Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Intergenerational transfers & human capital investments in children in the era of aging
(USC Thesis Other)
Intergenerational transfers & human capital investments in children in the era of aging
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Intergenerational Transfers & Human Capital Investments in Children in
the Era of Aging.
by
Kanika Aggarwal
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulllment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Economics)
August 2020
Copyright 2020 Kanika Aggarwal
Change is the only constant. To these unusual times, and us.
ii
Acknowledgements
I am very grateful to all of my professors, who have taught me, advised me and helped me
throughout my PhD journey. I would especially like to thank Professor Ayse Imrohoroglu,
my advisor, who has helped me in every possible way from start to nish, and has not only
taught me the ways of good and well-informed research but has also taught me the value of
perseverance while doing the research. I would like to thank Professor Caroline Betts, who
has been very supportive and encouraging throughout, and has always been willing to listen
and oer help and advise. I would like to thank Professor Jerey Nugent, who has shown
nothing but kindness and encouragement for every single endeavour of mine. I have learnt
a lot from him and am thankful for his classes where the seed of the work presented in this
thesis was rst planted. I would also like to thank Professor Neha Bairoliya, who has been
a great support in these past few months and her insights and suggestions has helped me
immensely in my work.
I am also grateful to all the sta members of the department of economics who have
always been supportive and friendly. Thank you, Alex and Young, for handling all my queries
and for your patience with me. I would also like to thank all of my friends, who have listened
to me patiently for hours, have helped me in the times of need, and have essentially become
my second family here.
I would like to thank each and every person in my family, my grandfather who passed
away recently, and who had been proud of all my accomplishments, big or small. I would
like to thank my parents, Mahesh and Anjna, without whom none of this would be possible,
and my brother Tushar who always has my back. I would also like to thank my boyfriend
iii
Anupam who has put up with me with innite patience, wisdom and love during all this
time. Last but denitely not the least, I would like to give a big thanks to my cat Kali, she
is the only one who endured sleepless nights with me, curled up in my lap, while I worked.
iv
Table of Contents
Dedication ii
Acknowledgements iii
List Of Tables vii
List Of Figures ix
Abstract xi
Chapter 1: Introduction 1
Chapter 2: Intergenerational education gap and old-age support 4
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 Empirical specication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Chapter 3: Does population aging matter for human capital investments in
children ? 27
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Quantitative Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
v
3.4.1 Parametrization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4.3 Model Variants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4.4 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.5 Conclusion & Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Chapter 4: A cross-country analysis of population aging and educational ex-
penditure on children 62
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Data description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3.1 Trends in aging and educational expenditure . . . . . . . . . . . . . . 69
4.4 Empirical Specication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.5.1 OECD Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.5.2 OECDplus Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Chapter 5: Conclusion 86
Bibliography 89
Appendix A
Appendix to chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Appendix B
Appendix to chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Appendix C
Appendix to chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
vi
List Of Tables
2.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Distribution of population by education level (% of the sample population) . 13
2.3 Average transfers by education gap . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 Average transfers by child's age & education gap . . . . . . . . . . . . . . . 14
2.5 Results with child's level of income and education . . . . . . . . . . . . . . . 18
2.6 Results with income gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.7 Results with education gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.8 Results with education gap by regions . . . . . . . . . . . . . . . . . . . . . . 22
2.9 Results with education gap: wave 2013 . . . . . . . . . . . . . . . . . . . . . 23
2.10 Results with education gap: heckman estimation . . . . . . . . . . . . . . . . 25
3.1 Parameter values for the basic model . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Comparing household investment and saving rate across the three demo-
graphic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3 Comparisons across models with and without the social security sector . . . 54
3.4 Comparison of the aggregate investment rate across models with endogenous
transfer share . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.5 Comparisons across models with dierent values of . . . . . . . . . . . . . 58
4.1 Description of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
vii
4.2 Descriptive Statistics: 1998-2018 . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.3 Aging trends across countries for the years 2000 and 2015 . . . . . . . . . . . 70
4.4 Trends in educational expenditures and age-demography for selected countries 72
4.5 Results with Old-age dependency ratio . . . . . . . . . . . . . . . . . . . . . 78
4.6 Results with Median age . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.7 Results with Age eects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.8 Results with Old-age dependency ratio for the extended sample . . . . . . . 82
4.9 Results with life expectancy at age 60 for the extended sample . . . . . . . . 83
A.1 Distribution of population by education categories . . . . . . . . . . . . . . 97
A.2 Distribution of population by income categories . . . . . . . . . . . . . . . . 98
C.1 Descriptive statistics: indicators of aging . . . . . . . . . . . . . . . . . . . . 103
C.2 Correlation between household education expenditure and age composition . 104
C.3 Results with Median age for the extended sample . . . . . . . . . . . . . . . 104
C.4 Results with Age eects for the extended sample . . . . . . . . . . . . . . . . 105
viii
List Of Figures
2.1 Average transfers by education level . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Average transfers by child's age . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.1 Demographic trends across the world . . . . . . . . . . . . . . . . . . . . . . 30
3.2 Timeline of an agent's life . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Demographic trends for the baseline model . . . . . . . . . . . . . . . . . . . 38
3.4 Household Investment and Saving rate in the baseline model . . . . . . . . . 42
3.5 Demographic trends for the constant survival model . . . . . . . . . . . . . . 44
3.6 Household investment and saving rate in the baseline and constant survival
model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.7 Total investment rate in the baseline and constant survival model . . . . . . 47
3.8 Old-age dependency ratio for the constant survival plus constant n model . . 48
3.9 Household investment and saving rate in the three demographic models . . . 49
3.10 Household investment and saving rate with social security system . . . . . . 52
3.11 Household investment rate with endogenous transfers . . . . . . . . . . . . . 55
3.12 Aggregate investment and savings rate in the baseline model with dierent
values of . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.1 Household expenditure on education and old-age dependency ratio . . . . . . 73
4.2 Household expenditure on education and median age . . . . . . . . . . . . . 74
ix
4.3 Household expenditure on education and life expectancy at age 60 . . . . . . 75
B.1 Age-distribution of population . . . . . . . . . . . . . . . . . . . . . . . . . . 99
B.2 Household investment and saving rate in the three demographic models with
social security sector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
B.3 Household investment and saving rate in the three demographic models with
= 0:8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
B.4 Household investment and saving rate in the three demographic models with
= 0:7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
x
Abstract
The dissertation studies the nature of intergenerational transfers in a family and how it is
aected by changing age-demography. Chapter 2 conducts a study on determinants of old-age
support to parents and makes use of the China Health and Retirement Longitudinal Study
to document how the dierence in the educational attainment of the child and the parents
is associated with the size of the transfers received in the old-age from the child. Estimation
of the relationship using pseudo-poisson maximum likelihood technique reveals that the
transfers are positively and signicantly associated with the education gap, after having
accounted for other family, demographic and geographic eects. Chapter 3 studies the impact
of changing age-structure or population aging on the investments made by parents on building
their child's human capital. This question is explored using an overlapping-generations
framework with human capital accumulation in the childhood and intergenerational transfers
between children and parents, in the context of uncertain lifetimes and declining mortality.
Simulation results, based on the Chinese experience of demographic transition between 1871-
2009, show that population aging or decline in mortality has had a detrimental impact on
human capital investments. Results also suggest that household saving rates increase as a
result of aging. Chapter 4 analyzes the relationship between aging and household educational
expenditure for a panel of countries to verify the results from the model simulations of
Chapter 3. It exploits the cross-country variations in the demographic-make up of dierent
countries at dierent times. It nds systematic dierences between older and younger nations
concerning the educational expenditure by households. Specically, per-child educational
expenditures are lower for countries with a greater proportion of older people, or higher
xi
old-age dependency rate. These results conform to the ndings in Chapter 3 of the causal
relationship between aging and human capital investments in children.
xii
Chapter 1
Introduction
The central problem of economics is the allocation of scarce resources to competing uses. It
is not a simple problem to solve; uses or the ends are innite while the means are limited.
Almost all research in economics involves dealing with this problem one way or the other.
This dissertation on intergenerational transfers, human capital investments, and aging, is no
exception. The central theme of this compilation is the problem of intergenerational resource
allocation and the implications of evolving age-demography for these allocations.
Inquiry into the economic outcomes of population aging is a highly relevant pursuit
in the present-day world. The old-age population is steadily increasing as mortality and
fertility rates are declining world over. More people will be over 60 years of age than the
number of children under age ten by 2030. Moreover, the pace of aging has gone up in recent
times. More advanced nations (several European countries) experienced the demographic
transition to low fertility rates and mortality over a long period; however, several emerging
market economies are making the transition more quickly, leading to a substantial increase
in the old-age population. Estimates suggest that almost 22 % of the world's population
will be over 60 years of age by 2050
1
. As the demographic structure of society evolves, the
nature of intergenerational transfers also potentially change.
1. Source: UN World Population Prospects 2019
1
Intergenerational transfers can take various forms, such as bequests from parents to chil-
dren (accidental and otherwise), resources spent on children's upbringing, expenditure on
education and skill development, and transfers made by children to old-age parents both
of monetary and non-monetary nature. An aging society is more likely to focus on old-
age pension, old-age support, and health care for the elderly compared to a younger one.
Therein lies the intergenerational con
ict in the allocation of resources for an aging society.
The dissertation consists of three inter-related essays on this theme. These essays have the
household sector at the center of their analysis of intergenerational dynamics, and they study
how the households respond to the changing demographic and economic environment, and
in turn, shape it.
The structure of the dissertation is as follows. The second chapter conducts an empir-
ical analysis of the determinants and correlates of transfers from children to old-age parents.
It adds to the existing pool of research on inter-vivos transfers by elucidating the importance
of intergenerational dierences in educational attainment between parents and children, as
a factor in determining these transfers. The paper evaluates if the intergenerational dier-
ences in education matter for transfers using a recent dataset on Chinese households, which
contains detailed information on inter-vivos transfers and other family characteristics.
The third chapter undertakes an in-depth analysis of the impact of the aging popula-
tion on human capital investments in children with the help of a theoretical framework. This
paper focuses on intergenerational transfers from parents to children in the form of parental
investments in their children. It seeks to answer, how exactly does the demographic change
(aging of the population) cause changes in the human capital investments in children? The
analysis makes use of an overlapping generation framework with two-way altruism and hu-
man capital accumulation in children to study how saving and investment decisions change
with a decline in mortality over time. The quantitative simulations of the model using
2
the Chinese population data nd that the aging population has led to lower investments in
children's human capital.
Finally, the fourth chapter explores the cross-country evidence on the relationship
between aging and human capital investments suggested in the model analysis in Chapter
3. Data on education spending (which serves as a proxy for human capital investments) and
population statistics of the OECD countries is used in the primary analysis to nd if there is
a systematic link between these expenditures and the population's age structure. It exploits
variations in the age-demography of dierent countries over space and time. It nds some
evidence that older countries (in terms of age) have lower rates of household expenditure on
education per child. These results are also consistent for a larger sample of countries that
includes selected non-OECD countries.
The three essays share a common objective of creating a better understanding of
intergenerational transfers between parents and children in the era of aging, which is the
primary goal of the thesis.
3
Chapter 2
Intergenerational education gap and old-age support
Abstract
Old-age support from children is a crucial part of the income for old-age parents in sev-
eral parts of the world, including China. It is important, therefore, to enquire into the
correlates and determinants of these transfers. Data suggests intergenerational dierences
in the educational levels are also an important determinant and not just the current socio-
economic characteristics of children and parents, such as income, age, educational level, and
geographical location. This paper explores if transfers to parents from children vary with
dierences in educational attainment between them. Results based on the survey of old-age
individuals in China suggest that households with more signicant inter-generational gaps
in educational attainment, systematically receive more transfers from adult non-coresident
children even after controlling for child and parent-specic characteristics including income
levels. Results also show that the education gap is not a proxy for the income gap, and
carries more information about the intergenerational dierences in socio-economic status.
2.1 Introduction
Financial security and support in the old age are of great importance. People invest in
pension schemes, and governments spend on social security so that people are nancially
4
adequate and stable in their old age. Another vital contributor to old-age support is di-
rect transfers from children. Many papers have explored the old-age security motive behind
child-bearing (Nugent (1985), Willis (1979)), implying that the transfers from children form
a crucial part of the old-age income source. In China, children are legally required to take
care of the parents in their old age, as enshrined in the Chinese Constitution, the Marriage
Law, and the Law on Protection of the Rights and Interests of Older Persons (Chou (2011)).
It is in conformance with the tradition in the Chinese society of providing support to the
old-age parents (Lin & Yi (2011), Gu & Vlosky (2008)). One estimate from a recent Chinese
household survey suggests that almost 60% of the income of the parents (who are receiv-
ing old-age support from their children) can be attributed to the transfers from children
(Choukhmanne, Coeurdacier, & Jin (2013)).
Given that children are an essential source of income for old-age parents, it is useful
to investigate those family characteristics that are likely to be correlated with the transfers
from children to parents. These characteristics could be parents' socio-economic status,
their health condition, their age, location, and child's socio-economic status, health, and
so on, as well as non-tangible characteristics such as the nature of upbringing of children,
attitudes towards risk and old-age. There are also social characteristics that may in
uence
these transfers, such as social norms, legal requirements, better societal support to the old-
age adults, and so on. In sum, there are a plethora of factors that may play a role in
determining the transfers from children to their parents in their old age.
This paper seeks to contribute to the expanding literature on intergenerational trans-
fers and old-age support. It aims to investigate the extent to which intergenerational dier-
ences in educational attainment is related to old-age support from children in the context of
China. Using household survey data from China, I explore whether children who do better
than their parents in terms of educational status also provide more nancial and in-kind sup-
port to their parents. I nd that a larger educational gap is associated with larger transfers
by children, all else equal. Also, this result holds even after controlling for the income gap
5
between the children and the parents, hence establishing the importance of socio-economic
factors in determining the transfers over and above the nancial status.
Analysis of these factors gains further importance due to the prevailing pension system
and population aging in China. It is estimated that by 2050, the proportion of people aged
60 years and above in China will constitute almost 36% of the total population, the current
gure being around 20% (UN World population prospects 2019). The old-age dependency
ratio in China is about 15% (world average being 13.5%), having risen by ve percentage
points since 2000. Moreover, the pension systems in China provide limited support to their
elderly population, especially in rural areas (Sin (2005)). Even though eorts to reform
the pension system in China is underway, transfers from family still constitute a signicant
portion of parents' income in the old-age (Zimmer & Kwong (2003)). The increased burden of
the old-age population and declining support ratios seriously compromises the government's
ability to provide old age security.
The rest of the paper is organized as follows: the next section contains a discussion
of the relevant literature. Section 3 discusses the data and presents some trends relating
to parent and child characteristics and the transfers received in the old-age. Section 4
species and describes the regression model, and section 5 shows the results. The last
section concludes.
2.2 Literature
As discussed above, various social, economic, and cultural factors are responsible for de-
termining the direction and magnitude of intergenerational transfers. Traditional and lial
norms for providing support to old-age parents are well established in the East-Asian cul-
ture. Lin & Yi (2011), in their study on old-age support in China and Taiwan, nd that
patriarchy and family bonds play a role in dictating intergenerational transfers. Filial norms
6
are an important determinant of support even after accounting for available resources with
children. Theerawanviwat (2014) estimates that almost 60% of parents in Thailand receive
nancial support from their adult children based on a survey in 2009. He also nds that the
likelihood of transfers increases with the number of children; however, the magnitude does
not.
Xie & Zhu (2009) discuss patriarchal-aspect and gender-aspect of the transfers using a
dataset on urban China. They nd that married daughters provided more nancial support
to parents than married sons in the late 1990s. Most of these transfers could be traced to the
daughters' level of income and education. Another study (Zhu (2016)) on the same dataset
also nds positive links between a child's education and transfers to parents. The paper
also discusses the possible exchange or reciprocity motive wherein the more educated child
also transfers more in order to repay the parental investment made in his/her education.
Ma (2017) nds that child's educational level positively impacts parents' health in old age
through better provision of social support and improved access to resources. It exploits the
natural variation in the implementation of the compulsory education law to pin down the
impact of aging on parents' later-life outcomes.
Literature also suggests the prevalence of need-based support from children. Lee &
Xiao (1998) analyzed data from a national survey conducted in China in 1992 and found
that usually, support from children compensates for the inability to access public resources
by the elderly parents. Parents with lower income levels and poor health status received
more old-age support from children. They also nd a clear dierence in the rural and the
urban regions; transfers from children were found to be more prevalent in the rural areas.
It could also be an outcome of poor access to public pensions in rural areas, along with the
prevalence of strong lial norms (Gu & Vlosky (2008)).
Lei et al. (2012) study the correlates of transfers to parents using a fairly recent dataset (the
study is based on CHARLS pilot survey 2008). They establish that even with the advent of
7
urbanization and shrinking family sizes, the direction of transfers in China is primarily from
children to parents. They nd that characteristics such as the number of siblings, education
level, and marital status are correlated to transfers. They also nd that transfers are positive
for low-income parents and become negative as income goes up.
China's aging population and low fertility have also created concerns regarding the
old-age support for future generations. The probability of receiving support in old-age is
positively correlated to the number of children (Lei et al. (2012), Lee & Xiao (1998)). Zimmer
& Kwong (2003) analyze the role played by the family size in determining the transfers and
predicts that future transfers (nancial and instrumental) will only moderately decline with
shrinking family sizes and not in the same proportion as the total fertility.
There have also been several theoretical studies in the eld of intergenerational trans-
fers. Pioneering research in this eld began with models that had features of intergenera-
tional and intra-household transfers to study the household decision-making in-depth (Barro
& Becker (1989), Becker (1974)). Others have sought to theorize the motives behind these
family transfers. Altonji, Hayashi & Kotliko (1997) used specic criteria to show that inter-
generational transfers between parents and children are not entirely altruistic based on the
PSID dataset. Raut & Tran (2005), on the other hand, nd evidence of two-sided altruism
between parents and children from the Indonesian Family Life Survey data.
In this paper, it is analyzed if intergenerational dierences in educational attainment
matter for transfers, after controlling for various other characteristics of children and parents
documented in the existing literature. Studies have shown how transfers to low-income
parents dier from transfers to high-income parents, with the former group receiving more
(Lei et al. (2012), Imrohoroglu & Zhao (2018b)). An important lesson from these studies is
that children of low-income parents compensate by transferring more, given their income. It
is consistent with the two-sided altruism motive. If intergenerational dierences in the social
8
and economic status matter for transfers, then educational dierences between parents and
children could be a signicant factor.
2.3 Data
The paper uses the China Health and Retirement Longitudinal Study (CHARLS) data for
2015, which is the most recent wave of this panel data set. This latest wave consists of
21,095 households with respondents aged 45 and above. CHARLS focuses on the middle-to-
old population and their economic and health outcomes, making it an ideal data set for the
study. It has detailed questions on nancial and in-kind transfers received and given by the
respondents, as well as the sources of these transfers. It also contains dis-aggregated data
on the respondents' income as well as the educational attainment of all the family members,
along with basic demographic information.
Sample
The data on children of the respondents from the CHARLS 2015 wave is used. The sample
is restricted to adult working-age children (between ages 18-60). Also, parents who are less
than 60 years old (minimum age between a couple) are excluded. One non-coresident child
is randomly chosen from each household so that the sample consists of only one child per
family. All observations with 0 or missing income for the children are dropped. Transfers
from children to parents include both nancial transfer and non-monetary support (value
evaluated by the individual respondents). Transfer data is available for only non-coresident
children. It is desirable since transfers are comparatively easy to measure for non-coresident
children. To eliminate nonsensical values and extraordinary transfers from the sample, top
1%, and bottom 1% of the gross transfer values are excluded. The nal sample consists of
2,415 observations (parent-child pairs) for the year 2015.
9
Trends in educational attainment and transfers
Table 2.1 presents summary statistics for parent and child characteristics. In China, the
maximum years of schooling possible is 21 years. CHARLS provides data on educational
attainment by the level of schooling and not by the years. For this study, the level of schooling
is converted to years of schooling. For details on these levels and the corresponding years of
schooling, refer to Table A.1 in the appendix. The average educational attainment of parents
is around four years, whereas it is nine years for children. It is a substantial dierence. The
average parent in the sample is born around 1947 and the average child around 1974. Hence,
a good portion of children is aected by the educational reforms in China in the 1980s. An
important component of these reforms was the compulsory education law, which made it
compulsory for children to acquire at least 9 years of schooling (Ma (2017), Tsang (1996)).
It is likely to have played a role in not only the intergenerational dierence in educational
attainment observed but also in the future old-age transfers.
Table 2.1: Descriptive Statistics
Variable N Mean Std. Dev. Min Max
Parents' years of schooling 2415 4 3 0 18
Parents' age 2155 68 6 60 90
Parents' income 1988 18226 35987 6 803280
No. of children 2415 3 1 1 13
Child's years of schooling 2415 9 4 0 21
Child's age 2415 41 8 18 60
Child's income 2415 43354 45507 1000 300000
Gross transfers 2415 2119 3858 0 30000
Parents' schooling and age respectively is the average of the schooling and age of both parents.
All incomes and transfers are expressed in current Yuans.
For details on how income levels are computed please see below
The summary statistics also show the dierence between the income levels of parents and
children. However, note that these are current income levels and for retired parents current
incomes would naturally be lower. I calculate parental income as the sum of earned income,
pensions, government transfers, interest income and income from other sources for both
10
parents in the last year. Child's income is the total income earned in the last year as
reported by the parent or the head of the household. CHARLS only provides information
about the income bracket in which the the child's income falls, with the highest bracket
of 300,000 Yuans and more. In order to maintain comparability across the income levels
of children and parents, I classify the income of the parents into the same income brackets
as children's, in the data, for all further analysis. Table A.2 in the appendix provides an
overview of these income brackets and the income distribution in the sample.
Finally, the summary table also reports the transfers received by the parents. Gross
transfers are dened as the sum of total money support and total in-kind support from the
non-co-resident child to the parents in the past year. Net transfer is obtained by subtracting
total support (monetary and in-kind) provided by the parents to the child from the gross
transfers. We see that for the overall sample, average gross transfers account for approx-
imately 12 % of parents' current average income, and 5% of the child's current average
income.
Figure 2.1 shows how gross transfers and net transfers change with the educational
attainment of children in panels (a) & (b) respectively. Child's education is measured as
number of years of schooling. Not surprisingly, more educated children transfer more. This
increase is most pronounced for the children who have more than 10 years of schooling.
Sample is further divided into two, on the basis of parental education: the rst group being
the high education parents (the ones who have completed atleast the upper secondary level)
and the second group consists of the low education parents (the ones with educational level
of less than lower secondary). Children of low education parents transfer more than children
of high education parents irrespective of their own education level. The gap in the education
level of the parents and the child matters because the amount of transfers go up exponentially
for low education parents as the education level of their child increases. This is illustrated
by the fanning out of the curves towards the end in the two plots. It is an indicator of how
11
0 1000 2000 3000 4000 5000
Average gross transfer (yuans)
0 5 10 15 20
Child's education (years)
Low education parents Whole sample
High education parents
(a)
0 5000 10000
Average net transfer (yuans)
0 5 10 15 20
Child's education (years)
Low education parents Whole sample
High education parents
(b)
Figure 2.1: Average transfers by education level
12
intergenerational mobility in education contributes towards old-age support and nancial
assistance for the old-age parents.
Table 2.2: Distribution of population by education level (% of the sample population)
Child Parent 1 Parent 2
Less than lower secondary 36.7 90.3 93.5
Upper secondary and vocational training 48.4 7.5 5.1
Tertiary 14.9 2.2 1.4
From the summary statistics we know that there is a substantial dierence in the educational
attainment across generations. Table 2.2 gives a dis-aggregated view of these dierences. It
shows the population distribution across education levels, separately for children and parents.
Majority of the children are in the upper secondary and vocational training category (9-12
years of schooling). In contrast to that, more than 90% of parents are in the rst category
(less than 6-8 years of schooling). Given the dierence in the education level in the two
generations, an overwhelming percentage of parents are likely to receive greater nancial
support from children, going by the trends in gure 2.1. Table 2.3 reports the average amount
of gross transfers and net transfers by the education gap between parents and children.
Table 2.3: Average transfers by education gap
Edu. gap Child's edu.(years) Parents' edu.(years) Gross trans. Net trans.
1 5.2 4.1 1521.1 826.5
2 8.8 3.7 1770.1 827.9
3 12.1 2.4 2260.6 835.9
Transfers are measured in current Yuans.
The education gap or intergenerational dierence in education is measured as the dierence in
the years of schooling between the child and the parents. Education gap is positive (children
are more educated than their parents) for an overwhelming majority of the sample
1
. The
1. Parents who are more educated than their children form less than 6% of the total sample
13
three categories are created using the three quantiles of the education gap. Category 1
denotes the smallest gap and category 3 denotes the highest gap. In the rst category the
maximum dierence in schooling is 3.25 years, the corresponding number for second and
third categories are 7 and 18 respectively. Both gross and net transfers go up as the gap in
education increases.
2
Table 2.4: Average transfers by child's age & education gap
Gross transfers Net transfers
Edu. gap < 30 yrs. 30-45 yrs. > 45 yrs. < 30 yrs. 30-45 yrs. > 45 yrs.
1 1849.3 1960 1070.1 663.1 976.1 574.4
2 1891.2 2314.35 1637.5 582.7 1184.2 791
3 3195.5 2770.3 1697.9 877.9 1339.4 1259.2
Note: Both kinds of transfers are measured in current Yuans
Literature on old-age support also points to the importance of child's age in determining
the transfers to parents (Lei et al (2012)). It will be interesting to see if the trend of
higher transfers with higher education gap diers for children of dierent age groups. Table
2.4 shows average gross and net transfers by child's age and the education gap. Transfers
increase with the education gap for all the age-groups. Also, there is an inverted-U shaped
relation between transfers and age. This pattern holds across all levels of education gap
except in one case; for the highest gap, gross transfers are highest for the youngest age
group.
Figure 2.2 plots these trends. While for the high education gap category, average gross
transfers are highest for children below 30 years of age, in the low education gap category,
children transfer most between ages 30 and 45. This could be due to high marginal utility
from transfers for the high education gap children at the beginning of their work life.
2. Also note here that gures for gross and net transfers are the averages over large categories in some
cases as compared to others. For instance the last category ranges between 7 to 18 years of dierence in
schooling with the average dierence (within the category) of around 9-10 years, whereas the second category
ranges between 3.5 to 7 years of dierence. Hence, the average transfers do not show a huge leap as in the
gure above. However, this table does establish the trend of increasing transfers with education gap in the
data.
14
0 1,000 2,000 3,000 4,000
Average gross transfer
< 30 yrs. 30-45 yrs. > 45 yrs.
low edu. gap high edu. gap
Figure 2.2: Average transfers by child's age
2.4 Empirical specication
Trends in the data point to an important role played by education, and more so by the edu-
cation gap between parents and children, in determining the old-age support from children.
Apart from this, child's age and other demographic and family characteristics have been
shown to be highly correlated with the old-age transfers, in the existing literature. In order
to analyze the role of the education gap alongside the other established determinants, the
following specication is used for the regression analysis:
log(
i;j;p
) =+
x
log(x
i
)+
g
log(gap
i;j
)+
a
log(a
i
)+
n
log(n
j
)+
Z
i;j
+
p
+
i;j;p
: (1)
Here, i denotes the child providing the transfers, j stands for the parents of the child
3
, and
p is the province of the parents.
i;f;p
denotes the transfers by childi to parentsj, belonging
3. Parents (j) are treated as a joint entity which includes both mother and father. In case where both
parents are present, parents' income is the joint income of both parents and parental age and educational
attainment is the average of the age and education, respectively, of both parents. In case of single parent,
single values are used.
15
to provincep. x
i
stands for the child's socio-economic status as measured by his/her income
or education level. a
i
is the age of the child. gap
i;j
denotes the gap or the dierence in
the socio-economic status of the parents and the child, measured by the dierence in their
income or education levels. n
j
stands for the number of children, and Z
i;j
are other parent-
child characteristics which includes child's gender and parents' age. I also include province
xed eects;
p
.
Given the sample and the empirical specication, following exercises are performed.
Firstly, it is assessed how the transfers are aected by child-level characteristics such as
income, education and age (as seen in previous literature). Then I proceed to analyze if
inter-generational dierence in the socio-economic status can play a role in determining
transfers over and above these characteristics, using the dierence in educational attainment
(between the parents and the child) as a measure.
Note here that changes in the magnitude of transfers with respect to changes in the
various explanatory variables is seen as a result of correlation between them, since it is
dicult to control for endogeneity arising from reverse causality or omitted variables. In
this case, endogeneity from reverse causality arises if parents are believed to invest more in
children with an expectation of support in the old-age. An important omitted variable could
be characteristics of other children in the family, that may play a role in determining the
transfers. As an example, Theerwanviwat (2014) nds that children collude to pool resources
for old-age support of their parents. Theory and intuition does suggest that the underlying
child and parent characteristics and other social characteristics are principal factors that
determine the magnitude of the transfers. Hence, one should look at the patterns of cor-
relation between the suggested characteristics and the transfers along with the magnitudes
of these correlation to get a better idea of how these transfers are shaped, along with using
theoretical knowledge to interpret the results.
16
In equation (1), the transfers (log()) are measured as the log of gross transfers, in order to
circumvent the issues related to negative values. From the description in the data section we
know that qualitatively, net and gross transfers behave in a similar way. Gaps are calculated
as the dierence in the education or the income level between children and parents
4
.
Finally for the estimation technique, the approach similar to Choukhmane et al.
(2013) is adopted. The equation (1) is estimated using the Poisson-pseudo maximum like-
lihood estimator (PPML). It provides consistent estimates of log-linear models for cross-
sectional data. This estimator has also been shown to perform more eciently than other
techniques that deal with censored dependent variables (dependent variables with zero or
missing values). The data need not follow a Poisson distribution and the dependent variable
needn't even be an integer to be able to use PPML technique. See Silva & Tenreyro (2006)
for further details. In the robustness section, I also employ Heckman selection method to
estimate equation (1) and compare the results.
2.5 Results
To begin with, I run the regression in equation (1) using only the income level or the education
level of the child as the indicator for child's socio-economic status, that is holding
g
= 0.
This specication is similar to Choukhmane et al. (2013). He used CHARLS pilot survey
(2008) and CHARLS 2011 wave to analyze the correlation between transfers and child's
`quality'; that is their income or education level. Columns (1) and (2) in Table 2.5 report the
elasticities with respect to the child's income level and the child education level respectively.
Coecient on income is positive and signicant. An increase in income by 1% is associated
with an increase in gross transfers by almost 1.7%. For education the relationship is unit
elastic. More children in the family reduce the transfers from a particular child. Transfers
4. To calculate income gap, the dierence in the income categories to which the child and the parents
belong is measured, see Table A.2 for details
17
are also declining in child's age. Columns (3) and (4) repeat the previous exercise with
additional controls of parents' age and child's gender. The magnitude of elasticity goes up
for both income and education.
5
Table 2.5: Results with child's level of income and education
(1) (2) (3) (4)
Dependent variable : Log(Gross transfers)
Log(Child's income level) 1.727
- 1.801
-
(0.161) (0.203)
Log(Child's edu. level) - 1.038
- 1.097
(0.102) (0.135)
Log(Child's age) -0.112
-0.0915
-0.179
-0.131
(0.0243) (0.0247) (0.0471) (0.0482)
Log(No. of children) -0.426
-0.370
-0.460
-0.343
(0.111) (0.114) (0.137) (0.143)
Other controls No No Yes Yes
Observations 1790 1788 1161 1161
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
I now add further elements of inter-generational dierences within a family in order to
account for parents' socio-economic status along with that of the child's. Table 2.6 presents
results of complete equation (1) with the gap variable being measured by the dierence in
the income levels of the parents and the child. Income gap is a signicantly associated with
the gross transfers to the parents. In fact, transfers go up by approximately 0.3 % for every
1% increase in the dierence in income levels. Column (3) introduces child's education level
as an additional explanatory variable, the coecient on the income gap remains signicant
and is of a greater magnitude. The result for the education hasn't changed much from the
previous table and it is also signicant. This highlights the importance of intergenerational
educational dierences. Even after controlling for child's educational status, the dierence
5. These results are slightly dierent from the original estimation in Choukhmane et al. (2013), where
they nd the relationship to be almost unit elastic in all cases.
18
in the income levels of the parents and the child proves to be an important factor. Income
gap, however, only signals the current economic gap or the current gap in the standard of
living between children and parents. Inclusion of educational dierences could potentially
proxy for long-term dierences in the standards of living. It is explored in the next set of
results.
Table 2.6: Results with income gap
(1) (2) (3)
Dependent variable : Log(Gross transfers)
Log(income gap) 0.284
0.319
0.356
(0.0721) (0.0871) (0.0840)
Log(No. of children) -0.645
-0.690
-0.386
(0.114) (0.141) (0.142)
Log(Child's age) -0.101
-0.182
-0.132
(0.0250) (0.0484) (0.0474)
Log(Child's edu. level) - - 1.120
(0.134)
Other controls No Yes Yes
Observations 1790 1161 1161
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
Table 2.7 presents results with the education gap as the main regressor. From the rst
two columns it can be seen that the gap in education between children and parents is a
signicant and positive in
uence on gross transfers. The elasticity of transfers with respect
to the education gap is slightly higher than the one in case of the income gap. But this is
only the case when child's income is not taken into account.
In column (3), income level of the child is added, coecient on education gap is still
signicant, although smaller in magnitude. This indicates two things: rstly, similar to the
previous table, intergenerational socio-economic gap is an important factor for transfers, and
secondly, intergenerational dierences matter not just in terms of current earnings but also
19
Table 2.7: Results with education gap
(1) (2) (3) (4) (5)
Dependent variable : Log(Gross transfers)
Log(Education gap) 0.462
0.424
0.320
0.321
0.395
(0.106) (0.107) (0.104) (0.104) (0.107)
Log(No. of children) -0.564
-0.581
-0.418
-0.393
-0.626
(0.140) (0.144) (0.139) (0.140) (0.144)
Log(Child's age) -0.143
-0.166
-0.168
-0.168
-0.168
(0.0365) (0.0487) (0.0467) (0.0467) (0.0482)
Log(Child's income level) - - 1.735
1.705
-
(0.201) (0.202)
Log(Parent's income level) - - - 0.0945 -
(0.0860)
Log(income gap) - - - - 0.292
(0.0875)
Other controls No Yes Yes Yes Yes
Observations 1165 1161 1161 1161 1161
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
20
in terms of education, which encompasses other characteristics such as inter-personal skills,
social background and so on, apart from signaling the earning ability. In column (4), parents'
income level is included as an additional control to conrm if the education gap is not merely
substituting for income gap. Results do not change much from column (4). Finally, Column
(5) the income gap is included directly in place of child's and parents' income levels. Results
indicate that both income and education dierences are relevant. This is important since it
indicates the role played by inter-generational mobility from socio-economic point of view
in determining the old-age support from children. Overall, it is seen that the education
gap elasticity is positive and ranges between 0.32 to 0.46. In the next section, I discuss
some alternative samples and methodology to check if the results are robust to alternative
specications.
2.6 Robustness
Following robustness checks are conducted. Firstly, equation (1) is estimated separately for
the following sub-samples; sample of parents from urban regions only, sample of parents from
rural regions only and, sample of children residing in the main cities only. In the second
exercise, another wave of the CHARLS dataset is used: households from wave 2013 with the
same sample selection criteria as with wave 2015 are analyzed. Lastly, as an alternative to
the PPML technique, Heckman estimation method is used to estimate equation (1).
Several papers point to the urban-rural dierences in China with respect to the old-
age support received by the parents due to the dierences in the access to public resources
such as pensions and the dierences in the socio-economic status. They also conduct separate
analysis for the two regions (see Zimmer & Kwong (2003) and Lee & Xiao (1998)). In order
to analyze the dierences across rural and urban regions with respect to education gap and
transfers, I also run equation (1) separately for the two regions using education gap as the
primary regressor. Columns (1) and (2) of Table 2.8 show the results for the sample of
21
Table 2.8: Results with education gap by regions
(1) (2) (3) (4) (5) (6)
Dependent variable : Log(Gross transfers)
Log(Education gap) 0.416
0.388
0.369
0.312
0.310
+
0.273
(0.170) (0.172) (0.134) (0.132) (0.179) (0.178)
Log(No. of children) -0.597
-0.632
-0.371 -0.348 -0.507
-0.584
(0.180) (0.178) (0.246) (0.247) (0.203) (0.203)
Log(Child's age) -0.156
-0.153
-0.161
-0.167
-0.0889 -0.0851
(0.0743) (0.0739) (0.0605) (0.0593) (0.0626) (0.0607)
Log(income gap) - 0.175 - 0.510
- 0.318
(0.124) (0.132) (0.124)
Other controls Yes Yes Yes Yes Yes Yes
Observations 481 481 680 680 417 417
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
parents residing in the urban areas only. There are a total of 481 observations. In the
second column I also add the income gap. Coecient on the education gap is positive and
signicant and similar in magnitude to the one in Table 2.7. Coecient on the income gap is
not signicant anymore. However, for the rural sample; that is columns (3) and (4), income
and education gap are both signicant. Education gap plays a more important role for the
urban residents, also the education and the income gap are more closely related to each other
for the urban sample. The coecient on child's age is negative and signicant everywhere.
However, it can be seen that the negative relationship between transfers and the number of
children is only signicant for the urban sample, implying that the negative family size eect
on transfers, mainly holds for the urban parents.
Finally, columns (5) and (6), show results for the sample of children living in the main
cities. Results here are less conclusive, education gap is signicant only when income gap is
not included.
22
Table 2.9: Results with education gap: wave 2013
(1) (2) (3) (4) (5)
Dependent variable : Log(Gross transfers)
Log(Education gap) 0.458
0.446
0.327
0.327
0.440
(0.108) (0.108) (0.106) (0.106) (0.109)
Log(No. of children) -0.841
-0.891
-0.763
-0.719
-0.906
(0.126) (0.127) (0.125) (0.127) (0.127)
Log(Child's age) -0.0650
+
-0.112
-0.118
-0.116
-0.112
(0.0332) (0.0464) (0.0462) (0.0462) (0.0465)
Log(Child's income level) 1.763
1.712
(0.242) (0.245)
Log(Parent's income level) 0.173
+
(0.0984)
Log(income gap) 0.108
(0.0958)
Other controls No Yes Yes Yes Yes
Observations 1187 1187 1187 1187 1187
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
23
Table 2.9 runs the exact same specication as that of Table 2.7, but with the data from
the wave 2013 of the CHARLS dataset. Running the regression on 2013 data gives us the
benet of looking at the same households or similar households at another point in time.
Sample selection criteria remains the same. The coecient on number of children is larger for
this sample and still highly signicant. The coecient for the education gap is also slightly
bigger; a percent change in the education gap is associated with almost 0.45 % increase
in transfers. Overall, the results are quite similar to 2015 wave. However, note that the
coecient on income gap in the last column is no longer signicant and the coecient on
the education gap is almost as large as the rst two columns. Inclusion of the income gap
hasn't been able to add any extra information to the analysis of the equation (1).
Finally for the last robustness check, an alternative maximum likelihood estimation
approach is used; the Heckman selection model. This model corrects for the bias arising from
non-random selection of the dependent variable. In other words, it accounts for the zero
values in the dependent variable, by simultaneously evaluating the conditional probability
of each of the non-missing (or non-zero) value of the dependent variable
6
, along with the
calculation of the conditional expectation of the dependent variable in equation (1) (Heckman
(1979)). This approach is equivalent to using a Tobit model for censored dependent variables.
Columns (1) and (2), of the Table 2.10 present results for education gap only. The coecient
on education gap is signicant and smaller by roughly, 25 % than the value in Table 2.7.
Qualitatively, we arrive at similar results as before. In column (5), education gap remains
signicant even after the inclusion of the income gap.
6. conditional on a given set of independent variables which determine the probability of non-zero transfers.
In our sample these are parents' age, child's age, child's gender, parents' income level, and the number of
children.
24
Table 2.10: Results with education gap: heckman estimation
(1) (2) (3) (4) (5)
Dependent variable : Log(Gross transfers)
Log(Education gap) 0.318
0.305
0.234
0.219
0.284
(0.0765) (0.0770) (0.0749) (0.0728) (0.0769)
Log(No. of children) -0.391
-0.413
-0.303
-0.360
-0.438
(0.0981) (0.101) (0.0988) (0.108) (0.100)
Log(Child's age) -1.061
-1.188
-1.200
-1.423
-1.184
(0.260) (0.356) (0.345) (0.381) (0.355)
Log(Child's income level) - - 1.203
1.154
-
(0.135) (0.130)
Log(Parent's income level) - - - 0.0968 -
(0.0677)
Log(income gap) - - - - 0.214
(0.0640)
Other controls No Yes Yes Yes Yes
Observations 1555 1555 1555 1555 1555
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
25
2.7 Conclusion
This paper analyzes the potential correlates of the transfers from children to old-age par-
ents. While many dierent factors come together to determine how much children transfer
to parents, the above analysis focuses on studying the impact of socio-economic mobility in
the family from parents' generation to the child's generation. Intergenerational dierences in
terms of educational attainment is an important factor as far as the transfers to old-age par-
ents are concerned. A larger educational gap between parents and children is associated with
larger transfers from children to parents. The result holds after controlling for geographical
factors, family and child-specic characteristics, and even the parents' and children's income
levels. Thus education is a more comprehensive measure of a person's standard of living and
also of long-term productivity and transfers are not only determined by current nancial
status but also by the long-term economic and social status.
This paper also has the advantage of using a fairly recent dataset, which adds to the
knowledge from the previous literature on transfers and intergenerational characteristics for
more recent times. However, an important caveat here is that the patterns from the technical
exercise above should not be confused with causal relations, and must be supplemented with
knowledge from theoretical models.
In-fact these results can also provide useful information for modeling purposes. The
information provided by the data, as well as the insights from the above analysis, is used
in an overlapping generation model (presented in full detail in the next chapter), which
studies the household decision-making process concerning investing in children, saving, and
providing for old-age parents. This exercise has shown how old-age support is correlated with
not only individual education and income level, but also the inter-generational dierence in
these levels. This analysis is also helpful for policy decisions on old-age pensions, social
security taxes, and expenditure on education.
26
Chapter 3
Does population aging matter for human capital
investments in children ?
Abstract
Population aging is steadily becoming a pan-world phenomenon. As the population ages,
household decision-making process concerning how much to save, how much to consume
and how much to spend on children, changes accordingly. This paper analyzes the impact
of population aging on human capital investments in children using an overlapping genera-
tions framework with intergenerational transfers and two-way altruism. Parents decide on
how much to invest in children, along with other lifecycle decisions, while facing uncertain
lifetimes with declining mortality in the old-age. Simulations of the model using the popu-
lation data from China show a detrimental impact of aging on human capital investments in
children.
3.1 Introduction
We live in an era of progressively aging societies world over. As we fast transition towards an
aging world, with rapidly falling death rates along with declining fertility, several challenges
are cropping up in the form of slowing economic growth, deepening economic inequalities,
27
and other economic challenges (Deaton & Paxson (1994, 1998), Maestas, Mullen & Pow-
ell(2016)). Aging of the population and the demographic transition has led to an intergener-
ational redistribution of resources towards the old-age population (Song et al. (2015), Walker
(1990)). This alteration in resource allocation is likely to aect human capital investments in
children, among other things. Accumulation of human capital matters for future economic
growth and development and is essential for the long-run prosperity of an economy.
Over time, the age structure of the population has shifted towards older age groups for
an increasingly large number of countries. As of 2015, the average old-age dependency ratio
for the world was 12.5%, having climbed a total of 4 percentage points from 1960. Moreover,
the rate of aging has gone up with better health care, technology, and attitudinal shifts.
By 2050, almost 22 % of the world's population would be above 60 years of age.
1
Given
these projections, we expect households to optimize on the distribution of resources between
present and future, keeping in mind the longer life spans (Ehrlich Lui (1991)). Investing in
children's human capital is one of the many lifecycle decisions made by households facing
nite and uncertain lifetimes. The opportunity cost of investing in children's human capital
is the foregone savings for the future minus the potential benets from these investments.
The question is, how would aging impact this opportunity cost? Would it exacerbate the
potential benets and costs associated with investing in the human capital of children, and
how would that alter the course of these investments?
To this end, this paper develops a simple overlapping generations model wherein
successive generations face lower and lower mortality in the old age, and parents invest in
the human capital of their children who form the future generations. The model is then used
to simulate an economy with real-life population data to obtain optimal trajectories for the
household portfolio of decisions. These decisions are evaluated under dierent demographic
scenarios. It is found that population aging, as seen in the real world, has led to lower
investments in the human capital of children when compared to a counterfactual scenario
1. Source: UN World Population Prospects 2019
28
with a constant mortality rate. The analysis with alternative demographic scenarios also
helps pin down the roles played by declining mortality and declining fertility in shaping
these investments. In the next chapter, insights from the model simulations are also applied
to cross-country data to nd evidence for systematic dierences across older and younger
countries with respect to the investments in children's human capital.
The next section provides a brief overview of the problem being explored in the paper
and relevant literature. Section 3 presents the model, and section 4 discusses the quantitative
exercise undertaken and the results obtained from the simulations. The last section provides
the concluding remarks.
3.2 Background
As discussed in the introduction, the elderly population is burgeoning everywhere. While
there are cross-country variations in the age composition of the populations, the world as a
whole is making a demographic transition with declining fertility and mortality rates. As
incomes rise and societies develop, they move from a high birth rate and high death rate
system to a low birth rate and low death rate system (Lee (2003)). An obvious by-product
is the heightened proportion of older people in the population.
Figure 3.1 gives a snapshot of the demographic evolution in a selected set of countries
between 1950-2050
2
. There has been a steep increase in the world-average for the old-age
dependency ratio since 2010. Meanwhile, the fertility rates are projected to fall below 2,
for all major economies of the world by 2025, with the oldest countries having the lowest
rates. The pattern of mortality decline over time can be clearly seen from the life expectancy
graph. The current world average is well above 20 years, driven mostly by the developed
nations. China is fast catching up and is projected to surpass the world average in another
ten years. China's rapid demographic transition is evident in these graphs, with a steep fall
2. Source: UN World Population Prospects 2019
29
0 20 40 60 80
Old-age dependency ratio
1950 1970 1990 2010 2030 2050
World India China
Italy Mexico USA
Japan
(a)
0.00 2.00 4.00 6.00 8.00
Total fertility rate
1950 1970 1990 2010 2030 2050
World India China
Italy Mexico USA
Japan
(b)
10.00 15.00 20.00 25.00 30.00
Life expectancy at age 60
1950 1970 1990 2010 2030 2050
World India China
Italy Mexico USA
Japan
(c)
Figure 3.1: Demographic trends across the world
30
in the fertility around the 1970s and a rapid increase in the old-age dependency and life
expectancy rates.
The central point here is that the age demographic structure is evolving. Studies that
look into the impact of population aging on the economy have provided useful insights on
the impact of changing demography on various macroeconomic indicators. Several works
have looked at the eects of aging on various important factors such as labor productiv-
ity (Maestas, Mullen & Powell (2016)), capital markets (Boersch-Supan & Winter (2001)),
inequality (Deaton & Paxson (1994, 1997, 1998)), social security systems (De Nardi et al
(1999)), savings (Imrohoroglu & Zhao(2018b), Deaton & Paxson (1994)), economic growth
and welfare (Lee(2016), Song et al. (2015), Ehrlich & Lui(1991)).
A key channel through which population aging impacts the macroeconomic environ-
ment is household decision-making. As population ages, household saving behavior changes,
with a greater inclination towards old-age security (Deaton & Paxson (1994), Boersch-Supan
& Winter (2001), Imrohoroglu & Zhao (2018b), Curtis et al. (2015)). By the same token, it
can be expected that aging alters the household's incentives with respect to human capital
investments in children. It can happen on two levels.
Firstly, investing in children's human capital is also one kind of investment for old-age.
Well-educated and well-earning children are likely to provide greater support to the parents.
It is a regular empirical nding, see chapter 2 for details. Transfers from children are of
greater importance in economies wherein old-age pension systems are not well developed, for
instance, developing nations like China (See Gu & Vlosky (2008) and Sin (2005)). Secondly,
investment in children is also undertaken for emotional gratication. Parents provide for
their children and invest in them because they intrinsically care about their children's well-
being (See Becker (1981)). However, with aging, the intrinsic utility derived from investing
in children might get altered, as individuals begin to discount the future dierently with
longer lives. These intergenerational links make human capital investments in children a
31
crucial aspect of the household decision-making process. In fact, its inclusion in the models
that study household behavior with respect to demographic changes is essential not only
because it adds a realistic aspect to the model but also because it helps understand the
interplay between household preference for saving versus investing in children. To the best
of my knowledge, Ehrlich & Liu (1991) is the only other paper that discusses the impact
of increased longevity in the old age on human capital accumulation in children using a
theoretical model of economic growth. However, it does not show how investments in the
human capital of children change with aging in the presence of savings. Here, I am able
to do so through a quantitative simulation of the model, which is designed to look at this
problem.
Finally, also note that most of the literature that discusses human capital investments
in children is centered around fertility choices (Barro & Becker (1989), Manuelli & Seshadri
(2009), Choukhmane, Coeurdacier, & Jin (2013), Daruich & Kozlowski (Forthcoming). In
this paper, the focus is on the other aspect of the demographic change; declining mortality.
The objective here is to analyze if declining mortality in the old age has impacted human
capital investments in children. The goal is to pin down the role of changing age demography
in shaping these investments, while also distinguishing between the contribution of declining
mortality and declining fertility.
3.3 Model
Consider an over-lapping generations economy wherein a continuum of identical agents are
born every period. An agent enters economic life at age 20 and lives for a maximum of
85 years. A household consists of an agent and its children (born to the agent at age 20).
Children stay with the parent between ages 0-19, at age 20 they move out to make their
own household. Agents work till age 63 and retire at 64. All individuals are endowed with
1 unit of labor which they supply inelastically to earn labor income. A working household
32
saves, consumes and invests in children's human capital. It also transfers funds to retired
parents. A retired household
3
consumes, saves and receives transfers from its children. After
retirement agents face a time-specic probability of survival each year. Upon the death
of an agent, his remaining assets are distributed equally amongst its children as accidental
bequests.
There are three stages in the life of an economically-active agent: working parent with
co-residing dependent children (20-39 years of age), working parent with non co-residing
non-dependent children (40-63 years of age), retired parent with non co-residing children
providing old-age transfers (64- max. 85 years of age). This timeline is summed up in Figure
3.2.
Figure 3.2: Timeline of an agent's life
Stage 1
All agents of a given cohort are ex-ante identical. Per-period utility function of a working
parent belonging to cohort c and age j2 [20; 39] is:
u
c;j
=
c
1
c;j
1
+
1
(n
c;j
)
2
h
k
c;j
1
1
(3.1)
3. Note that a retired household consists of only the agent since the children move out when the agent
turns 40
33
Here, c
c;j
is the consumption of the agent of cohort c and age j. h
k
c;j
is the human capital
of each child of the the agent of cohort c and age j
4
. The term
1
n
2
c;j
denotes the discount
rate or the weight on child's human capital.
1
2 [0; 1] determines how much the agent
cares about human capital of its kids.
2
2 [0; 1]
5
is the curvature on number of children.
This formulation, with a warm-glow motive for the child's well-being is similar to Barro &
Becker (1989). The parameter denotes the coecient of risk aversion or the inverse of
inter-temporal elasticity of substitution.
Agents invest an amount x in children per period and transfer a fraction of their income
(
c;j
) to their parents. They face the following budget constraint:
c
c;j
+n
c
x
c;j
+a
c;j+1
= (1
c;j
)w
c;j
+ (1 +r)a
c;j
+b
c;j
; j2 [20; 39] (3.2)
w
c;j
is the labor income of the agent which is a function of his own human capital and
experience (discussed in detail later). Agents transfer funds to parents only after their
parents retire, hence,
c;j
> 0 if parents are retired, and 0 otherwise. Agents also invest in a
risk-free asset a
c;j+1
, and they are not allowed to borrow. Lastly, b
c;j
denotes the accidental
bequest received upon the death of the parent, and are calculated as follows:
b
c;j
=
c20;j+20
(s
c20;j+20
) (3.3)
Here,
c20;j+20
is the probability of the agent's parent dying at the age j + 20, that is when
the agent is j years old, and s
c20;j+20
is the total saving of the parent at the time of their
death. Note that before retirement age all agents live with certainty and hence = 0 for
ages below 64.
4. superscript k is used to denote the human capital of the kid.
5. indicating diminishing returns to utility with number of children.
34
Stage 2
After the children have moved out, agents no longer invest in their human capital. Everything
else remains the same as before. The Stage 2 per-period utility function and the budget
constraint are as follows:
u
c;j
=
c
1
c;j
1
; j2 [40; 63] (3.4)
c
c;j
+a
c;j+1
= (1
c;j
)w
c;j
+ (1 +r)a
c;j
+b
c;j
; j2 [40; 63] (3.5)
Stage 3
Finally, after the agent retires he faces a time and cohort-specic probability of survival,
denoted by
c;j
. An agent lives with certainty before retirement, hence,
c;j
= 18j < 64,
and,
c;j
< 18j2 [64; 85]. The per-period utility is the same as stage 2 and is given by the
equation (3.4). The agent now consumes out of the transfers received from his children, and
the budget constraint looks like:
c
c;j
+a
c;j+1
=f
c;j
+ (1 +r)a
c;j
+b
c;j
;j2 [64; 85] (3.6)
f
c;j
denotes the support from the children. f
c;j
=n
c
c+20;j20
(w
c+20;j20
). Here,w
j20;c+20
is
the wage of agent's child when the agent is j years old, and
c+20;j20
is the transfer share.
Human capital production function and earnings
Individuals stay with parents between ages 0-19 and accumulate human capital. For every
unit of investmentx
c;j
made by a parent of cohortc and agej, the human capital produced
is:
h
k
c;j
=(x
j;c
)
; 0<< 1 (3.7)
35
indicates the initial ability of the child and is the returns to scale on investment in the
human capital of the child. We assume 0 depreciation in accumulation of human capital so
that the total human capital accumulated at the beginning of the agent's economic life, that
is at age 20, is equal to
h
c+20
=
P
39
j=20
h
k
c;j
. Subscriptc + 20 denotes the cohort of the child.
Once the agent turns 20, he/she begins working and carries the same human capital stock
for life. Earnings are a function of this human capital stock;
w
j;c
= w
j
(1 +
h
c
)
; 0<< 1 (3.8)
w is the common component in the wage function which is xed over time and across indi-
viduals.
j
denotes the age-eciency prole.
Maximization problem
A 20-year old agent of cohort c maximizes its lifetime utility subject to the age-specic
budget constraints:
Max
fc
c;j
;a
c;j+1
;x
c;j
g
U = E [
P
39
j=20
^
j20
(
1
(n
c
)
2
h
k
c;j
1
1
+
c
1
c;j
1
) +
P
85
j=40
^
j20
(
c
1
c;j
1
)],
^
j20
=
8
>
>
<
>
>
:
j20
j2 [20; 63]:
c;j
j20
j2 [64; 85]:
subject to,
(A) Budget constraints; eqns. (3.2), (3.5) and (3.6)
(B) Human capital production function; eqn. (3.7)
(C) Human capital accumulation;
h
c+20
=
P
39
j=20
h
k
c;j
The above maximization problem is solved recursively for each cohort, in order to obtain the
optimal values of investment in human capital of children along with savings and consump-
tion trajectories. The two demographic factors that vary across the cohorts are : number
36
of children (n) and the survival probabilities (
j
). Also note that there are two intergener-
ational links in the model. Firstly, parents invest in children's human capital and therefore
determine their future income. Secondly, children support their parents in the old age by
transferring funds to them. Human capital investment in children is determined by the in-
come level of the parents when they are working. The investments that they make in their
kids, in turn, determine their income when retired. The objective is took look at the invest-
ments made by the parents in their children, given this network of intergenerational transfers
and changes in the demographic structure. In order to do so, a series of generations or co-
horts are simulated that face given mortality and fertility rates (based on real-life population
data), and the economic problem of the representative agent of each cohort is then solved
successively. The successive cohorts are linked to each other due to the intergenerational
transfers discussed above.
3.4 Quantitative Exercise
In this section, demographic data is used to run the model simulations. This involves solving
for the life-cycle problem of a representative agent in each cohort, for all the cohorts suc-
cessively. The changes in the investments and the savings across cohorts are mapped over
time, and compared across dierent demographic scenarios to understand the role of aging
in driving human capital investments by households.
3.4.1 Parametrization
The baseline model uses the Chinese population data (1871-2009) to estimate cohort and age-
specic survival probabilities (
), the cohort-specic child-parent ratio or the fertility rate
(n), and proportion of population in dierent age groups (childhood, working and retired).
37
400 600 800 1000 1200 1400
Total population (in mill.)
1870 1890 1910 1930 1950 1970 1990 2010 2020
year
(a)
1 1.5 2 2.5 3 3.5
Child-parent ratio (n)
1870 1890 1910 1930 1950 1970 1990 2010
year
(b)
.85 .9 .95
Average Survival probability post retirement
1870 1890 1910 1930 1950 1970 1990 2010
year
(c)
0 .05 .1 .15 .2
Old-age dependency ratio
1870 1890 1910 1930 1950 1970 1990 2010 2020
year
(d)
Figure 3.3: Demographic trends for the baseline model
38
Figure 3.3 shows the demographic trends in China beginning 1871. These are based on the
Chinese population data obtained from the UN World Population Prospects 2019. Total
population growth in China accelerated in the 1950s. From the graphs in the background
section (Figure 3.1), it can be seen that China has had a unique experience as far as the
demographic transition is concerned. This is primarily due to the family-planning measures
adopted by the Chinese government in the second half of the twentieth century. Between
1950 and 1975 China witnessed a period of rapid mortality decline attributable to economic
development and improvement in health and education services. The fertility rate remained
high till the early 1970s, after which it began to decline in the earnest due to the introduction
of strict family-planning measures (Peng (2011)). In 1980, China adopted a nation-wide one-
child policy regime to limit the population growth.
The fertility rate or the child-parent ratio
6
, shown in panel (b), is declining for the
sample period till the early 1930s. Beginning 1930s, it started to go up and kept rising
till 1970. The adoption of one-child policy in 1980 is marked by the red line. Average
survival probabilities have consistently gone up with the sharpest rise post 1950s. Average
survival for individuals aged 64 plus has increased by more than 5 percentage points in the
60 years beginning 1950. Finally, in panel (d), old-age dependency ratio has shown a similar
acceleration post 1950s. Please see appendix Figure B.1 for the age-wise distribution of
population for the sample period to get a clear picture of the evolution of population over
time.
Values for the other parameters in the model are assigned as follows. The coecient
of relative risk aversion = 1:5. The discount rate is calibrated to match the long-term
average capital-output ratio (between 1950-2009) in China
7
. The value obtained is 0.99. The
interest rate on risk-free bonds is xed at 4%.
6. This is the ratio of total number of children to total number of parents for each year in the sample
period. Note that, in the model, children population is population from ages 0 to 19, and parent population
is population from 20 to 39
7. Average capital output ratio in China for the period 1950-2009 is equal to 2.3, source: Penn World
Tables 9.1
39
For the discount rate on children's investment in the utility function, we need to estimate two
parameters: weight on the investment
1
, and the curvature on number of kids
2
. Former
is estimated using the value of average expenditure share of households on child's education.
Based on dierent sources the values lies between 15% 20%
8
. I set
1
= 0:18. In order
to set the value of
2
, I use the estimates from a study that analyzes parental expenditure
on children and how it varies with number of children, using a Danish household data-set
(Bonke & Browning (2011)). The paper nds that parents increase their expenditure by
almost 41% with a second (equivalent) child and increase it by another 22% with a third
(equivalent) child. Using these estimates,
2
is set equal to 0.5 to re
ect the change in
expenditure by number of children.
The ability parameter in the human capital production function (HCPF) is nor-
malized to 1, in order to carry forward the assumption of homogeneity across agents of the
same cohort. For the baseline model, is set equal to 0.93 using the value estimated by
Manuelli & Seshadri (2009), of the degree of returns to scale on investment for a similar
human capital production function. Browning et al. (1999) estimate the value of to lie
between 0.5 and 1, based on dierent formulations of the human capital production function.
Sensitivity analysis is undertaken later in this section with alternative values of .
I follow the same approach as Choukhmane et al. (2013) to estimate the value of
the human capital productivity parameter, by using the estimates provided by the growth
accounting literature. Bernanke & Gurkaynak (2001), use cross-country panel data for the
period 1960-1996 to estimate the growth accounting equation from Mankiw, Romer & Weil
(1992), and nd that productivity of human capital is between 0.3 & 0.5 for dierent set of
economies. For China, I set equal to 0.5.
8. Estimates of average household expenditure on education from the `Study of Family Life in Urban
China, 1999' vary between 18-20 % for families living in urban areas with 1 school-going child, Choukhmane
et al. (2013) estimates average education expenditure by children's age using another dataset and nds
that it ranges from 5% for younger kids to 16% for young adults. Chi & Qian (2016) estimate the average
education expenditure to vary between 8 % - 19 %, depending on the income of the parents.
40
Finally, transfer share is estimated from the CHARLS data-set which is a survey on Chinese
households and carries information on the amount of transfers between dierent members
of the family, including those between parents and children (See chapter 2 for details).
Using samples from cross-sectional waves 2013 and 2015 it is found that gross transfers from
children to parents account for almost 4-4.8 % of the child's income and net transfers from
children to parents is around 2-2.4%. Hence, is set equal to 4%. This estimate is also in
line with the ndings from the previous literature (Curtis et al. (2015), Choukhmane et al.
(2013)). In the model variants section, later in the chapter, transfer share is endogenized
and estimated as a function of the dierence between the human capital stocks of the parent
and the child, since a higher dierence is seen to be associated with higher transfers (This
is documented in the second chapter).
Table 3.1 summarizes the values of parameters used in the simulation of the baseline
model.
Table 3.1: Parameter values for the basic model
Parameter Symbol Value
Coef. of relative risk aversion 1.5
Discount factor 0.99
Weight on children
1
0.18
Curvature on children
2
0.5
Transfer share 0.04
Interest rate r 0.04
Human capital productivity 0.5
HCPF coecient 1
HCPF curvature 0.93
3.4.2 Results
The aggregate household investment rate and savings rate for the period 1955-2009
9
for the
baseline model is shown in Figure 3.4.
9. 1955 being the year when the rst cohort completes its life cycle, that is, turns 85
41
.075 .08 .085 .09 .095 .1
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
(a)
0 .05 .1 .15 .2 .25
Aggregate savings rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Data
(b)
Figure 3.4: Household Investment and Saving rate in the baseline model
42
In the model, changes in the survival probabilities and the child-parent ratio are the drivers
of change in the aggregate investment and saving levels. From panel (a), it can be seen that
investments rst go up and then start falling roughly from the early 1990s. Note that the
investment rate is the household investment rate per child. Panel (b) graphs the household
savings rate obtained from the model and compares it to the actual observed rate between
1955 and 2009. The objective here is to document the contribution of the age-demography
in shaping the trajectory of the household savings over time. While demographic features
alone cannot fully explain the trajectory of savings rate, they do explain a substantial part
of the increase in the savings rate from the 1990s, as will be seen later.
We can trace some of the trends seen here to the trends in fertility and the old-age
dependency ratio. Beginning 1970s, n began to decline rapidly, which coincided with an
increase in the investment rate per child. This trend continued roughly till the end of the
1980s. In the 1990s, n continued to decline, accompanied by an acceleration in the old-age
dependency levels (with the aging of the baby-boomer population) after 20 year period of
no signicant change. The investment rate declined, and the savings rate rose beginning
1990s. The old-age dependency levels languished between 1970 and 1990, which coincides
with minimal changes in the savings rate for that period. A case can be made here of
investments and savings being strongly in
uenced by the old-age dependency ratio. Note
however, the above are just correlational trends between the investment rate, the savings
rate, and the age-demography.
From the above analysis we have gathered an idea of how investments in children
and savings evolved with demographic change over time, but in order to causally pin down
the role of declining mortality and fertility, there is a need to look at the trajectories of
these investments and savings under some alternative demographic scenarios. Since the
objective is to measure the impact of declining mortality in the old-age, I undertake a
43
counter-factual simulation exercise where the probability of survival is xed
10
. Specically,
survival probabilities are held constant at 1871 cohort level (this is the rst cohort in our
sample). Denote this counterfactual as the constant survival (constant
) model. Holding
mortality constant at the 1871 level for the entire sample period, implies that the economy
no longer experiences aging in the sense of increased chances of living longer post retirement,
as in the baseline model. Note that everything else is same as the baseline model, including
the child-parent ratio.
400 600 800 1000 1200 1400
total population (in mill.)
1870 1890 1910 1930 1950 1970 1990 2010 2020
Baseline
Constant survival
(a)
1 1.5 2 2.5 3 3.5
Child-parent ratio (n)
1870 1890 1910 1930 1950 1970 1990 2010 2020
(b)
.85 .9 .95
Average survival probability post retirement
1870 1890 1910 1930 1950 1970 1990 2010 2020
Baseline
Constant survival
(c)
0 .05 .1 .15 .2
Old-age dependency ratio
1870 1890 1910 1930 1950 1970 1990 2010 2020
Baseline
Constant survival
(d)
Figure 3.5: Demographic trends for the constant survival model
10. A counterfactual exercise involves modifying a particular feature of the model (in the present case, that
feature is the demographic structure (such as the mortality rate or the fertility rate) while keeping everything
else exactly the same as before.
44
Before presenting the simulation results, some demographic implications of the constant
survival model are discussed in order to facilitate a better understanding of the results.
Figure 3.5 gives an overview of the population in the counter-factual model, that is if the
survival probabilities are constant at the 1871 level and how it diers from the baseline.
Dashed lines depict the trends in the constant survival model. Since the probability of
survival is held constant, average survival is now a straight line (panel (c)). The change
can be most comprehensively seen in panel (d). Old-age dependency ratio is now lower and
ares out for the later periods. It appears to be almost constant between 1970 and 2010.
Figure 3.6 compares the investment rate and the savings rate in the constant survival
model to the ones in the baseline model. As can been seen in panel (a), aging (in the sense of
declining mortality) has led to lower investments in the human capital of children. In other
words, had the mortality rate not declined, investments in children would have been higher.
The gap has widened after 1990, also the peak (before the decline) is reached much later in
the constant survival model. Note thatn is still declining for this counterfactual, hence, the
shape of the investment rate plot in the two models can be attributed to changes in n (this
is discussed in detail later). Moving onto the savings rate, following trends can be clearly
discerned. Firstly, the savings rate curve is now much
atter with the increase post 1900s
also less pronounced. Hence, a good portion of the increase in the savings rate for the post
1990s period can be purely attributed to declining mortality. Secondly, savings are higher
between 1970 and 1990 for the constant survival model. Note that investment in children is
undertaken by the people in the working age group between ages 20-39, on the other hand,
all agents between ages 20-85 save. Hence, a simultaneous increase in the investment and
the savings rate is possible, given the rates of return on investing in kids and saving. It is
likely that the agents with dependent children choose to save less in the baseline in order
to invest more in children in the specic period mentioned. Also, a lower (and more or less
unchanging) dependency ratio between 1970-1990 in the constant survival model, could have
implied more freed up resources implying higher savings and investments in this period.
45
.07 .08 .09 .1 .11
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
(a)
0 .05 .1 .15 .2 .25
Aggregate savings rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
Data
(b)
Figure 3.6: Household investment and saving rate in the baseline and constant survival
model
46
.08 .1 .12 .14 .16
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
Baseline: per child rate Constant γ: per child rate
Baseline: total rate Constant γ: total rate
Figure 3.7: Total investment rate in the baseline and constant survival model
Up until this point, we have only looked at how per-child investment rate changes. In the
model, it is assumed that households spend equally on all kids in the family. Since households
lower expenditure across all children with aging, total expenditure is expected to be lower
for the aging economy compared to the constant survival one. It can be seen in Figure 3.7,
which plots the total investment rate for the baseline and the constant survival model. The
total investments are lower in the baseline model. Also note that the total investments peak
before the per-child investments, indicating the fall in fertility in the 1970s. In the post-2000
period, total investments coincide with the per-child investment as the maximum number of
children per household falls to 1.
In the constant survival model, changes in the investment and savings rate over time
can be solely attributed to changes in n. In the next counterfactual exercise, this channel
is also turned o. Here, neither the survival rate nor the child-parent ratio (n) changes;
both are held constant at the 1871 cohort level. This counterfactual is denoted as the
constant survival plus constant n model. This exercise helps in understanding the role of
overall demographic change in shaping savings and investments and how much do survival
probabilities contribute versus the child-parent ratio.
47
0 .05 .1 .15 .2
Old-age dependency ratio
1870 1890 1910 1930 1950 1970 1990 2010 2020
Baseline
Constant survival
Constant survival + constant n
Figure 3.8: Old-age dependency ratio for the constant survival plus constant n model
Figure 3.8 shows the old-age dependency ratio across the three models. Since,n is now xed
at the initial high level, old-age dependency ratio is initially higher in the constant survival
plus constant n model, as compared to the baseline, due to a smaller group of working-age
population. With time as more people enter the working age, dependency ratio plateaus.
Investment and savings rate for the three models together, is shown in Figure 3.9.
As expected, in the constant survival plus constant n model, curves for investment rate and
savings rates are
at lines. Moving from the third model to the second, investment curve
jumps up, and is also no longer
at. This change in the investment is due to the changing
fertility levels as observed in the data. Savings rate is also no longer constant, it is higher
and it is increasing post-1990s. This implies that lower fertility in the post 1990 period has
contributed towards an increase in savings. Finally, when we move from the second model to
the baseline, we witness a pervasive fall in the investment rate across the sample period. This
change is due to the mortality decline only. Savings on the other hand are much higher for
the post-1990 period. In this period, the substitution eect is stronger as individuals choose
to invest in risk-free assets and substitute away from investing in children. Notice that in
48
.07 .08 .09 .1 .11
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
Constant γ + constant n
(a)
0 .05 .1 .15 .2 .25
Aggregate savings rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
Constant γ + constant n
Data
(b)
Figure 3.9: Household investment and saving rate in the three demographic models
49
Figure 3.5, the gap between the dependency ratio in the two models widens substantially
after 1990.
These trends are summed up in Table 3.2. As seen in the previous gures, the
aggregate investment rate follows a hump-shape in the baseline as well as in the constant
survival model, with average investment rate peaking around 1990. Note that the dierence
in the average investment rate between the rst two models widens time. In the last sub-
period the average investment rate is lower in the baseline by full 2 percentage points.
Average levels of savings are higher for the latter two periods, with a substantial jump
(almost by higher 7 percentage points), post 2000. in the savings rate.
Table 3.2: Comparing household investment and saving rate across the three demographic
models
1955-1979 1980-1999 2000-2009
Aggregate investment rate (%)
Baseline 8.2 9.1 7.6
Constant
(at 1871 level) 8.7 9.8 9.6
Constant
, constant n (at 1871 level) 8.1 8.1 8.1
Aggregate saving rate (%)
Baseline 9.7 10.4 21.3
Constant
(at 1871 level) 10.7 10.2 14.6
Constant
, constant n (at 1871 level) 7 7 7
3.4.3 Model Variants
In this section I discuss two extensions or variations to the basic model economy discussed
above. The rst is the addition of a social security system, and the second is endogenizing
the transfer shares, that is the fraction of the income that children transfer to their retired
parents.
50
Introducing social security system.
Simulations of the basic model with alternative population scenarios reveal that aging has
led to lower investments in children and higher savings (for a given span of time). However,
in the basic model households have no external insurance to ensure nancial support in
the old-age. In the real world, the government provides social security benets to retired
individuals. Mu & Du (2015) nd that the expansion of the pension coverage in certain parts
of urban China led to an increase in the education expenditure by parents. This implies that
agents save less for their retirement with the provision of pensions, and hence are able to
spend more on their children. I analyze if this is the case here, by introducing the social
security system in the basic model.
The social security system works by making social security payments to the retired
workers and collecting taxes from the working individuals. After retirement all workers
receive a pension which is a xed fraction of their last wage before retirement. I set the
replacement rate (rr) at 15 %, following Imrohoroglu & Zhao (2018a). The tax rate is
calibrated to target the balanced budget for the baseline economy for the pre-economic
reform period in China, that is the period between 1955-1978 (See Curtis et al. (2015)).
Balanced budget implies that the tax revenue exactly pays for the social security transfers
to the retired workers. The tax rate (denoted by ) comes out to be 1.5%.
Hence, the budget constraints (3.2), (3.5) and (3.6), now include the social security taxes
in stages 1 & 2, and pensions in stage 3. The modied budget constraints are mentioned
below. Here, the pension p
c;j
=rr(w
c;63
), and denotes the social security tax rate.
c
c;j
+n
c
x
c;j
+a
c;j+1
= (1
c;j
)w
c;j
+ (1 +r)a
c;j
+b
c;j
; j2 [20; 39] (3:2a)
c
c;j
+a
c;j+1
= (1
c;j
)w
c;j
+ (1 +r)a
c;j
+b
c;j
; j2 [40; 63] (3:5a)
c
c;j
+a
c;j+1
=f
c;j
+p
c;j
+ (1 +r)a
c;j
+b
c;j
; j2 [64; 85] (3:6a)
51
.07 .08 .09 .1 .11
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
Baseline: Without SS.
Baseline: With SS.
Constant γ: With SS.
(a)
.05 .1 .15 .2 .25
Aggregate savings rate
1950 1960 1970 1980 1990 2000 2010
Baseline: Without SS.
Baseline: With SS.
Constant γ: With SS.
(b)
Figure 3.10: Household investment and saving rate with social security system
Figure 3.10 presents results for the baseline and the constant survival model for the economy
with the social security system. Note that the denitions of the two demographic models
are same as before. The results are also compared to the baseline with no social security
sector (shown in red). Firstly, aging result still holds; people tend to invest less in kids due
to declining mortality even after receiving pension in the old-age. Hence pensions are not
sucient to reverse the impact of aging on the investments. Secondly, comparing the two
52
baseline models; with and without the social security, we see a marked dierence in post 1990
period. The investments are much higher in the model with the social security sector
11
. This
conforms to the ndings in Mu & Du (2015). For the same period, savings in the baseline
model with no pensions or social security are higher. Hence, pensions provide a safety net
in the old age as agents choose to save less in the presence of a social security system.
Also, comparing the baseline and the constant survival model with the social security
sector, it can be seen that savings are pretty much similar throughout, up til after 2000,
when it starts to rise in the baseline. As before, this is the declining mortality eect. We no
longer have lower savings in the baseline for 1970-1990 period as before. I also compare the
results in the baseline and constant survival model to the constant survival plus constant
n model. The trends are same as before. For detailed graphs on the three counterfactual
population models with social security, please see Figure B.2 in the appendix.
Table 3.3 summarizes the investment and savings rate across the three models for the
basic economy and the economy with the social security sector. As noted before, introduction
of the social security has reduced the need to save and has led to higher investments in
children. This can be seen from the rst row of the table. Also note that the dierence
between the baseline and the constant survival model in the economy with social security
is not as big as with the basic economy. The dierence in the investment rate for the last
sub-period is only 1%, as compared to the 2% before. Similarly savings haven't shown a big
increase either. In sum, the inclusion of the social security sector has helped reduce some of
the precautionary behavior associated with aging by providing an external insurance for the
old-age.
11. Note here that for the period between 1970-1990 investment rate in the baseline is almost the same,
with and without the social security. Individuals would like to invest more in children in the absence of
social security during this period, hence they also save less. It is likely that they are restricted by the zero
borrowing limit
53
Table 3.3: Comparisons across models with and without the social security sector
Aggregate investment rate Aggregate savings rate
1955-1979 1980-1999 2000-2009 1955-1979 1980-1999 2000-2009
Baseline Without SS 8.2 9.1 7.6 9.7 10.4 21.3
With SS 8.5 9.5 8.8 9.4 8.4 14.6
Constant
Without SS 8.7 9.8 9.6 10.7 10.2 14.6
With SS 8.7 9.7 9.9 9.5 8.8 10.9
Constant
, Without SS 8.1 8.1 8.1 7 7 7
constant n With SS 8 8 8 6.6 6.6 6.6
Endogenizing transfers
Transfers from children to parents are xed at 4% of the child's income in the model de-
scribed above. The analysis in chapter 2 showed how education gap played an important
role in determining the transfers from children to parents. This is relevant for our current
analysis since parents with relatively more educated children (more human capital) are likely
to receive larger transfers. Given this setup, parents may also choose to invest more in their
children in expectation of higher transfers
12
. Using this insight, transfers are now endoge-
nized to re
ect the role played by the intergenerational gap in human capital accumulation.
They take the following functional form:
c;j
=
8
>
>
<
>
>
:
0
(
h
c
=
h
c20
)
g
; if parents are retired
0; otherwise
Here,
h
c
is the human capital of the child and
h
c20
is the human capital of the parent.
0
is the average transfer share in the population. It is set equal to 4%, as before.
g
is
the elasticity of transfers with respect to the human capital gap or the skill gap. I use the
estimates from the regression analysis in Chapter 2 to measure the elasticity parameter
g
.
12. while in the regression analysis conducted in chapter 2, I did not control for endogeneity between the
transfers and child's education level, in the model it is possible to do so, since parents optimize on investments
after taking into account all possible future returns
54
It comes out to be equal to 0.42
13
. The idea here is that if there is no inter-generational
dierence in the human capital stocks then the child transfers at the average rate (which
is 4%), however as the skill gap increases the child begins to transfer more (by the same
token, he transfers less in case the skill gap decreases). Therefore, transfers evolve in the
overlapping generations framework alongside human capital investments.
Figure 3.11 presents the results for three alternative demographic models with en-
dogenous transfers. Here, again we nd that declining mortality has led to a downward shift
in the investment rate in children. These results are also compared to the baseline model
with constant (shown in red). Interestingly, it can be seen that parents now invest more
in kids, since a higher level of human capital of children leads to a larger transfer share.
.07 .08 .09 .1 .11
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
Baseline: Constant τ
Baseline : Endogenous τ
Constant γ: Endogenous τ
Figure 3.11: Household investment rate with endogenous transfers
Table 3.4 compares average investment rate across models with the endogenous transfer
share and constant transfer share. With endogenous transfers, the magnitude of change
between the baseline model and the constant survival model is the same as before (that is
with constant transfers). Note that investment rates are higher everywhere with endogenous
transfers as compared to constant transfers. This result shows that parents value higher
13. See table 2.7 in section 2.5 for further details
55
transfers from children along with being altruistically motivated to invest in their kids. An
increase in transfer share leads to higher investments, even when the weight on child's human
capital in the parent's utility function is the same as before.
Table 3.4: Comparison of the aggregate investment rate across models with endogenous
transfer share
1955-1979 1980-1999 2000-2009
Baseline Constant 8.2 9.1 7.6
Endogenous 8.7 9.7 8.2
Constant
Constant 8.7 9.8 9.6
Endogenous 9.2 10.4 10.2
Constant
, constant n Constant 8.1 8.1 8.1
Endogenous 8.5 8.5 8.5
3.4.4 Sensitivity Analysis
In this section, I evaluate the model using alternate values of returns to scale on investment
in human capital (). In the literature there has been a lengthy discussion on the value of
. Browning et al. (1999) estimate the value using the Ben-Porath model of human capital
(Ben-Porath (1967)). They nd varying estimates ranging from 0.5 to 1. Choukhmane et
al. (2013) adopt a Cobb-Douglas CRS human capital production function with two inputs;
investments made by parents and the human capital of the parents. Their estimate of
returns to investment is 0.8. Huggett (2011) on the other hand, also using a Ben-Porath
human capital production function estimates the value of to be 0.7. I use these two values
in the sensitivity analysis.
Figure 3.12 presents the results for the baseline model with dierent values of . As
returns to investment in human capital diminish, so do the rates of investment. On the other
hand, with the savings rate there is no clear discernible trend. Between 1970-1990, savings
seem to be highest for = 0:7 which implies substitution between investments in children
56
and saving. But by the late 1990s, this trend reverses and savings rate is highest for the
highest value of = 0:93.
.04 .06 .08 .1
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
Baseline:φ = 0.93
Baseline:φ = 0.8
Baseline:φ = 0.7
(a)
.05 .1 .15 .2 .25
Aggregate savings rate
1950 1960 1970 1980 1990 2000 2010
Baseline:φ = 0.93
Baseline:φ = 0.8
Baseline:φ = 0.7
(b)
Figure 3.12: Aggregate investment and savings rate in the baseline model with dierent
values of
Table 3.5 summarizes the trends across dierent values of . Average investment rate is
highest for = 0:93 across all population models. Comparing dierent population models
for = 0:8 & 0:7, it can be seen that the aging result still holds, and the major trends are
similar to those with = 0:93, even with lower returns to scale on investments in the human
57
capital. Appendix gures B.3 and B.4 show the trends in investment and savings across the
three population models for the alternative values of . However, with smaller values of ,
the magnitude of upward shift in the investment rate is smaller between the baseline and the
constant survival model. For = 0:7, this shift diminishes further. This is not surprising
given that the value of governs as to how much can an individual gain from investing
in the human capital of their children. As the returns diminish, the marginal benet from
investing in children also fall in the constant survival model. Also note that for the constant
survival model, the saving rate pattern is not as ambiguous as the baseline. The saving rate
goes up and the investment rate falls as the value of decreases (second row of the table).
This pattern doesn't exist for the baseline model.
Table 3.5: Comparisons across models with dierent values of
Aggregate investment rate Aggregate savings rate
1955-1979 1980-1999 2000-2009 1955-1979 1980-1999 2000-2009
Baseline = 0:9 8.2 9.1 7.6 9.7 10.4 21.3
= 0:8 6.2 6.9 5.9 10.4 9.7 19.2
= 0:7 4.8 5.3 4.9 11.4 11.8 16.3
Constant
= 0:9 8.7 9.8 9.6 10.7 10.2 14.6
= 0:8 6.4 7.2 7 11 10.5 15
= 0:7 4.9 5.5 5.4 12.2 12.2 15.3
Constant
, = 0:9 8.1 8.1 8.1 7 7 7
constant n = 0:8 6.3 6.3 6.3 7 7 7
= 0:7 4.6 4.6 4.6 10 10 10
3.5 Conclusion & Discussion
This paper analyzes whether population aging can potentially impact human capital invest-
ments in children. It is seen that all else equal, age-demography plays a vital role in shaping
investments in children. In particular, an aging society spends less on children's human
capital development. Specically, the declining mortality rate has caused human capital
58
investments in children to shrink. For the period 1955-2009, the investment rate shrank
by almost 12 % on average (compared to constant mortality scenario). For the last decade
(2000-2009), the decline in the investment rate was 21 % while the savings rate increased by
almost 46 %. It is the period when the dierence in human capital investments due to aging
is most pronounced.
Dierent variants of the basic model are studied. The result of lower human capital
investments with aging holds even after the introduction of social security transfers in the
old age. Endogenizing the old-age support from children boosts investments in children's
human capital.
In the sensitivity analysis, dierent parameter values for the returns to investment
in human capital are used. With lower returns to investment in the human capital, the
magnitude of investments is also lower. However, the aging result still holds; mortality
decline leads to lower investments in the children, although the dierence in the investment
rate between the declining mortality and the constant mortality model is now smaller.
This paper has adopted a simple framework to analyze the role of age-demography
in shaping household behavior towards investment in children. However, several important
components that can potentially impact these investments and the savings and alter the
trajectories projected by the demographic change are missing. Some of the model's short-
comings are: rstly, households are assumed to be identical in terms of their initial abilities
or learning abilities. These dierences can lead to dierential outcomes within a cohort and
also across generations. Here, the impact of aging will dier with the ability. Secondly,
income is certain. Households face no risk of unemployment. Again this is an important
aspect since it has direct implications for savings for current as well as for future generations
(see Boar (2020)). Lastly, health care risks associated with the old-age adults are missing
from the analysis. Given the current setup, these risks are likely to exacerbate savings for the
old age and amplify the lowering of investments in children with aging. However, in reality,
59
children play a vital role in caregiving to old-age parents, especially those with medical prob-
lems. Modeling these characteristics would have helped analyze the problem of investing in
children more comprehensively. Incorporating these features can empower this analysis and
make it a more practical tool for policy analysis.
The foregoing analysis looks at the singular role of demographic change in shaping
human capital investments in children in China's context. The analysis is meant to be a
general one, using the simulations with Chinese data as an example to illustrate the role of
population aging. However, it is important to acknowledge that the results obtained from the
simulation exercise conducted above might be aected by other contextual factors related
to the Chinese society and economy. Firstly, estimates suggest that returns to education
in China substantially increased from being less than 5% in the early 1990s to more than
20% during 1997-2006 (Fang et al. (2012)). In so far that the human capital investments
imply investments in children's education, this increase in the returns is likely to play an
important role in parents' decision to invest in children. Secondly, educational reforms in
China, especially the compulsory education law promulgated in 1986, potentially impacted
the household investments in children in that period. Moreover, the dierence in the timing
and the intensity of enforcement across dierent regions might have also led to cross-regional
disparities in household-level investments
14
. Thirdly, rapid urbanization and rural-urban
migration in China led to a steep increase in housing prices, especially after the introduction
of the housing reforms and dismantling of the welfare housing system beginning mid-1990s
(Chen, Guo & Wu (2011)). Expensive housing may impact human capital investments
in children by inducing working households to save more for future housing needs. Note
that population aging itself is known to aect housing prices (Li & Shen (2013)). Aging
can, therefore, also indirectly impact the course of investments in children. Lastly, rising
migration, urbanization, and higher educational attainment have escalated the geographic
14. for further details on dierential exposure to the compulsory education law and its likely impacts please
see Ma (2017)
60
distances between children and parents. The proximity is a strong positive predictor of old-
age support from children (Bian et al. (1998)). Increased distance may change the parental
incentives associated with investing in the human capital of children (Fu (2019)).
61
Chapter 4
A cross-country analysis of population aging and
educational expenditure on children
Abstract
Population aging changes the patterns of household saving and spending. Results from the
model analysis in Chapter 3 reveals that the aging process has resulted in lower investments
in children's human capital. This paper undertakes a cross-country analysis of household
expenditure on education to explore its potential links with population aging in the data.
It uses cross-country variations in the age structure of the population in dierent countries
and across time. It nds that households of older economies tend to spend less on education
per child.
4.1 Introduction
In Chapter 3, we pin down the impact of aging on human capital investments in children.
The model analysis helps answer the question posed at the beginning about the nature of
change in the investments in children's human capital, due to declining mortality. However,
it is important to test whether the model has been able to generate a correct understanding
of the relationship between aging and potential investments in children's human capital. In
62
other words, do we see a systematic change in the household-level investments in human
capital in the aging societies in the real-world data?
The model in the previous chapter suggests that all else equal, the aggregate house-
hold human capital investment rate (fraction of household income spent on human capital
investments) is lower for an aging economy (relative to a non-aging economy). We learn
from the model that at any point in time if we were to compare two economies that are
otherwise identical except that one of them has younger demography compared to the other,
we will nd lower investments in human capital in the older economy. The plan is to use
the cross-country variations in the age composition of populations to assess if human capital
investments dier across time and space with changing age demography.
The caveat `all else equal' is critical. Isolating the impact of aging in the data is
less straightforward compared to the simulations from the previous chapter. In practice,
there are too many underlying forces that work together to determine the investments made
in children's human capital. These forces are also inter-related; countries with very low
mortality rates also have high levels of development, literacy, and human capital. This issue
is mitigated to a certain extent by including country-specic characteristics that are likely to
impact the investments made in the human capital of children for a given economy at a given
time. Results from this exercise should be interpreted as merely indicative of the relationship
under evaluation. At the same time, the model analysis is responsible for providing insights
as to why a change in the age structure would change the investments in the way that it
does. Data on household-level expenditure on education is used to measure human capital
investments in children. The analysis shows that the cross-country evidence supports the
model predictions; `age' of an economy is negatively associated with household expenditures
or investments in children's human capital.
Rest of the paper is organized as follows: Section 2 outlines the relevant literature.
Section 3 describes the data, its sources, and the cross-country comparisons of trends in
63
aging and educational expenditures. Section 4 species the regression exercise, and section
5 presents the results. The last section concludes.
4.2 Literature
Cross-country analysis, in general, is one of the most widely used methods in empirical
research of long-term macroeconomic trends. These are especially useful in testing out
theories related to macro-economic trends, such as the theory of convergence ((Mankiw,
Romer & Weil (1992), Barro (1991)), and the relationship between economic growth and
inequality (Barro (2000), Alesina & Perotti (1994)). It also has wide-ranging applications
for empirical analysis involving demographic changes, especially aging. Lyons, Grable &
Joo (2018) study how population aging is related to the nancial security parameters in
various countries. They nd that older countries are more likely to save and that nancial
inclusion and pension spending by government matters more for old-age security in these
countries. Li, Zhang & Zhang (2007) use a cross-country panel data-set to analyze the
impact of longevity and higher dependency rates on savings and investment. They nd that
cross-country dierences in the age structure of the population can explain the dierences
in aggregate savings rates.
Most of these works use a suite of aging parameters such as old-age dependency ratios,
life expectancy, fertility rates, and age composition of the population. Menz & Welsch (2012)
analyze life-cycle and cohort eects in order to study how population aging has impacted
carbon emissions in OECD countries. They show that age-composition of the population
as well as the birth cohorts matter; countries with older people and more people in specic
birth cohorts are emitting more per-capita.
While the cross-country aging literature has explored a wide range of topics includ-
ing the impact of aging on economic growth, nances, environment, previous literature on
64
a cross-country analysis of education spending is mostly focused on the relationship be-
tween educational spending and educational outcomes (Afonso & Aubyn (2006), Lee &
Barro (2001)). Some works study the relationship between population aging and education
spending; however, they mostly pertain to intergenerational con
ict and welfare spending
by the government. This strand of literature focuses on how the government chooses be-
tween addressing the needs of the young versus the old in society, and how these choices
and the preferences towards these choices change with population aging (Tosun, Williamson
& Yakovlev (2009), Gradstein & Kaganovich (2004) and Harris, Evans, & Schwab (2001)).
Sorensen (2013) studies the preferences for welfare spending in aging societies by undertak-
ing a cross-country analysis and nds that older people, not surprisingly, would prefer higher
spending on pensions and healthcare and less on education.
The present paper seeks to ll the gap in the literature by undertaking a study that
focuses on private or household educational spending instead of public spending and by
exploring how these are associated with population aging using a cross-country exercise.
4.3 Data description
Data on education nance is extracted from a publicly available cross-country panel data set,
which deals with various topics on education; UNESCO Institute of Statistics. It consists
of data categorized by educational expenditures undertaken by dierent entities such as
households and government and the expenditure undertaken at dierent education levels;
primary, secondary, and tertiary. This dataset also provides statistics on total years of free
schooling guaranteed by the law and the school-age population for all the countries in the
sample from this data source.
For the aging parameters, a set of age-demographic indicators are used, namely the
old-age dependency ratio, the median age in the population, life expectancy at age 60, total
65
Table 4.1: Description of variables
Variable Denition
HH expenditure on education Initial household expenditure on education per
student expressed as percentage of GDP per
capita
HH expenditure on primary education Initial household expenditure on education per
primary level student expressed as percentage of
GDP per capita
HH expenditure on secondary education Initial household expenditure on education per
secondary level student expressed as percentage
of GDP per capita
HH expenditure on tertiary education Initial household expenditure on education per
tertiary level student expressed as percentage of
GDP per capita
Government expenditure on education Initial Government expenditure on education
per student expressed as percentage of GDP per
capita
Govt expenditure on primary education Initial Government expenditure on education
per primary level student expressed as percent-
age of GDP per capita
Govt expenditure on secondary education Initial Government expenditure on education
per secondary level student expressed as percent-
age of GDP per capita
Govt expenditure on tertiary education Initial Government expenditure on education
per tertiary level student expressed as percent-
age of GDP per capita
Years of free guaranteed schooling Total no. of years of free schoolong guaranteed
by law (in years)
Old-age dependency ratio Number of people aged 65 years or above as a
proportion of number of people in working age
group (15-64) (in %)
Median age Median age of the population (years)
Life expectancy at age 60 Expected no. of years an individual of age 60
will live for
Total Fertility Rate (TFR) No. of live births per 1000 women in productive
age
Public expenditure on pension Total expenditure on pensions by government
expressed as a % of GDP
Private expenditure on pension Total expenditure on pensions by all private
sources expressed as a % of GDP
Long term interest rate Yield on 10 year government bonds (in %)
66
fertility rate and population shares for dierent age groups. These are obtained from the
UN World Population Prospects 2019. It is a comprehensive data set with population tables
dis-aggregated by age, and summary statistics on age structure, mortality, and fertility.
Other important determinants of expenditure on education are income, pension spending,
and long-term interest rates. Data on real GDP is obtained from Penn World Tables 9.0 and
the data on public and private expenditure on pensions, and yields on ten-year government
bonds are obtained from OECD.Stat for a given set of countries. For the exact denitions
and units for all the variables used in the study, please see Table 4.1.
Sample
The primary sample for the analysis constitutes of all the OECD countries for which the
data is available. There are several advantages of using the set of OECD countries as the
primary sample. Firstly data on household expenditure on education is available for a limited
set of countries and for fairly recent period of time: 1998-2018. Most OECD countries are
covered in this dataset. Secondly, data on pension spending and long-term interest rates
is more widely available for these countries. Lastly, OECD countries are seen as a more
cohesive group with similar characteristics which overcomes unobserved heterogeneity issues
to a certain extent and provide more reliable results. I also conduct the analysis using an
extended sample later. This extended sample (denoted as the OECDplus sample) includes
all the OECD countries and a set of selected non-OECD countries
1
The primary sample
consists of 35 OECD countries and the OECDplus sample consists of 51 countries, including
16 non-OECD countries. The list of all the countries and their relevant characteristics are
discussed below.
Table 4.2 summarizes key variables on educational expenditure and its determinants
for the primary and the extended sample. Household expenditure in OECD countries on an
1. Sample selection is done on the basis of the income level of the countries. All the available countries in
the data set are divided into three income groups (high, middle and low) using the data on real GDP, and
the countries that belong to the top two groups are chosen. The idea here is that the sample should consist
of countries that are similar enough for the comparisons to be reasonable.
67
Table 4.2: Descriptive Statistics: 1998-2018
Variable N Mean Std. Dev. Min Max
OECD countries
Total population (in mill.) 410 32 52 .28 320
Real GDP per capita
$
410 32531 12976 9936 83401
HH exp. on education per student:
Overall 410 3.1 2.5 .06 13
Primary 230 1.9 1.4 .01 7.2
Secondary 214 2.6 2 .025 12
Tertiary 207 7.4 7 .066 44
Govt. exp. on education per student:
Overall 410 24 6.2 7.9 42
Primary 314 20 4.3 10 36
Secondary 304 23 5.4 12 37
Tertiary 324 28 10 9.2 74
Yrs. of free schooling guaranteed 387 13 2.2 7 18
Pub. spending on pension 400 7.2 3.5 .77 17
Pvt. spending on pension 252 1.6 1.5 0 6
Yield (govt. 10 yr. bond, %) 370 4.4 2.3 -.36 22
Non-OECD countries*
Total population (in mill.) 133 84 191 2.1 1144
Real GDP per capita 133 14054 12259 1196 94572
HH exp. on education per student:
Overall 133 4.6 3 .046 11
Primary 87 3.5 1.7 .026 6.5
Secondary 80 5.6 3.1 .17 12
Tertiary 104 16 12 .9 64
Govt. exp. on education per student:
Overall 126 16 6.6 7.6 41
Primary 100 15 5.8 .24 36
Secondary 96 18 8.4 8.3 55
Tertiary 101 30 43 7.5 312
Yrs. of free schooling guaranteed 131 13 2.8 5 16
All countries
Total population (in mill.) 543 45 107 .28 1144
Real GDP per capita 543 28005 15064 1196 94572
HH exp. on education per student:
Overall 543 3.5 2.7 .046 13
Primary 314 2.3 1.6 .01 7.2
Secondary 292 3.4 2.7 .025 12
Tertiary 310 10 10 .066 64
Govt. exp. on education per student:
Overall 536 22 7.2 7.6 42
Primary 411 19 5.1 .26 36
Secondary 398 22 6.5 8.3 55
Tertiary 423 28 23 7.5 312
Yrs. of free schooling guaranteed 515 13 2.4 5 18
Refer to Table 1 for detailed denitions and units of the variables.
$
Real GDP per capita is in millions of constant 2011 USD.
*Non-OECD countries are the ones that are part of the extended sample: OECDplus
68
average is about 3.1% of the per-capita real GDP, highest expenditure being in the tertiary
education. Again there is great variation across countries. Comparing the OECD and non-
OECD sample, households in non-OECD countries spend more on an average on education,
the dierence being highest in the tertiary education. By the same token, government
expenditure on education is lower in the non-OECD sample (tertiary education being an
exception). Note that the non-OECD nations in the sample have larger population on
an average, and lower per-capita income. Data on pension spending and long term interest
rates is available only for OECD countries. Also, summary statistics for the key demographic
indicators are given in the appendix Table C.1. Non-OECD countries are relatively younger
with lower old-age dependency rates and lower life expectancy (at age 60), and higher fertility
rates. The average old-age dependency ratio for the OECD sample and for the non-OECD
countries diers by 8 percentage points. There is also greater variation within the set of
non-OECD countries. A dis-aggregated view of population statistics for dierent countries
and time periods is presented in detail in the next subsection.
4.3.1 Trends in aging and educational expenditure
Table 4.3 lists the aging indicators for all the sample countries for the years 2000 and 2015. It
is easily seen that there is a signicant variation in the age-demography both across time and
across countries. For the primary sample only, we can see much variation in the age structure,
with the median age ranging from 23 years (28 years) in Mexico to 41 years (47 years) in
Japan for the year 2000 (2015). Similar trends can be seen for the old-age dependency
ratio, with countries like Italy, Japan, Germany, and Sweden, having some of the highest
rates. The ratio has increased by at-least 15% between 2000 and 2015 for these countries. In
Japan alone, it went up to 43% in 2015 (an increase of almost 72% from 2000). The overall
OECD average for the median age and the dependency ratio has also gone up substantially
in fteen years. Comparing it to the non-OECD average, it is clear that that the non-OECD
69
Table 4.3: Aging trends across countries for the years 2000 and 2015
2000 2015
Country Old-age
dep. ratio
Median age Life exp. at
60
TFR Old-age
dep. ratio
Median age Life exp. at
60
TFR
OECD countries
Australia 18 35 24 1.8 22 38 26 1.9
Austria 23 38 22 1.4 28 43 24 1.5
Belgium 26 39 22 1.6 28 42 24 1.7
Canada 18 37 23 1.5 24 41 25 1.6
Chile 12 29 22 2.1 15 35 24 1.8
Czechia 20 37 20 1.1 27 42 22 1.6
Denmark 22 38 21 1.8 30 43 23 1.7
Estonia 22 38 19 1.3 29 42 22 1.6
Finland 22 39 22 1.7 32 43 25 1.7
France 25 38 24 1.8 30 42 26 1.9
Germany 24 40 22 1.4 32 46 24 1.5
Greece 24 38 22 1.3 32 44 25 1.3
Hungary 22 39 19 1.3 26 43 21 1.4
Iceland 18 33 23 2 21 37 25 1.9
Ireland 15 32 21 1.9 20 37 25 1.9
Israel 16 28 23 2.9 18 30 25 3.1
Italy 27 40 23 1.3 34 46 25 1.4
Japan 25 41 25 1.3 43 47 27 1.4
Latvia 22 38 19 1.2 30 43 21 1.6
Lithuania 21 36 19 1.3 28 44 21 1.6
Luxembourg 21 37 22 1.7 20 40 24 1.5
Mexico 8.5 23 22 2.7 10 28 21 2.2
Netherlands 20 38 22 1.7 27 43 24 1.7
New Zealand 18 34 23 1.9 22 38 25 2
Norway 24 37 23 1.8 25 39 25 1.7
Poland 18 35 20 1.3 23 41 22 1.4
Portugal 24 38 22 1.5 32 45 25 1.3
Korea 10 32 21 1.3 18 42 25 1.2
Slovakia 16 34 19 1.3 20 40 21 1.4
Slovenia 20 38 21 1.2 27 44 24 1.6
Spain 24 38 23 1.2 28 44 26 1.3
Sweden 27 39 23 1.6 31 41 25 1.9
Switzerland 23 39 24 1.4 27 43 26 1.5
Turkey 9.6 25 19 2.5 12 31 22 2.1
United Kingdom 24 38 22 1.7 28 40 24 1.8
United States of America 19 35 22 2 22 38 24 1.8
Total 20 36 22 1.6 25 40 24 1.7
Non-OECD countries
Argentina 16 28 20 2.5 17 31 21 2.3
Azerbaijan 9.3 26 17 2 8 31 19 2.1
Belarus 20 37 16 1.2 21 40 20 1.7
Bulgaria 24 40 18 1.2 30 44 20 1.5
Colombia 8.4 24 21 2.6 11 30 23 1.9
Ethiopia 6.2 17 16 6.5 6.4 19 18 4.6
India 7.2 23 17 3.3 8.5 28 18 2.3
Indonesia 7.3 24 16 2.5 8 29 18 2.4
Kazakhstan 10 28 16 1.9 10 30 19 2.7
Kuwait 2.3 28 17 2.8 2.8 35 18 2.1
Morocco 8.6 23 18 2.8 9.7 29 21 2.5
Peru 8 23 20 2.8 11 29 22 2.3
Romania 20 35 18 1.3 25 42 20 1.6
Russia 18 36 16 1.2 19 39 20 1.8
Ukraine 20 38 17 1.1 23 41 18 1.5
Viet Nam 10 24 21 2 9.5 31 22 2
Total 12 28 18 2.4 14 33 20 2.2
Total (OECDplus) 18 34 20 1.9 22 38 23 1.8
70
countries have much younger demography. The median age of countries like India and Peru
is comparable to that of Mexico. These countries also have a higher fertility rate as compared
to OECD countries in general. We see a substantial rise in the old-age dependency ratio
and the median age for the non-OECD sample as well by 2015. Life expectancy at age 60
has gone up by almost 9% in the OECD sample. There is a small increase in fertility rates
for OECD countries. In contrast, the average fertility rate in the non-OECD countries has
fallen by about 9% between 2000 and 2015. India registered a decline of 30% in these 15
years. Also, note that there has been a more signicant improvement in the life expectancy
for the non-OECD countries. The overall conclusion gathered from these trends is that the
world is aging, albeit in varying degrees, with most OECD countries at a later stage in the
aging process. The relatively younger economies with a smaller proportion of the elderly
population currently are also steadily moving towards lower fertility and lower mortality
stage as time passes by.
Table 4.4 shows educational expenditures and population shares for the elderly and
school-going children for a selected group of countries for years 2000 and 2015. We can infer
the following from the data. Firstly, household expenditure on education has gone up for
most countries between 2000 and 2015, in Spain the level of increase is almost 175%. A
notable exception to this rule is Chile. Secondly, government expenditures have also gone
up for almost all countries with the increase being quiet substantial in several cases. Note
that in most cases higher educational expenditures by government are associated with lower
household expenditure. Thirdly, with respect to the age-demography there is quiet some
variation in household expenditures; amongst countries with an elderly population of 20
% and above, expenditures vary between 0.32 % and 4.1 % in the year 2000. Similarly,
countries with larger school-age population vary in their educational spending. Last but not
the least, older countries are associated with lower household expenditure on education on
an average. In 2015, households in Italy, Greece and Finland spent an average of 2% on
education, while Mexico, Peru and Columbia spent 6.4 %, the corresponding gures for 2000
71
Table 4.4: Trends in educational expenditures and age-demography for selected countries
2000 2015
Country Pop.60
& abv.
School-
age
pop.
HH
edu.
exp.
Govt.
edu.
exp.
Pop.60
& abv.
School-
age
pop.
HH
edu.
exp.
Govt.
edu.
exp.
Australia 16 25 2.9 20 20 23 5.4 23
Canada 17 23 5 24 22 20 8 22
Chile 11 30 9.6 12 15 25 4.9 20
Columbia 7.6 32 6.4 11 11 27 6.9 16
Czechia 18 25 1.3 14 25 18 1.5 33
Finland 20 21 .32 27 27 19 .48 37
France 21 22 2.5 22 25 21 3.1 22
Greece 22 21 1 15 27 16 1.9 21
Italy 24 19 1.3 22 28 17 3.8 23
Latvia 21 25 4.1 21 26 17 3.4 31
Mexico 7.3 35 1.7 11 10 30 4 17
Netherlands 18 20 .26 23 24 20 .11 26
Norway 19 22 .28 29 22 22 .42 34
Peru 7.2 36 6.8 9 11 29 8.2 14
Spain 21 21 2.4 20 24 17 6.6 25
Sweden 22 21 .64 33 25 20 .92 38
Turkey 9 32 .11 7.9 12 28 3.5 13
United States of America 16 24 6.2 18 21 22 8.8 18
72
are 0.9 % and 2 % respectively. In order to explore the last point, correlations between the
household educational expenditure and aging indicators are looked at.
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AUT AUT AUT AUT
AZE AZE AZE AZE AZE AZE AZE AZE AZE
BEL BEL BEL BEL BEL BEL BEL BEL BEL BEL
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
BLR BLR BLR BLR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHE CHE CHE
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DEU DEU DEU DEU DEU DEU DEU DEU
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
FIN FIN FIN FIN FIN FIN FIN FIN FIN FIN
FRA FRA FRA FRA FRA FRA FRA FRA FRA FRA
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN
IND IND IND IND
IRL IRL IRL IRL IRL IRL IRL IRL IRL IRL
ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR
ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KOR KOR KOR KOR KOR KOR KOR KOR KOR
KWT KWT KWT
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU
LUX LUX LUX LUX
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD
NOR NOR NOR NOR NOR NOR NOR NOR NOR
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
ROU ROU ROU ROU
RUS RUS RUS RUS RUS RUS RUS RUS RUS
SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE
TUR TUR TUR TUR TUR TUR TUR
UKR UKR UKR UKR UKR UKR
USA USA USA USA USA USA USA USA USA USA
VNM VNM VNM VNM VNM
0 5 10
HH exp. on education
0 10 20 30
Old-age dependency ratio
Overall
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AUT AUT AUT AUT
BEL BEL BEL BEL BEL BEL BEL BEL BEL BEL
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
FIN FIN FIN FIN FIN FIN FIN FIN FIN FIN
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC
HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN IND IND IND IND
IRL IRL IRL IRL IRL IRL IRL IRL IRL IRL
ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KWT KWT KWT
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LUX LUX LUX LUX
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE
UKR UKR UKR UKR UKR UKR
USA USA USA USA USA USA USA USA USA USA
VNM VNM VNM VNM VNM
0 2 4 6
HH exp. on education
0 10 20 30
Old-age dependency ratio
Primary
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AUT AUT AUT AUT
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
BLR BLR BLR BLR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHE CHE CHE
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC
HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN
IND IND IND IND
ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR
ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KWT KWT KWT
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LUX LUX LUX LUX
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE
UKR UKR UKR UKR UKR UKR
0 2 4 6 8 10
HH exp. on education
0 10 20 30
Old-age dependency ratio
Secondary
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AZE AZE AZE AZE AZE AZE AZE AZE AZE
BEL BEL BEL BEL BEL BEL BEL BEL BEL BEL
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
BLR BLR BLR BLR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHE CHE CHE
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC
HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN
IND IND IND IND
IRL IRL IRL IRL IRL IRL IRL IRL IRL IRL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR
ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KOR KOR KOR KOR KOR KOR KOR KOR KOR
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
ROU ROU ROU ROU RUS RUS RUS RUS RUS RUS RUS RUS RUS SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
TUR TUR TUR TUR TUR TUR TUR
UKR UKR UKR UKR UKR UKR
VNM VNM VNM VNM VNM
0 10 20 30 40 50
HH exp. on education
0 10 20 30
Old-age dependency ratio
Tertiary
Figure 4.1: Household expenditure on education and old-age dependency ratio
Figures 4.1 to 4.3, graph the relationship between aging and household expenditure on
education for all countries in the OECDplus sample, using the average values for the sample
period. Looking at gure 4.1, we see a clear negative relation between average per-child
educational expenditure and old-dependency levels. This is true across all education levels
with varying degrees of intensity. Younger countries such as Mexico, Peru, Indonesia etc.
can be seen to lie in the upper left quadrant, whereas the older countries such as Italy,
Austria, Sweden etc. are found in the lower right quadrant. Another interesting aspect is
the convexity. The graphs indicate a u-shaped relation between old-age dependency and
educational expenditure. Similar trends can be seen with respect to the median age in gure
73
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AUT AUT AUT AUT
AZE AZE AZE AZE AZE AZE AZE AZE AZE
BEL BEL BEL BEL BEL BEL BEL BEL BEL BEL
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
BLR BLR BLR BLR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHE CHE CHE
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DEU DEU DEU DEU DEU DEU DEU DEU
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
FIN FIN FIN FIN FIN FIN FIN FIN FIN FIN
FRA FRA FRA FRA FRA FRA FRA FRA FRA FRA
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN
IND IND IND IND
IRL IRL IRL IRL IRL IRL IRL IRL IRL IRL
ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR
ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KOR KOR KOR KOR KOR KOR KOR KOR KOR
KWT KWT KWT
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU
LUX LUX LUX LUX
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD
NOR NOR NOR NOR NOR NOR NOR NOR NOR
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
ROU ROU ROU ROU
RUS RUS RUS RUS RUS RUS RUS RUS RUS
SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE
TUR TUR TUR TUR TUR TUR TUR
UKR UKR UKR UKR UKR UKR
USA USA USA USA USA USA USA USA USA USA
VNM VNM VNM VNM VNM
0 5 10
HH exp. on education
10 20 30 40 50
Median age
Overall
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AUT AUT AUT AUT
BEL BEL BEL BEL BEL BEL BEL BEL BEL BEL
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
FIN FIN FIN FIN FIN FIN FIN FIN FIN FIN
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC
HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN IND IND IND IND
IRL IRL IRL IRL IRL IRL IRL IRL IRL IRL
ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KWT KWT KWT
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LUX LUX LUX LUX
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE
UKR UKR UKR UKR UKR UKR
USA USA USA USA USA USA USA USA USA USA
VNM VNM VNM VNM VNM
0 2 4 6
HH exp. on education
10 20 30 40 50
Median age
Primary
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AUT AUT AUT AUT
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
BLR BLR BLR BLR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHE CHE CHE
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC
HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN
IND IND IND IND
ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR
ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KWT KWT KWT
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LUX LUX LUX LUX
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE
UKR UKR UKR UKR UKR UKR
0 2 4 6 8 10
HH exp. on education
10 20 30 40 50
Median age
Secondary
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AZE AZE AZE AZE AZE AZE AZE AZE AZE
BEL BEL BEL BEL BEL BEL BEL BEL BEL BEL
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
BLR BLR BLR BLR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHE CHE CHE
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC
HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN
IND IND IND IND
IRL IRL IRL IRL IRL IRL IRL IRL IRL IRL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR
ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KOR KOR KOR KOR KOR KOR KOR KOR KOR
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
ROU ROU ROU ROU RUS RUS RUS RUS RUS RUS RUS RUS RUS SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
TUR TUR TUR TUR TUR TUR TUR
UKR UKR UKR UKR UKR UKR
VNM VNM VNM VNM VNM
0 10 20 30 40 50
HH exp. on education
10 20 30 40 50
Median age
Tertiary
Figure 4.2: Household expenditure on education and median age
74
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AUT AUT AUT AUT
AZE AZE AZE AZE AZE AZE AZE AZE AZE
BEL BEL BEL BEL BEL BEL BEL BEL BEL BEL
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
BLR BLR BLR BLR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHE CHE CHE
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DEU DEU DEU DEU DEU DEU DEU DEU
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
FIN FIN FIN FIN FIN FIN FIN FIN FIN FIN
FRA FRA FRA FRA FRA FRA FRA FRA FRA FRA
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN
IND IND IND IND
IRL IRL IRL IRL IRL IRL IRL IRL IRL IRL
ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR
ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KOR KOR KOR KOR KOR KOR KOR KOR KOR
KWT KWT KWT
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU
LUX LUX LUX LUX
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD
NOR NOR NOR NOR NOR NOR NOR NOR NOR
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
ROU ROU ROU ROU
RUS RUS RUS RUS RUS RUS RUS RUS RUS
SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE
TUR TUR TUR TUR TUR TUR TUR
UKR UKR UKR UKR UKR UKR
USA USA USA USA USA USA USA USA USA USA
VNM VNM VNM VNM VNM
0 5 10
HH exp. on education
16 18 20 22 24 26
Life expectancy at age 60
Overall
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AUT AUT AUT AUT
BEL BEL BEL BEL BEL BEL BEL BEL BEL BEL
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
FIN FIN FIN FIN FIN FIN FIN FIN FIN FIN
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC
HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN IND IND IND IND
IRL IRL IRL IRL IRL IRL IRL IRL IRL IRL
ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KWT KWT KWT
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LUX LUX LUX LUX
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE
UKR UKR UKR UKR UKR UKR
USA USA USA USA USA USA USA USA USA USA
VNM VNM VNM VNM VNM
0 2 4 6
HH exp. on education
16 18 20 22 24 26
Life expectancy at age 60
Primary
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AUT AUT AUT AUT
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
BLR BLR BLR BLR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHE CHE CHE
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC
HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN
IND IND IND IND
ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL ISL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR
ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KWT KWT KWT
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LUX LUX LUX LUX
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD NLD
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE SWE
UKR UKR UKR UKR UKR UKR
0 2 4 6 8 10
HH exp. on education
16 18 20 22 24 26
Life expectancy at age 60
Secondary
ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG ARG
AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS AUS
AZE AZE AZE AZE AZE AZE AZE AZE AZE
BEL BEL BEL BEL BEL BEL BEL BEL BEL BEL
BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR BGR
BLR BLR BLR BLR
CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN CAN
CHE CHE CHE
CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL CHL
COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL COL
CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE CZE
DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK DNK
ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP ESP
EST EST EST EST EST EST EST EST EST EST
ETH ETH
GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR GBR
GRC GRC GRC GRC GRC GRC GRC GRC GRC GRC
HUN HUN HUN HUN HUN HUN
IDN IDN IDN IDN IDN IDN IDN IDN
IND IND IND IND
IRL IRL IRL IRL IRL IRL IRL IRL IRL IRL
ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR ISR
ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA ITA
KAZ KAZ KAZ KAZ KAZ KAZ KAZ
KOR KOR KOR KOR KOR KOR KOR KOR KOR
LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU LTU
LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A LV A
MAR MAR MAR MAR MAR MAR MAR MAR
MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX MEX
NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL NZL
PER PER PER PER PER PER PER PER PER PER PER PER PER PER PER
POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL POL
PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT PRT
ROU ROU ROU ROU RUS RUS RUS RUS RUS RUS RUS RUS RUS SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK SVK
SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN SVN
TUR TUR TUR TUR TUR TUR TUR
UKR UKR UKR UKR UKR UKR
VNM VNM VNM VNM VNM
0 10 20 30 40 50
HH exp. on education
16 18 20 22 24 26
Life expectancy at age 60
Tertiary
Figure 4.3: Household expenditure on education and life expectancy at age 60
75
4.2. The negative correlation is most discernible for the secondary level. In gure 4.3 average
household expenditures are plotted against life expectancy at age 60. Life expectancy at age
60 is only weakly negatively correlated to the household expenditures on education, and
not at all for the overall level. Table C.2 in the appendix presents correlation coecients
for all the aging indicators. High negative correlation can be seen with respect to old-age
dependency ratio, and population shares in the older age-groups.
4.4 Empirical Specication
It can be clearly discerned from the data patterns in the previous section that there is a
negative correlation between older economies and household-level expenditure on education.
I use the following regression specication to analyze this relationship.
hhexp
c;t;d
=
0
+
1
govexp
c;t;d
+
2
A
c;t;d
+
3
POP
c;t;d
+
4
PCGDP
c;t;d
+
d
+
t
+C
c;t;d
+
c;t;d
(4.1)
Subscripts c & t denote country and year respectively. Subscript d denotes the develop-
ment group that a country belongs to
2
. hhexp & govexp denote per student expenditure
on education by the households and the government respectively. These are expressed as
percentage of per-capita real GDP. POP stands for total population and PCGDP denotes
per-capita real GDP. A stands for the `aging' indicator: median age, old-age dependency
ratio, life expectancy at age 60, and age shares in the total population.
d
&
t
stand for
development group xed eect and time xed eects respectively. Finally, controls (C) in-
clude total fertility rate, yield on 10 year government bonds, public and private spending on
pensions, and total years of free schooling guaranteed by the law. The controls vary by the
sample, details are given in the results section.
2. Countries are classied into development groups on the basis of the per-capita real GDP in year 2012
76
The primary objective is to study the relationship between the dependent variable, per-
student household expenditure on education, and the aging variables listed above. We expect
2
< 0, given the results in Chapter 3. Higher government expenditure on education implies
a lower need for households to spend on education, irrespective of the age structure. Also, a
larger population will imply a larger work-force, everything else equal. Hence the expected
signs are;
1
< 0;
3
> 0.
Other controls, such as pension spending and yields on government bonds, indicate the
safety net of individuals in their old age. Better safety nets will reduce the opportunity costs
associated with spending on children. Hence controlling for these ensure that the impact of
aging is not wrongly estimated. Year xed eects take care of all time-specic variability
across countries other than the demographic changes. Slotting the countries in dierent
development groups helps further insure against the biases from unobserved heterogeneity.
Given the type of the sample under study (small T unbalanced panel dataset), random eects
panel regression model is adopted to estimate equation (4.1).
4.5 Results
4.5.1 OECD Sample
We begin with a discussion of results for the OECD sample. The results are presented sepa-
rately for the expenditure on the three education levels (primary, secondary and tertiary), as
well as for the overall educational expenditure. Note that, depending on the education level
being analyzed, the dependent variable; household expenditure on education per child is ex-
penditure on either overall education, or the primary education, or the secondary education
or tertiary education. This is also true for the government expenditure on education. Apart
from the primary regressors mentioned in equation (4.1), we also control for total years of
free schooling guaranteed by law, total fertility rate, pension spending (public and private)
77
and long-term interest rate. Year xed eects and development group xed eects are also
included. The above information applies to all results presented for the OECD sample.
Table 4.5: Results with Old-age dependency ratio
Overall Primary Secondary Tertiary
Old-age dep. ratio -1.778
-1.166
-1.522
-4.626
(0.423) (0.303) (0.249) (0.941)
Old-age dep. ratio sq. 0.0362
0.0231
0.0336
0.105
(0.00870) (0.00609) (0.00489) (0.0205)
Population 0.0192
0.00170 0.0181 0.0740
(0.00360) (0.00995) (0.0117) (0.0236)
Per-capita real GDP -0.0502
0.0281 0.0676
-0.335
+
(0.0236) (0.0376) (0.0318) (0.188)
Govt. exp. on education -0.0135 0.0849 -0.0866 -0.0397
(0.0508) (0.0559) (0.0543) (0.0785)
N 236 123 120 111
Standard errors in parentheses.
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
Table 4.5 presents results for equation (4.1), with the old-age dependency ratio as the ex-
planatory aging variable. The four columns present results for the household expenditure
at four dierent educational levels as mentioned in the rst row of the table. It is evident
that higher old-age dependency ratio is associated with smaller household expenditure on
education per child. For the overall education level, an increase in the old-age ratio by 1 % is
associated with a decline of almost 1.8% in the household spending, all else equal. This gure
is substantial given the statistics in Table 4.3, which show that the old-age dependency rate
went up by 5% (on an average), in 15 years. The coecient for tertiary level is quiet high.
Intuitively this makes sense as households might perceive tertiary education spending to be
a relatively non-essential spending and hence income elasticity is high. Positive and signif-
icant coecient on squared dependency ratio points to the U-shaped relationship between
the expenditures and the old-age dependency ratio. This validates the trends observed in
78
gure 4.1. As expected, we nd a negative correlation with government expenditure, since
economies that have better public education infrastructure are likely to have lower household-
level spending on education. We nd similar trend with per-capita income. Population is
positively related to spending, as expected. This could be from the scale eect of a larger
working-age population.
Table 4.6: Results with Median age
Overall Primary Secondary Tertiary
Median Age -0.163 -0.149
0.132 0.287
(0.123) (0.0716) (0.140) (0.340)
Population 0.0208
0.00874 0.0335
0.137
(0.00483) (0.00632) (0.00949) (0.0312)
Per-capita real GDP -0.0883
0.00153 0.0329 -0.443
(0.0229) (0.0236) (0.0325) (0.144)
Gov. exp. on education -0.0610 0.0343 -0.173
-0.150
(0.0585) (0.0278) (0.0420) (0.0563)
N 236 123 120 111
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
Table 4.6 shows results using median age as the explanatory aging variable. Results with
median age are weak and insignicant. We only nd a negative and signicant coecient
for educational expenditure at the primary level. Coecient for the total population here
is positive and highly signicant. Note also, the negative and signicant coecient on per-
capita GDP in columns (1) and (4). Table C.2 shows that correlation coecient between the
median age and the educational expenditure is negative and moderately strong in terms of
the magnitude. It appears that, after controlling for country-specic characteristics as well
as the incomes the correlation seems to break down.
79
Menz & Welsch (2012) have discussed the importance of life-cycle eects for observed vari-
ations in the macroeconomic data across countries. I incorporate the life-cycle eects in
Table 4.7, by using the shares of population in various age groups as the explanatory vari-
able. Population share in ages below 15 serves as the benchmark group.
Table 4.7: Results with Age eects
Overall Primary Secondary Tertiary
Sample: OECD
Age 75+ -0.149 -0.103 -0.114 -0.554
(0.361) (0.312) (0.300) (0.934)
Age 50-74 -0.586
-0.380 -0.366 -1.823
(0.161) (0.270) (0.344) (0.797)
Age 15-49 -0.550
-0.259 -0.700
-2.830
(0.276) (0.342) (0.335) (1.092)
Age 15- - - - -
Population 0.0221
0.000510 0.0322
0.110
(0.00400) (0.0102) (0.0107) (0.0200)
Per-capita real GDP -0.102
-0.0212 -0.00644 -0.502
(0.0212) (0.0369) (0.0443) (0.215)
Gov. exp. on education -0.0373 0.0521 -0.151
-0.197
(0.0523) (0.0699) (0.0757) (0.0985)
N 236 123 120 111
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
Not surprisingly, the coecients for the three age groups (that is the population share in
ages 15-49, ages 50-75, and ages 75 and above) are negative everywhere. In other words,
economies with relatively smaller school-age going population spend less on education per
student. The coecient for middle to old-age group (ages 50-74) is highly signicant for
columns (1) and (4). Most of the impact is coming from cutbacks in higher education
80
spending; that is at the secondary and the tertiary-level. The negative coecient for the
population in the age group 50-74, also establishes how a higher proportion of not only
old-age but also middle to old age individuals in the population can have a negative impact
on household expenditure on education. The coecient on total population is positive and
signicant, except for the primary level. As before, the government expenditure and the
per-capita income are negatively related to the household expenditure.
In sum, it can be seen that for the OECD countries, there is a systematic dierence
between older and younger economies with regard to the household spending on education
per student. Older economies have been found to be associated with lower expenditures,
controlling for other economy-specic factors when age-composition or old-age dependency
rates are used as explanatory variables.
4.5.2 OECDplus Sample
We now examine the results for the extended sample which includes selected non-OECD
countries along with the OECD countries. Some of the advantages of the extended sample
are: higher number of total observations, and increased variation with respect to the age
structure, total population and expenditures on education. Due to unavailability of that
data on certain control variables for the non-OECD countries, the regression exercise for the
extended sample only includes years of free schooling and the total fertility rate as controls
along with year xed eects and development group xed eects.
Table 4.8 presents results with old-age dependency ratio as the explanatory variable,
for the extended sample. We nd the same trend as with the primary sample; older countries
have lower household spending on education per child. As before the highest change can
be seen with tertiary education. A point of dierence between the OECD sample and the
extended sample is that the strength of the relationship between old-age dependency and
educational expenditure has diminished in the latter case. The coecient on the dependency
81
Table 4.8: Results with Old-age dependency ratio for the extended sample
Overall Primary Secondary Tertiary
Old-age dep. ratio -0.432 -0.499
-0.932
-2.366
+
(0.292) (0.186) (0.285) (1.371)
Old-age dep. ratio sq. 0.00708 0.00802
+
0.0210
0.0602
(0.00674) (0.00451) (0.00728) (0.0301)
Population 0.0137
0.00183 0.0128
+
-0.00575
(0.00481) (0.00207) (0.00762) (0.0175)
Per-capita real GDP -0.0433
+
-0.0470
-0.0801
-0.288
+
(0.0243) (0.0158) (0.0241) (0.164)
Govt. exp. on education 0.00584 0.0149 -0.0188 0.000882
(0.0525) (0.0307) (0.0386) (0.0360)
N 509 292 277 289
Standard errors in parentheses.
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
ratio for the overall education expenditure is no longer signicant. The coecient on income
per-capita is universally signicant and negative. The relationship between the household-
level spending and the government level spending on education is less clear as compared to
OECD sample.
We also explore the results with median age and the life-cycle eects for the extended
sample given in the appendix tables C.3 and C.4 respectively. The results are qualitatively
similar to the primary sample results. For the median age, the sign of the coecients for
dierent education levels is ambiguous like before. Generally, it is found that median age
has a negative coecient for the overall, primary and secondary education levels, although
it is only signicant for the primary education. Tertiary education on the other hand, shows
a signicant and positive relationship, which is in contrast to what is expected. For the life
cycle eects, we nd that, overall higher proportion of older-age population is associated
with lower per-child expenditure on education. For the extended sample, the coecients on
the age groups 15-49 and 50-74 are highly signicant for the secondary level.
82
Table 4.9: Results with life expectancy at age 60 for the extended sample
Primary Secondary Tertiary Primary Secondary Tertiary
Life expectancy at 60 0.06 0.32 0.747 -0.625
-1.134
-2.491
(0.146) (0.208) (0.815) (0.230) (0.556) (1.844)
Gov. exp. on education -0.0348 -0.0414 -0.011 0.0192 -0.00888 -0.0311
(0.0321) (0.0524) (0.0443) (0.0264) (0.0325) (0.0470)
Population 0.00585
0.0228
-0.007 0.00153 0.00479 -0.00802
(0.0023) (0.00724) (0.0195) (0.00277) (0.00620) (0.0248)
Real GDP per capita -0.0311 -0.048 -0.452
-0.0433
-0.0873
-0.470
(0.0272) (0.0384) (0.201) (0.0160) (0.0218) (0.205)
Age 60+ -1.220
-2.133
-2.738
(0.338) (0.608) (1.962)
Life exp*Age 60+ 0.0468
0.0910
0.153
+
(0.0148) (0.0298) (0.0871)
N 292 277 289 292 277 289
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
In the nal analysis, the eect of life expectancy (at age 60) on the educational expenditure
is analyzed, at the primary, secondary, and tertiary levels, respectively. The rst three
columns of Table 4.9 present the results with life expectancy at age 60, as the explanatory
aging variable. The coecients turn out to be insignicant. The coecients on all other
regressors are of the expected sign. These results indicate that life expectancy in itself fails
to have a signicant impact on household expenditures. However, this regression is missing
information on the age composition of the population of a given economy. It is reasonable
to assume that life expectancy at age 60 is more relevant to countries that are experiencing
aging. In particular, if there is a higher proportion of the population that is 60 years or older,
the life expectancy in the later period is likely to matter more. The last three columns add
the proportion of population aged 60 and above, as another explanatory variable, along with
an interaction term. The coecient on life expectancy is now of the expected sign. Similarly,
the proportion of people above 60 also carries a negative impact on educational expenditure.
83
Overall, given the proportion of people who are 60 years and above, higher life expectancy
(at age 60) negatively impacts the household expenditures on education. These results are
statistically signicant (except for the tertiary level) and are in line with previous results
that used other aging measures. Note that the interaction term between the life expectancy
and population share of age group 60 and above is positive and signicant. It implies that
the growing proportion of the old-age population would lower the negative impact of higher
life expectancy at age 60.
4.6 Conclusion
This paper supplements the analysis of Chapter 3 by providing empirical evidence of how
aging could lower investments in education or human capital of children. Analysis using
a cross-country panel dataset of 51 countries, nds that prominent aging parameters are
negatively associated with per-child spending on education by the households. Resources
allocated to children's education are lower at the intensive margin for countries with a higher
proportion of old-age adults. This result, while not implying causality, rearms the ndings
of the model, which showed that the human capital investments in children would have been
higher had the old-age mortality not declined.
The analysis is conducted initially for a smaller sample of the OECD countries only
and later for an extended sample that included selected non-OECD nations as well. The
overall result of a negative relationship between aging and educational investments stands
for dierent aging indicators, but the magnitude of the coecients vary across samples
and education levels. Mostly, the old-age dependency ratio is found to have a signicant and
stable negative coecient. Age-eects are also found to be signicant. Lastly, life expectancy
at age 60 is also found to be signicant and negatively related, given the population over
age 60.
84
The primary motivation behind using the cross-country data in this paper is to make use of
the cross-sectional variation in the age structure of dierent countries at the dierent stages
of the demographic transition. The results are encouraging to the extent that the analysis
can control for other country-specic variations. However, the cross-country analyses must
be looked at with care. Attempts to nd patterns between dierent variables at the cross-
country level is fraught with problems since comparing dierent economies is a humongous
task given the size of the unobserved variations. With this in mind, cross country analyses
are also useful in showcasing prominent patterns in the data that seem to hold irrespective
of the heterogeneity.
85
Chapter 5
Conclusion
The analysis in the preceding three essays has added to the understanding of intergenera-
tional transfers and intra-household allocation of resources with the prospect of population
aging. Various aspects of transfers are looked at, ranging from the old-age support provided
by children to the investments in the human capital of children made by parents. Aging
alters the incentives associated with these intergenerational transfers, thus altering the pat-
tern of distribution of resources across generations. Importantly, this research showcases the
role of household-decision making in an aging society.
Chapter 2 examines the determinants of the old-age support to parents. Existing
literature has shown that children's levels of income, education, age, gender, and other
characteristics play a role in determining transfers to old-age parents. This paper evaluates
the role of the dierences in the educational attainment of the parents and the children.
Using the data from China Health and Retirement Longitudinal Study (CHARLS), it nds
that the gap in the educational level across generations is positively correlated to transfers
from children. The correlation is statistically signicant, and its magnitude is substantial.
The elasticity of transfers to parents with respect to the education gap ranges between 0.3
% to 0.5 %. This result holds even after accounting for the child's and the parents' current
income levels and also the gap in their income levels. The implications of this result are
86
straightforward; less-educated parents of more educated children receive maximum support
in their old age. It also highlights the fact that education and intergenerational mobility
in education are essential independent factors in the determination of old-age support since
they signal more than just the individuals' earning capacity. This exercise also provides
insights into the mechanics of the household decision-making process.
The third chapter takes the study of household behavior further by looking at the
changes in parental transfers to children with population aging. The premise of the inter-
temporal choice set of any household is the knowledge of nite and uncertain lifetimes. Hence,
the choice set is bound to change with the decline in the mortality rate and progression in
aging. An overlapping generations model is used to study how the households belonging to
successive generations invest in their children's human capital as the demography changes.
Comparison with an alternative scenario where there is no change in the mortality rates over
time shows that the demographic change in the actual world has led to lower investments
in the human capital of children and an increase in the household-level savings rate. Simu-
lations using the Chinese population data are used as an example to illustrate the impact of
population aging. The analysis stands true of any economy going through the demographic
transition and progressing towards an aging population. The primary takeaway here is that
the demographic structure matters for investments in children's human capital and that
aging of the population (declining mortality rates) has caused these investments to shrink.
Variations to the basic model, such as endogenizing the old-age transfers and introducing
the social security system, do not alter the main results.
The nal essay presents an empirical study that evaluates if the trends in the real-
world data support the results produced in the previous chapter. Analysis of a cross-country
panel data-set suggests that the countries at a more advanced stage of population aging have
lower rates of household educational expenditure per-child compared to the relatively younger
countries in the sample. This negative relationship is not causal but provides evidence of the
87
existence of the causal relationship established in Chapter 3. A higher proportion of old-age
adults in the population or a higher old-age dependency ratio is negatively and signicantly
associated with household expenditure on education per child. Also, a higher life expectancy
at age 60 is associated with lower expenditures given the proportion of population above age
60. Results for other aging measures such as the median age of the population are slightly
ambiguous but are indicative of the same negative relationship.
The big picture here is that the study of the household's decision-making process
oers clues as to how dierent macro-economic variables will behave as the demographic
structure undergoes a change and the population ages. The analysis in this dissertation
looks at this decision-making process from dierent angles. It nds that population aging
has implications for intergenerational transfers, especially those related to investment in the
human capital of future generations. Investment in human capital is critical to the future
trajectory of economic growth, development, and inequality. A comprehensive look at the
generational exchanges of resources helps inform policies on not only social security and
pensions but also public expenditure on education and child care. More informed policies
will help resolve the intergenerational con
ict in the division of resources in a better manner.
Lastly, there is much scope for further research in the current analysis presented in this
dissertation. The future endeavors to study these topics can improve upon the methodologies
adopted. In the empirical analyses, the use of identication techniques and detailed data can
help establish causality, which will enable more meaningful quantication of the relationships
studied. Secondly, in the model analysis, additional elements such as inequality in income,
income risk, and health-care risks could be added to the model framework to arrive at
more realistic conclusions. Also, population data of other countries at dierent stages of the
demographic transition could be used to compare the projected dierences in the investments
in children across dierent countries. This comparison will also be interesting from the point
of view, dierent social security systems, and education systems in dierent countries. Last
88
but not least, intergenerational transfers and aging dynamics can also be explored in the
context of dierent cultural backgrounds. In the United States, for instance, the norms of
old-age support are dierent from those in Asian countries. Even the aging of the population
within the US is likely to yield dierent results for people of dierent ethnic backgrounds.
The relevance of such an analysis grows as the populations become more ethnically and
culturally diverse.
89
Bibliography
Afonso, Ant onio, and Miguel St Aubyn. \Cross-country eciency of secondary education
provision: A semi-parametric analysis with non-discretionary inputs." Economic mod-
elling 23, no. 3 (2006): 476{491.
Aizer, Anna, and Flavio Cunha. The production of human capital: Endowments, investments
and fertility. Technical report. National Bureau of Economic Research, 2012.
Alesina, Alberto, and Dani Rodrik. \Distributive politics and economic growth." The quar-
terly journal of economics 109, no. 2 (1994): 465{490.
Altonji, Joseph G, Fumio Hayashi, and Laurence J Kotliko. \Parental altruism and inter
vivos transfers: Theory and evidence." Journal of political economy 105, no. 6 (1997):
1121{1166.
Auerbach, Alan J, and Laurence J Kotliko. \Evaluating scal policy with a dynamic sim-
ulation model." The American Economic Review 77, no. 2 (1987): 49{55.
Bairoliya, Neha, David Canning, Ray Miller, and Akshar Saxena. \The macroeconomic and
welfare implications of rural health insurance and pension reforms in China." The Jour-
nal of the Economics of Ageing 11 (2018): 71{92.
Barro, Robert J. \Economic growth in a cross section of countries." The quarterly journal
of economics 106, no. 2 (1991): 407{443.
. \Inequality and Growth in a Panel of Countries." Journal of economic growth 5, no.
1 (2000): 5{32.
Barro, Robert J, and Gary S Becker. \Fertility choice in a model of economic growth."
Econometrica: journal of the Econometric Society, 1989, 481{501.
Becker, Gary. A Treatise on the Family. National Bureau of Economic Research, Inc, 1981.
https://EconPapers.repec.org/RePEc:nbr:nberbk:beck81-1.
Becker, Gary S. \A theory of social interactions." Journal of political economy 82, no. 6
(1974): 1063{1093.
90
Behrman, Jere R, Robert A Pollak, and Paul Taubman. \Parental preferences and provision
for progeny." Journal of Political Economy 90, no. 1 (1982): 52{73.
Ben-Porath, Yoram. \The production of human capital and the life cycle of earnings." Jour-
nal of political economy 75, no. 4, Part 1 (1967): 352{365.
Bernanke, Ben S, and Refet S G urkaynak. \Is growth exogenous? taking mankiw, romer,
and weil seriously." NBER macroeconomics annual 16 (2001): 11{57.
Bian, Fuqin, John R Logan, and Yanjie Bian. \Intergenerational relations in urban China:
Proximity, contact, and help to parents." Demography 35, no. 1 (1998): 115{124.
Boar, Corina. Dynastic precautionary savings. Technical report. National Bureau of Eco-
nomic Research, 2020.
Boersch-Supan, Axel H, and Joachim K Winter. Population aging, savings behavior and
capital markets. Technical report. National bureau of economic research, 2001.
Bonke, Jens, and Martin Browning. \Spending on children: Direct survey evidence." The
Economic Journal 121, no. 554 (2011): F123{F143.
Browning, Martin, Lars Peter Hansen, and James J Heckman. \Micro data and general
equilibrium models." Handbook of macroeconomics 1 (1999): 543{633.
Cervellati, Matteo, and Uwe Sunde. \Human capital formation, life expectancy, and the
process of development." American Economic Review 95, no. 5 (2005): 1653{1672.
Chen, Junhua, Fei Guo, and Ying Wu. \One decade of urban housing reform in China: Urban
housing price dynamics and the role of migration and urbanization, 1995{2005." Habitat
International 35, no. 1 (2011): 1{8.
Chi, Wei, and Xiaoye Qian. \Human capital investment in children: An empirical study
of household child education expenditure in China, 2007 and 2011." China Economic
Review 37 (2016): 52{65.
Chou, Rita Jing-Ann. \Filial piety by contract? The emergence, implementation, and im-
plications of the \family support agreement" in China." The Gerontologist 51, no. 1
(2011): 3{16.
Choukhmane, Taha, Nicholas Coeurdacier, and Keyu Jin. \The one-child policy and house-
hold savings," 2013.
Curtis, Chadwick C, Steven Lugauer, and Nelson C Mark. \Demographic patterns and house-
hold saving in China." American Economic Journal: Macroeconomics 7, no. 2 (2015):
58{94.
91
Daruich, Diego, and Julian Kozlowski. \Explaining Intergenerational Mobility: The Role of
Fertility and Family Transfers." Review of Economic Dynamics, Forthcoming.
De Nardi, Mariacristina, Selahattin Imrohoro glu, and Thomas J Sargent. \Projected US
demographics and social security." Review of Economic dynamics 2, no. 3 (1999): 575{
615.
Deaton, Angus S, Pierre-Olivier Gourinchas, and Christina Paxson. \Social security and
inequality over the life cycle." In The distributional aspects of Social Security and Social
Security reform, 115{148. University of Chicago Press, 2002.
Deaton, Angus S, and Christina Paxson. \Saving, growth, and aging in Taiwan." In Studies
in the Economics of Aging, 331{362. University of Chicago Press, 1994.
Deaton, Angus S, and Christina H Paxson. \Aging and inequality in income and health."
The American Economic Review 88, no. 2 (1998): 248{253.
Ehrlich, Isaac, and Francis T Lui. \Intergenerational trade, longevity, and economic growth."
Journal of Political Economy 99, no. 5 (1991): 1029{1059.
Fang, Hai, Karen N Eggleston, John A Rizzo, Scott Rozelle, and Richard J Zeckhauser.
The returns to education in China: Evidence from the 1986 compulsory education law.
Technical report. National Bureau of Economic Research, 2012.
Fu, Chien-Hao. \Living arrangement and caregiving expectation: the eect of residential
proximity on inter vivos transfer." Journal of Population Economics 32, no. 1 (2019):
247{275.
Gourieroux, Christian, Alain Monfort, and Alain Trognon. \Pseudo maximum likelihood
methods: Applications to Poisson models." Econometrica: Journal of the Econometric
Society, 1984, 701{720.
Gradstein, Mark, and Michael Kaganovich. \Aging population and education nance." Jour-
nal of public economics 88, no. 12 (2004): 2469{2485.
Gu, Danan, and Denese Vlosky. \Long-term care needs and related issues in China." Social
sciences in health care and medicine, 2008, 52{84.
Harris, Amy Rehder, William N Evans, and Robert M Schwab. \Education spending in an
aging America." Journal of public economics 81, no. 3 (2001): 449{472.
Heckman, J. \Sample Selection Bias as a Specication Error." Econometrica 47, no. 1 (1979):
153{161.
92
Hirazawa, Makoto, and Akira Yakita. \Labor supply of elderly people, fertility, and economic
development." Journal of Macroeconomics 51 (2017): 75{96.
Huggett, Mark, Gustavo Ventura, and Amir Yaron. \Human capital and earnings distribution
dynamics." Journal of Monetary Economics 53, no. 2 (2006): 265{290.
. \Sources of lifetime inequality." American Economic Review 101, no. 7 (2011): 2923{
54.
_
Imrohoro glu, Ay se, and Kai Zhao. \Intergenerational transfers and China's social security
reform." The Journal of the Economics of Ageing 11 (2018a): 62{70.
. \The chinese saving rate: Long-term care risks, family insurance, and demographics."
Journal of Monetary Economics 96 (2018b): 33{52.
Kalemli-Ozcan, Sebnem, Harl E Ryder, and David N Weil. \Mortality decline, human capital
investment, and economic growth." Journal of development economics 62, no. 1 (2000):
1{23.
Lee, Jong{Wha, and Robert J Barro. \Schooling quality in a cross{section of countries."
Economica 68, no. 272 (2001): 465{488.
Lee, Ronald. \Macroeconomics, aging, and growth." In Handbook of the economics of popu-
lation aging, 1:59{118. Elsevier, 2016.
. \The demographic transition: three centuries of fundamental change." Journal of
economic perspectives 17, no. 4 (2003): 167{190.
Lee, Ronald, and Andrew Mason. \Population aging, intergenerational transfers, and eco-
nomic growth: Asia in a global context." In Aging in Asia: Findings from New and
Emerging Data Initiatives. National Academies Press (US), 2012.
Lee, Yean-Ju, and Zhenyu Xiao. \Children's support for elderly parents in urban and rural
China: Results from a national survey." Journal of cross-cultural gerontology 13, no. 1
(1998): 39{62.
Lei, Xiaoyan, John Giles, Yuqing Hu, Albert Park, John Strauss, and Yaohui Zhao. Patterns
and correlates of intergenerational non-time transfers: evidence from CHARLS. The
World Bank, 2012.
Li, Hongbin, Jie Zhang, and Junsen Zhang. \Eects of longevity and dependency rates on
saving and growth: Evidence from a panel of cross countries." Journal of Development
Economics 84, no. 1 (2007): 138{154.
93
Li, Meng, and Kunrong Shen. \Population aging and housing consumption: A nonlinear
relationship in China." China & World Economy 21, no. 5 (2013): 60{77.
Lin, Ju-Ping, and Chin-Chun Yi. \Filial norms and intergenerational support to aging par-
ents in China and Taiwan." International Journal of Social Welfare 20 (2011): S109{
S120.
Lyons, Angela C, John E Grable, and So-Hyun Joo. \A cross-country analysis of population
aging and nancial security." The Journal of the Economics of Ageing 12 (2018): 96{
117.
Ma, Mingming. \Does Children's Education Matter for Parents' Health and Cognition in
Old Age." Evidence from China. University of Southern California, Los Angeles, 2017.
Maestas, Nicole, Kathleen J Mullen, and David Powell. The eect of population aging on
economic growth, the labor force and productivity. Technical report. National Bureau of
Economic Research, 2016.
Mankiw, N Gregory, David Romer, and David N Weil. \A contribution to the empirics of
economic growth." The quarterly journal of economics 107, no. 2 (1992): 407{437.
Manuelli, Rodolfo E, and Ananth Seshadri. \Explaining international fertility dierences."
The Quarterly Journal of Economics 124, no. 2 (2009): 771{807.
Menz, Tobias, and Heinz Welsch. \Population aging and carbon emissions in OECD coun-
tries: Accounting for life-cycle and cohort eects." Energy Economics 34, no. 3 (2012):
842{849.
Mincer, Jacob A. \The human capital earnings function." In Schooling, Experience, and
Earnings, 83{96. NBER, 1974.
Mu, Ren, and Yang Du. \Pension coverage for parents and educational investment in children:
Evidence from urban China." The World Bank Economic Review 31, no. 2 (2017): 483{
503.
Nugent, Jerey B. \The old-age security motive for fertility." Population and development
review, 1985, 75{97.
Parsons, Donald O. \The cost of school time, foregone earnings, and human capital forma-
tion." Journal of Political Economy 82, no. 2, Part 1 (1974): 251{266.
Peng, Xizhe. \China's demographic history and future challenges." Science 333, no. 6042
(2011): 581{587.
94
Qian, Joanne Xiaolei, and Russell Smyth. \Educational expenditure in urban China: income
eects, family characteristics and the demand for domestic and overseas education."
Applied Economics 43, no. 24 (2011): 3379{3394.
Raut, Lakshmi K, and Lien H Tran. \Parental human capital investment and old-age trans-
fers from children: Is it a loan contract or reciprocity for Indonesian families?" Journal
of Development Economics 77, no. 2 (2005): 389{414.
Silva, JMC Santos, and Silvana Tenreyro. \The log of gravity." The Review of Economics
and statistics 88, no. 4 (2006): 641{658.
Sin, Yvonne. \Pension liabilities and reform options for old age insurance." World Bank
working paper 1 (2005).
Song, Zheng, Kjetil Storesletten, Yikai Wang, and Fabrizio Zilibotti. \Sharing high growth
across generations: pensions and demographic transition in China." American Economic
Journal: Macroeconomics 7, no. 2 (2015): 1{39.
Srensen, Rune J. \Does aging aect preferences for welfare spending? A study of peoples'
spending preferences in 22 countries, 1985{2006." European Journal of Political Econ-
omy 29 (2013): 259{271.
Theerawanviwat, Duanpen. \Intergenerational transfers and family structure: Evidence from
Thailand." Ageing International 39, no. 4 (2014): 327{347.
Tosun, Mehmet Serkan, Claudia R Williamson, and Pavel A Yakovlev. \Population aging,
elderly migration and education spending: Intergenerational con
ict revisited," 2009.
Tsang, Mun C. \Financial reform of basic education in China." Economics of Education
Review 15, no. 4 (1996): 423{444.
Walker, Alan. \The economic `burden'of ageing and the prospect of intergenerational con-
ict." Ageing & Society 10, no. 4 (1990): 377{396.
Willis, Robert J. The old age security hypothesis and population growth, 1979.
Xie, Yu, and Haiyan Zhu. \Do sons or daughters give more money to parents in urban
China?" Journal of Marriage and Family 71, no. 1 (2009): 174{186.
Yuan, Cheng, and Lei Zhang. \Public education spending and private substitution in urban
China." Journal of Development Economics 115 (2015): 124{139.
Zhu, Haiyan. \Adult children's characteristics and intergenerational nancial transfers in
urban China." Chinese Journal of Sociology 2, no. 1 (2016): 75{94.
95
Zimmer, Zachary, and Julia Kwong. \Family size and support of older adults in urban and
rural China: Current eects and future implications." Demography 40, no. 1 (2003):
23{44.
96
Appendix A
Appendix to chapter 2
In CHARLS, educational attainment is classied into 10 dierent levels. Table A.1 sum-
marizes the sample distribution by the education categories and the corresponding years of
schooling.
Table A.1: Distribution of population by education categories
Category Schooling
(yrs.)
Frac. of child
pop. (%)
Frac. of parent 1
pop. (%)
Frac. of parent 2
pop. (%)
No formal education/ illit-
erate
0 2.73 26.14 59.55
Did not nish primary
school but can read
2 10.42 17.91 12.12
Elementary school 6 20.80 28.12 14.89
Middle school 9 32.63 17.08 8.89
High school/ Vocational
training
12 17.66 8.4 3.56
Two/Three year college/
Associate degree
14.5 6.66 1.61 0.66
Four year college/ Bache-
lor's degree
16 8.02 0.70 0.33
Post-graduate 19 0.91 0.04 0
PhD 21 0.17 { {
For parents, the last two categories are combined in the data-set because of the negligible population in them.
97
In CHARLS wave 2015, reported income of children are classied into 12 categories. The
income categories and the sample distribution are given in Table A.2.
Table A.2: Distribution of population by income categories
Category Total Income last year Fraction of child pop.
(%)
Fraction of parent pop.
(%)
1 0 - -
2 Less than 2000 Yuan 1.82 27.63
3 2000-5000 Yuan 4.80 14.02
4 5000-10000 Yuan 7.32 7.32
5 10000-20000 Yuan 18.40 8.93
6 20000-30000 Yuan 20.60 6.95
7 30000-50000 Yuan 22.00 9.26
8 50000-100000 Yuan 17.29 6.45
9 100000-150000 Yuan 4.84 1.12
10 150000-200000 Yuan 1.24 0.37
11 200000-300000 Yuan 0.79 0.08
12 More than 300000 Yuan 0.91 0.17
98
Appendix B
Appendix to chapter 3
.2 .4 .6 .8 1
Fraction of population
1870 1890 1910 1930 1950 1970 1990 2010 2020
year
Ages 64-85
Ages 20-63
Ages 0-19
Figure B.1: Age-distribution of population
99
.07 .08 .09 .1 .11
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
Constant γ + constant n
(a)
0 .1 .2 .3
Aggregate savings rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
Constant γ + constant n
Data
(b)
Figure B.2: Household investment and saving rate in the three demographic models with
social security sector
100
.055 .06 .065 .07 .075
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
Constant γ + constant n
(a)
0 .1 .2 .3
Aggregate savings rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
Constant γ + constant n
Data
(b)
Figure B.3: Household investment and saving rate in the three demographic models with
= 0:8
101
.045 .05 .055 .06
Aggregate educational investment rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
Constant γ + constant n
(a)
0 .1 .2 .3
Aggregate savings rate
1950 1960 1970 1980 1990 2000 2010
Baseline
Constant γ
Constant γ + constant n
Data
(b)
Figure B.4: Household investment and saving rate in the three demographic models with
= 0:7
102
Appendix C
Appendix to chapter 4
Table C.1: Descriptive statistics: indicators of aging
Variable N Mean Std. Dev. Min Max
OECD countries
Old-age dep. ratio 410 22 5.7 8.5 35
Median Age 410 38 4.6 23 46
Life exp. at 60 410 23 1.7 18 26
TFR 410 1.7 .4 1.1 3.1
Non-OECD countries
Old-age dep. ratio 133 14 6.4 2.5 29
Median Age 133 30 6.2 17 42
Life exp. at 60 133 19 1.8 16 23
TFR 133 2.2 .59 1.1 5
All countries
Old-age dep. ratio 543 20 6.9 2.5 35
Median Age 543 36 5.9 17 46
Life exp. at 60 543 22 2.2 16 26
TFR 543 1.8 .49 1.1 5
103
Table C.2: Correlation between household education expenditure and age composition
HH expenditure on Education: Overall Primary Secondary Tertiary
Old-age dep. ratio -0.36 -0.42 -0.45 -0.62
Median age -0.31 -0.36 -0.48 -0.60
TFR 0.12 0.12 0.39 0.44
Life exp. at 60 0.06 -0.15 -0.16 -0.30
Population shares in age groups :
75 & abv. -0.35 -0.41 -0.45 -0.62
50-74 -0.34 -0.39 -0.53 -0.63
15-49 0.39 0.48 0.35 0.50
Below 15 0.27 0.31 0.51 0.59
Table C.3: Results with Median age for the extended sample
Overall Primary Secondary Tertiary
Median Age -0.144 -0.146
-0.185 1.224
(0.0978) (0.0656) (0.142) (0.371)
Population 0.0140
0.00292
+
0.0168
0.0108
(0.00434) (0.00168) (0.00788) (0.0186)
Per-capita real GDP -0.0346 -0.0303 -0.0380
+
-0.596
(0.0294) (0.0224) (0.0228) (0.225)
Gov. exp. on education -0.00645 -0.0159 -0.0359 -0.0264
(0.0534) (0.0352) (0.0444) (0.0481)
N 509 292 277 289
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
104
Table C.4: Results with Age eects for the extended sample
Overall Primary Secondary Tertiary
Age 75+ -0.478
+
-0.253 -0.0930 0.136
(0.280) (0.279) (0.466) (1.203)
Age 50-75 -0.231 -0.267 -0.798
-0.375
(0.142) (0.198) (0.194) (0.987)
Age 15-49 -0.330
+
-0.179 -0.660
-1.211
(0.178) (0.294) (0.215) (0.903)
Age 15- - - - -
Population 0.0130
0.00164 0.0123 0.00597
(0.00396) (0.00212) (0.00801) (0.0192)
Per-capita real GDP -0.0223 -0.0275 -0.0152 -0.436
(0.0246) (0.0212) (0.0264) (0.202)
Gov. exp. on education 0.00786 0.00989 -0.0267 -0.0324
(0.0540) (0.0229) (0.0321) (0.0484)
N 509 292 277 289
Standard errors in parentheses
+
p< 0:10,
p< 0:05,
p< 0:01,
p< 0:001
105
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Inter-temporal allocation of human capital and economic performance
PDF
The determinants and measurement of human capital
PDF
Essays on health and aging with focus on the spillover of human capital
PDF
Essays in macroeconomics
PDF
Three essays on economics of early life health in developing countries
PDF
Long-term impacts of childhood adversity on health and human capital
PDF
Intergenerational transmission of values and behaviors over the family life course
PDF
Internet communication use, psychological functioning and social connectedness at older ages
PDF
Three essays on human capital and family economics
PDF
Social determinants of physiological health and mortality in China
PDF
Essays on development and health economics
PDF
The impact of globalization, economics, and educational policy on the development of 21st-century skills and education in science, technology, engineering, and mathematics in Costa Rican schools
PDF
Essays in labor economics: demographic determinants of labor supply
PDF
Essays on health economics
PDF
Subclinical carotid atherosclerosis, psychosocial measures, and cognitive function in middle- to older-aged adults
PDF
Age differences in diffusivity in the locus coeruleus and its ascending noradrenergic tract
PDF
Self-perceptions of Aging in the Context of Neighborhood and Their Interplay in Late-life Cognitive Health
PDF
Corporate reputation crisis in the digital age: a comparative study on Abercrombie & Fitch’s reputation crisis in the U.S., China and Taiwan
PDF
The influence of child day care on cognitive and psychosocial functioning in adolescence
PDF
The role of globalization, science, technology, engineering, and mathematics project‐based learning, and the national science and technology fair mandate in creating 21st‐century-ready students i...
Asset Metadata
Creator
Aggarwal, Kanika
(author)
Core Title
Intergenerational transfers & human capital investments in children in the era of aging
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
06/22/2020
Defense Date
05/06/2020
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
age demography,aging,CHARLS,cross-country,education,educational expenditure,human capital,human capital investment,intergenerational transfers,OAI-PMH Harvest,old age support,overlapping generation,Retirement
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Imrohoroglu, Ayse (
committee chair
), Bairoliya, Neha (
committee member
), Betts, Caroline (
committee member
), Nugent, Jeffrey B. (
committee member
)
Creator Email
kanika1207@gmail.com,kanikaag@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-319963
Unique identifier
UC11665927
Identifier
etd-AggarwalKa-8605.pdf (filename),usctheses-c89-319963 (legacy record id)
Legacy Identifier
etd-AggarwalKa-8605.pdf
Dmrecord
319963
Document Type
Dissertation
Rights
Aggarwal, Kanika
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
age demography
CHARLS
cross-country
education
educational expenditure
human capital
human capital investment
intergenerational transfers
old age support
overlapping generation