK 0.20 0.51 0.38 0.40 1.00 FIN 0.52 0.64 0.63 0.62 0.52 1.00 ISL 0.42 0.26 0.37 0.41 0.33 0.58 1.00 JAP 0.02 0.08 0.16 0.25 0.23 0.22 0.13 1.00 NLD 0.33 0.72 0.75 0.47 0.50 0.58 0.29 0.13 1.00 NOR 0.10 0.18 0.19 0.38 0.54 0.27 0.38 -0.10 0.30 1.00 PRT 0.19 0.53 0.56 0.07 0.05 0.47 0.26 0.19 0.45 -0.10 1.00 KOR -0.23 -0.05 -0.07 0.13 0.25 -0.06 -0.18 0.42 0.02 -0.02 0.06 1.00 ESP 0.15 0.45 0.58 0.32 0.26 0.61 0.39 0.14 0.58 0.12 0.76 0.12 1.00 SWE 0.36 0.60 0.67 0.66 0.69 0.82 0.47 0.40 0.63 0.26 0.39 0.12 0.57 1.00 GBR 0.26 0.34 0.33 0.62 0.52 0.60 0.33 0.40 0.43 0.10 0.17 0.28 0.36 0.62 1.00 USA 0.15 0.05 0.23 0.74 0.37 0.29 0.09 0.25 0.47 0.38 -0.09 0.28 0.17 0.46 0.54 1.00 Chi2 (120) = 703.254***
(*=significant at 0.10 level, **=significant at 0.05 level, ***=significant at 0.01 level)
5.1.2. Stationarity
The stationary process is a stochastic process that of joint probability distribution does not change over time (Min & Choi, 2014). As a consequence, its mean and variance also do not change along with time evolution (Min & Choi, 2014). The stationarity is the crucial
consideration when the panel data analysis follows autoregressive models which expect the lag of a variable consistently affect the result of the future value (Min & Choi, 2014). It is led to spurious regressions if non-stationary process is directly used for the autoregression tests (Min & Choi, 2014). The autoregressive model with lag 1: AR(1) looks like:
𝑌𝑖,𝑡= 𝛼𝑖𝑌𝑖,𝑡−1+𝛽𝑖𝑋𝑖,𝑡+𝜀𝑖,𝑡
It is generally tested to check the stationarity by employing the unit root test. In case the variable contains the unit root: I(1), the cointegration relationship should be further examined as this facilitates a linear transformation to a stationary process (Min & Choi, 2014).
The interest in this paper is not the autoregressive model. Thus the discussion of
advanced methodology over the autoregressive model finishes here. Nonetheless, it is important to mention it here that there exist two generations of the unit root tests for the future research regarding both subsections 5.1.1 and 5.1.2, as the cross sectional dependence is normally disregarded. The first generation assumes cross sectional independence across panel entities. Hence aforementioned discussion in subsection 5.1.1 that the assumption of cross panel independence in the pretest becomes a serious problem in investigating the stationarity of the panel time series for the first generation. More importantly, most of the unit root tests
recommended in the literature are the first generation such as Levin-Lin-Chu (LLC),
Im-Pesaran-Shin (IPS), and Phillips and Perron (PP) (Min & Choi, 2012). The second generation relaxes the assumption of cross sectional independence within the unit root test. For the unit root test with cross sectional dependence, Pesaran (2007) suggested Cross-sectional Augmented Dickey-Fuller (CADF) test with panel data. The second generation models will significantly reduce the cost of choosing incorrect models.
5.1.3. Autocorrelation
Similar to the discussion of stationarity, autocorrelation of residuals is also one of the most salient considerations in the pretest. The differentia with stationarity is an application of a linear regression model focusing on residuals (Min & Choi, 2012). The autocorrelation is applied to fixed and random effects models which are the main focus of this paper. To obtain unbiased and consistent estimates from the linear panel regression, absence of autocorrelation of residuals within panel groups is generally assumed (Min & Choi, 2012). However, as the panel data contains the characteristic of time series, there exists a possibility of having autocorrelation (Min
& Choi, 2012). It is written when the residual 𝜀i,t has first-order autocorrelation, AR(1):
ei,t= 𝜌ei,t-1 + 𝜈i,t
The wooldridge test is generally conducted for the investigation of autocorrelation in panel data, and it is employed in this paper. The null hypothesis of the Wooldridge test is no first-order autocorrelation in the panel data and the alternative hypothesis is the presence of the autocorrelation (Wooldridge, 1960). The test is based on the F-test and the variables used for the test are the same as subsection 5.1.1 for both groups. The table 5.1.3 exhibits the results of the
Wooldridge test and null hypotheses are rejected for both groups (p<0.01).
Table 5.1.3 Wooldridge test for group 1 and 2
F-statistic p value
Group 1 37.623 0.0000***
Group 2 25.936 0.0000***
* p<0.05, ** p<0.01, *** p<0.001
Source: the table is constructed by author based on the complete dataset
Hence, it is concluded that both cross-sectional dependence and AR(1) exist in the panel data
from the pretests. For the fix effect model with AR(1), 𝜌 is estimated by Cochrane-Orcutt
random effect model, the generalised least squares (GLS) estimates is applied as suggested by Baltagi and Wu (1999).
5.1.4. Hausman test
Before the selection of the empirical models, Hausman test is performed. As mentioned in subsection 5.1.3, this paper focuses on linear panel regressions to examine more generalized overall effects of population ageing in the long run. The important issue in the selection of the linear panel regression model is an assumption of the error term. Fixed and random effects models which are taken in empirical models of this paper, provide different assumptions of the error term within the model (Min & Choi, 2012). The panel regression model follows as:
𝑌𝑖,𝑡= 𝛼𝑖 + 𝛽𝑖𝑋𝑖,𝑡 + 𝜈i,t + 𝜀𝑖,𝑡, i=1,2,...,n and t=1,2,...,T (1)
Except for 𝜈i,t , other variables are identical to the normal linear regression model with respect to
panel data setting. 𝜀𝑖,𝑡 here refers to the error term that changes according to i and t. However, 𝜈i,t
is assumed differently in the fixed and random effects models (Min & Choi, 2012). The fixed
effect model expects 𝜈i,t does not change over time denoting individual panel’s characteristic is
fixed at each individual (Min & Choi, 2012). Thus, in the fixed effect model, 𝜈i,t is not a random
variable but it is a parameter to estimate and it considers each country’s overall difference (Min & Choi, 2012). The estimation idea of the fixed effects models follows two steps of
transformations from the equation (1). Firstly, it supposes each variable consists of the average and it is called ‘between’ transformation as (2):
𝑖,𝑡= 𝛼𝑖 + 𝛽𝑖 𝑖,𝑡 + 𝜈i,t + 𝑖,𝑡 (2)
Y X ε
Deduction of (2) from (1) eliminates the term 𝜈i,t , and it is called ‘within’ transformation as (3):
Therefore it has an advantage of estimating consistent estimators using OLS method when
Cov(𝑋𝑖,𝑡, 𝜈i,t ) ≠ 0 (Min & Choi, 2012).
On the other hand, the random effects models assume the individual characteristics are not fixed for each individual and thus the intercepts are independent from explanatory variables (Min & Choi, 2012). Furthermore, it is known that the random effects models are more
statistically meaningful if the condition of Cov(𝑋𝑖,𝑡, 𝜈i,t ) = 0 is satisfied (Min & Choi, 2012).
Hence, if the condition Cov(𝑋𝑖,𝑡, 𝜈i,t ) = 0 is met, the random effects models are preferred.
However, there exists systematic difference in the random effects models as the estimates are
inconsistent in the case of Cov(𝑋𝑖,𝑡, 𝜈i,t ) ≠ 0 (Min & Choi, 2012). In other words, the selection of
the model between the fixed and random effects is contingent on the assumption of Cov(𝑋𝑖,𝑡, 𝜈i,t )
= 0.
There exists a test model so called ‘Hausman test’ that examines the argument above between fixed and random effects models. This estimates the preference between the fixed and random effects models (Min & Choi, 2012). The null and alternative hypotheses are following:
H0: Cov(𝑋𝑖,𝑡, 𝜈i,t ) = 0
H1: Cov(𝑋𝑖,𝑡, 𝜈i,t ) ≠ 0
Hausman test was conducted for both groups, and the results in Table 5.1.4 exhibit the fixed effects models are preferred over the random effects models for group 2 and the random effects models were preferred for group 1. Hence both models are defined in the following subsection.
Table 5.1.4 Hausman test
Chi2 p value
Group 1 1.19 0.5520
Group 2 37.50 0.0000***
* p<0.05, ** p<0.01, *** p<0.001
5.2. Empirical models
In accordance with the pretests, the three empirical models are defined: fixed effects models with AR(1), random effects models with AR(1) and feasible generalized least squares (FGLS) models with AR(1) and cross sectional dependence. Unfortunately, the method of controlling
autoregression and cross sectional dependence at the same time using the fixed and random effects models could not be found, hence the only method of controlling autocorrelation is applied in the fixed and random effects models. In addition, controlling autocorrelation with the fixed and random effects models have no option for robustness check. The reason for including the third model is that it facilitates controlling the cross sectional dependence, heteroscedasticity and autocorrelation. The FGLS model is based on the normal OLS regression with deviations regarding autocorrelation and heteroscedasticity treating all selected countries as one entity with pooled data. This method is limited in the consideration of each country’s characteristic but it controls autocorrelation, heteroscedasticity and cross dependence. Therefore, the results of three models will be compared. The models for both groups before the transformation are as:
Group 1: YG𝑖,𝑡= 𝛼𝑖+ 𝛽𝑖TFR𝑖,𝑡+ 𝛽𝑖LE𝑖,𝑡+𝜈i,t + 𝜀𝑖,𝑡
Group 2: YG𝑖,𝑡= 𝛼𝑖+ 𝛽𝑖TFR 𝑖,𝑡+ 𝛽𝑖LE𝑖,𝑡+ 𝛽𝑖OD𝑖,𝑡+ 𝛽𝑖LOGPC𝑖,𝑡+𝛽𝑖LOGHE𝑖,𝑡+
𝛽𝑖SAV𝑖,𝑡+𝛽𝑖CR𝑖,𝑡+𝛽𝑖CRSAV𝑖,𝑡+𝜈i,t + 𝜀𝑖,𝑡
6. Empirical results and interpretation
This section interprets the empirical results built on the all arguments above. Table 6 displays the empirical results of three panel regression methods for group 1 and 2. FE with AR(1) and RE with AR(1) refer to fixed and random effect models with the control of autocorrelation respectively. As discussed in subsection 5.1.3, first order autocorrelation is controlled by Cochrane-Orcutt estimation in the fixed effect model and Baltagi-Wu model of GLS in the random effect model. FGLS with AR(1) & CD & rob stands for feasible generalised least square model with the control of autocorrelation, cross-sectional dependence and robustness. For both groups, crisis-dummy was estimated followed by the estimations of independent variables of life expectancy and total fertility rate. Additionally, measurements with control variables were
conducted for group 2. The interpretation first looks at 15 regressions to check any significant variation in the result and focuses on the result of Hausman test: random effect of group 1 and fixed effect of group 2.
Table 6 Panel data analysis with autocorrelation and cross sectional dependence Dependent variable: YG𝑖,𝑡
Variable Group 1 Group 2
FE with AR(1) RE with AR(1) FGLS with AR(1)
& CD & rob
FE with AR(1) RE with AR(1) FGLS with AR(1) & CD & rob
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) LEi,t -0.303*** -0.226*** -0.327*** -0.269*** -0.253*** -0.204*** -0.361*** -0.265*** -0.354 -0.334*** -0.260*** -0.0539 -0.225*** -0.163*** 0.0178 (0.0471) (0.0468) (0.0411) (0.0409) (0.0284) (0.0282) (0.0592) (0.0588) (0.204) (0.0502) (0.0499) (0.0825) (0.0380) (0.0325) (0.0362) TFRi,t -0.472 -0.144 -0.678* -0.461 -0.202 -0.0196 -3.022*** -2.273** -2.565*** -1.912** -1.559** -1.505** -1.104*** -0.767** -1.013** (0.339) (0.333) (0.285) (0.280) (0.181) (0.181) (0.784) (0.742) (0.769) (0.590) (0.567) (0.524) (0.316) (0.290) (0.324) CRi,t -3.915*** -3.855*** -2.837*** -1.852*** -1.577*** -1.891*** -1.583*** -1.505*** -1.257** (0.456) (0.457) (0.524) (0.404) (0.415) (0.395) (0.408) (0.415) (0.394) SAVi,t 3.619 4.559** 3.037*** (2.331) (1.738) (0.712) CRSAVi,t -0.0849 0.155 -1.412 (3.333) (3.061) (1.262) ODi,t -0.0663 -0.105** -0.0837*** (0.0653) (0.0326) (0.0194) LOGPCi,t 5.994*** -0.160 -0.0924 (1.546) (0.104) (0.0714) LOGHEi,t -4.462*** -0.558 -0.695*** (1.325) (0.306) (0.163) cons 26.31*** 20.00*** 28.56*** 23.89*** 21.79*** 17.83*** 35.10*** 26.50*** -4.336 31.09*** 24.89*** 16.83** 21.09*** 15.82*** 10.27*** (2.671) (2.625) (3.567) (3.543) (2.330) (2.296) (3.607) (3.814) (9.391) (4.325) (4.305) (5.263) (3.218) (2.751) (2.683) Rho 0.3497 0.3555 0.3497 0.3555 0.3115 0.2672 0.2844 0.3115 0.2672 0.2844 F-statistic 37.82 51.04 20.07 22.59 11.42 Wald-statistic 98.23 173.52 83.51 98.42 45.54 75.20 115.33 36.08 47.18 72.21 R2(Overall) 0.1561 0.2044 0.1594 0.2056 0.1265 0.1801 0.0183 0.1371 0.1878 0.2556
* p<0.05, ** p<0.01, *** p<0.001, standard errors in parentheses
Source: the table is constructed by author based on the complete dataset
The empirical results indicate that increases in life expectancy have negative effects
(P<0.01) on economic growth except for (9), (12) and (15). The effects of life expectancy on
effects of other variables are fixed. Focusing on the result of Hausman test, the effects of the group 1 is -0.327 percent and of group 2 is -0.361 percent without crisis, and -0.269 percent for group 1 and -0.265 percent for group 2 when the crisis is present at a significance level of 0.01. This implies that one year increase in life expectancy is expected to have negative effects of -0.344 (average of group 1 and 2) percent on economic growth without the crisis and it is
expected to have approximately negative effects of -0.267 (average of group 1 and 2) percent on economic growth with the crisis.
Yet, it is difficult to conclude that the estimates follow the pessimistic view. The equations (9), (12) and (15) show that the negative effects of life expectancy on economic growth are insignificant when the control variables are added. One of the possible interpretations for control variables is the consumption. In general, private consumption is a large part of the
GDP. Thus, private consumption possibly predisposed the effects of life expectancy to be
ambiguous with other control and possible omitted variables which are not covered in the literature. Furthermore, private consumption in (9) indicates 1 percentage change in private consumption is predicted to positively affect economic growth by 5.994 percent at a significance level of 0.01. However, there exists a possibility that other factors influenced the effects of independent variables which are omitted in this paper. Hence, it is concluded that the hypotheses 1 of group 1 and 2 are rejected and the hypothesis 3 of group 2 is ambiguous based on the empirical results.
The effects of total fertility rates of group 1 exhibit only one significant negative effect of
-0.678 percent on economic growth in (3) at a significance level of 0.10. In group 2, all the effects of the total fertility rates are predicted to have negative effects at a significance level of either 0.01 or 0.05. Focusing on the result of Hausman test, the regressions (7), (8), and (9) imply the effects of total fertility rates are -3.022 percent, -2.273 percent, and -2.565 percent
respectively. Notice here the reverse interpretation of total fertility rates that an increase in total fertility rates of 1 birth per woman have negative effects on economic growth of approximately -2.62 percent (average of (7), (8) and (9)). In other words, higher total fertility rates have negative effects on economic growth. Hence the interpretation should be opposite as lower fertility rates have positive effects on economic growth of 2.62 percentage. The effects of total
fertility rates were lessened with control variables as (9), (12) and (15) display but the variation is not significant. Hence it is concluded that the hypothesis 2 of group 1 and the hypotheses 2 and 4 of group 2 are not rejected based on the empirical results.
In conclusion, the empirical results support the pessimistic view regarding life
expectancy without control variables, whereas the optimistic view is supported by the empirical
results concerning total fertility rates for both with and without control variables. In the
hypotheses, control variables were expected to have effects on the relationship between the independent and dependent variables. The evidence indicates that the pessimistic view of life expectancy is ambiguous with other factors discussed in the literature such as consumptions, savings, expenditures and others. Nevertheless, in isolation, it is concluded that the pessimistic view of life expectancy on economic growth is supported theoretically and empirically. On the other hand, total fertility rates of group 1 were not supported by the empirical results whereas the optimistic viewpoint was supported by empirical results for group 2. The main difference
between the two groups in terms of data is the time period. 17 years of data were omitted in the second group and the decrease in total fertility rate during this period is more severe than other periods as figure 2.1 depicts. As can be seen in this figure 2.1, the marginal decreasing speed of the total fertility rates has been being mitigated since the 1980s. Hence, it seems the data structure of group 2 is closer to the present. Additionally, data obtained from the World Bank and OECD implied savings, consumptions, public expenditures, old age dependency have increased since the 1980s. According to the pessimistic view, consumption and saving patterns should have declined with the effects of population ageing. In this regard, theoretical
assumptions of the optimistic viewpoint concerning saving and consumption patterns are supported by the empirical data, whereas the theoretical assumptions of the pessimistic view of public expenditures and old age dependency ratio are supported by the empirical data. Hence, for the policy implication, the life expectancy should be considered with other factors based on economic reasoning and the total fertility rates should be considered with more realistic data structure. Furthermore, the orthodoxy of the pessimistic view of effects of population ageing on economic growth is worth to be reconsidered with the optimistic viewpoint as the empirical evidence indicated.
7. Limitations and discussion
This paper contains limitations. Firstly, data availability of control variables is weak.
Notwithstanding the empirical results support the optimistic view regarding total fertility rates,
the data of core variables within the optimistic view such as labor productivity and human capital were unavailable for the long run although it is significantly expected to affect the empirical results. Secondly, the control of cross sectional dependence was unavailable using fixed and
random effects models.A better solution could not be found by the time of finalising this paper.
This should be concerned in the future study. Thirdly, as this paper aims at generalisation of conflicting arguments, the empirical tests put their basis on the previous literature. However, I expect the reverse causality between variables. This should be further analysed in the future study. Furthermore, the time dynamic is simplified in this paper but advanced econometric models such as dynamic panel model or panel vector autoregression generalised by Abrigo and Love (2015) should be considered due to the expectation of the reverse causality and the time dynamic. Specifically, Christopher Sims (1980) argues ‘atheoretical’ methodology in
macroeconomics with vector autoregression (VAR) model that dynamics of macroeconomics are difficult to control with theories and statistical information should be maximised. This is in line with the assertion of Paul Romer (2016) who is one of the developers of new endogenous growth theory of the optimistic view that intrinsic problems in modern macroeconomics theories are usually a failure in explaining the real world by being dependent on authoritative theories. As the change in trend of total fertility rates since the 1980s exemplifies, dependence on an authoritative theory without criticism might lead to ‘a priori’ assumption from the beginning. A combination of critical theories and advanced methods will provide the potential to avoid this problem. Finally, most of the literature studied on developed countries and therefore the study on
developing countries for the policy implication regarding the difference with developed countries should be considered.
8. Conclusion
This paper investigated the research question “To what extent population ageing in most developed OECD countries has affected on economic growth during the period from 1961 to 2015?” so as to generalise conflicting opinions and to infer a policy implication. Population ageing has been accelerating in most developed OECD countries for a recent few decades and the number of studies have been done with respect to these countries. Howbeit, most studies were theoretical and were divided into positive and negative effects. Hence, this paper focalised on the empirical research of the conflicting arguments. To analyse the research question, the literature was reviewed and clarified the mechanism of the arguments. Pursuant to the literature, hypotheses were constructed. In addition, the data of 20 developed OECD members countries were collected and were divided into two groups to compare the effects of control variables for the empirical analysis. However, as pretests are crucial issues in the panel data, pretests were conducted before the empirical analysis and cross sectional dependence and autocorrelation were detected. To control the autocorrelation and cross sectional dependence, three types of panel data regression models were employed and the empirical results indicated life expectancy is
supported by the pessimistic view whereas total fertility rates were supported by the optimistic view. Therefore, the orthodoxy of the pessimistic view of population ageing on economic growth is not empirically verified. For the policy implication, the combination of the two opposite views are recommended.
