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

net/publication/288932276

The Impact of Education on Income Inequality and Intergenerational Mobility

Article  in  China Economic Review · December 2015

DOI: 10.1016/j.chieco.2015.12.009

CITATIONS READS

75 3,964

2 authors:

Juan Yang Muyuan Qiu

Beijing Normal University Peking University
12 PUBLICATIONS   235 CITATIONS    1 PUBLICATION   75 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Muyuan Qiu on 29 October 2019.

The user has requested enhancement of the downloaded file.

 China Economic Review 37 (2016) 110–125

Contents lists available at ScienceDirect

China Economic Review

The impact of education on income inequality and

intergenerational mobility
Juan YANG a, Muyuan QIU b,⁎
a
School of Economics and Management, Beijing Normal University, 19, Xinjiekouwai Dajie, Beijing 100875, China
b
Institute of World Economics and Politics, Chinese Academy of Social Sciences, No.5 Jianguomennei Street, Beijing 100732, China

a r t i c l e i n f o a b s t r a c t

Article history: This paper analyses the effects of innate ability, compulsory education (grades 1–9), and non-
Received 17 June 2015 compulsory education (grades 10–12 and higher education) on inequality and intergeneration-
Received in revised form 24 December 2015 al mobility of income, by constructing a four-period overlapping-generation model. We find
Accepted 24 December 2015
that innate ability and family investment in early education play important roles in explaining
Available online 31 December 2015
income inequality and intergenerational income mobility. Though children from the wealthiest
families are only 1.36 times ‘smarter’ that those from the poorest, the gap in human capital ex-
JEL classification: pands to 2.35 at the end of compulsory education and to 2.89 at the end of non-compulsory
I22
education. One important reason for the increase is that poor families invest relatively less in
I28
children's early education than do wealthy families; therefore, their children attend lower-
H31
J24 quality schools, which results in them being much less likely to participate in higher education.
By simulating policy experiments for different types of government education expenditure, we
Keyword:
find that direct subsidies to poor parents are the most efficient and effective policy for mitigat-
Education income inequality
Intergenerational mobility
ing poor families' budget constraints with regard to early-education investment in their
Educational expenditure children.
Early education investment © 2015 Elsevier Inc. All rights reserved.

1. Introduction

For quite a long time, economists have been more concerned with income inequality than with intergenerational income
mobility, because most economic models assume that income inequality is caused by the random assignments of ability and
other market factors. However, over the past several years, scholars have found that the assignment of high levels of ability is
not a random process but instead determined by parents' inheritance (Plug, 2004; Björklund, Lindahl, & Plug, 2006). Therefore,
intergenerational income mobility has become a popular topic, because it affects a country's long-run economic situation and
the equality of economic opportunities for the country's citizens. Becker and Tomes (1979) proposed that income distribution
should include two components: income inequality between generations (i.e., intergenerational income mobility) and income in-
equality among families within the same cohort. Combining intergenerational income mobility and income inequality in a model
may lead to a better understanding of income distribution.
Many empirical studies have found that China's income inequality has reached a warning level. Li and Luo (2011) showed that
the income of the wealthiest 10% of the population was 32.8 times the income of the poorest 10% and the average income in
urban areas was 3.87 times the average income in rural areas. In addition, the income gap is still widening (Ravallion & Chen,
2007; Meng, Shen, & Xue, 2013). As a developing country, China has much lower intergenerational income mobility than many
developed countries (e.g., Denmark 0.089, Sweden 0.141, UK 0.198, and US 0.357). Using data from Chinese Household Income

⁎ Corresponding author.
E-mail addresses: yangjuan@bnu.edu.cn (J. Yang), qiumuyuan@126.com (M. Qiu).

http://dx.doi.org/10.1016/j.chieco.2015.12.009
1043-951X/© 2015 Elsevier Inc. All rights reserved.
 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125 111

Project, Deng, Gustafsson, and Li (2013) estimates the urban income precipitancy of father-son pairs is 0.47 in 1995 and 0.53 in
2002. They found that intergenerational income mobility had a declining trend, similar to the findings of Eriksson and Zhang
(2012) who used rural data. Gong, Leigh, and Meng (2012) used the UHEE dataset and obtained an even higher persistence by
incorporating income fluctuation and life-cycle bias.
This situation may affect economic development or social stability. Therefore, it is urgent for Chinese government to put in
place public policies to decrease income inequality and promote intergenerational income mobility. However, designing effective
public policy requires that the sources of income inequality and intergenerational income mobility first be determined. Many
studies in China have found that the expansion of education plays an important role in explaining income inequality (Lai,
2004; Bai, 2004); however, they have not considered individuals' heterogeneity and the reasons for intergenerational income mo-
bility. This paper proposes an overlapping-generations model to analyse the effects of innate ability, compulsory education, and
non-compulsory education on income inequality and income persistence across generations and explores how to design effective
education policy based on the findings from the model.
Specifically, we construct an equilibrium model of income inequality and intergenerational income mobility. Among the factors
that determine the equilibrium of the model, innate ability and educational investment play the most important role while other
factors such as the tax rate and public education policy can also affect the equilibrium. Another main contribution of our model is
as education has been divided into compulsory and non-compulsory stage, we analyse separately the effects of innate ability,
compulsory education, and non-compulsory education on income inequality and intergenerational income mobility. By computing
early-education investment, senior high school attendance rates, and higher education attendance rates for various income
groups, we determine the sources of income inequality. Finally, we examine the effectiveness of different types of public educa-
tion investment policy in reducing income inequality and increasing intergenerational income mobility.
The rest of the paper is composed as follows: In Section 2, we summarize the literature about the effect of education on in-
come inequality and intergenerational income mobility; Section 3 describes the setting of the benchmark model; we calibrate
the parameters of the model in Section 4; analyse the sources of income inequality and income persistence across generations
in Section 5; Section 6 conducts an analysis of the model behaviour to some key parameters; Section 7 studies the model impli-
cation for different designs of public education expenditure and the conclusions are provided in Section 8.

2. Literature review

Early human capital theory indicates that education may increase income, and thus, educational expansion or increasing public
expenditure on education will decrease income inequality and increase intergenerational mobility. However, the empirical
evidence shows that educational expansion augments income gaps because the rate of return on higher education is much higher
than the rate of return on compulsory education. In addition, when more and more people acquire higher education, the rate of
return on higher education still remains at a comparatively high position; in other words, educational expansion does not reduce
income inequality (Mincer, 1974). Consequently many scholars have used different data and methodologies to explain the reasons
why increased higher education subsidies or decreased tuition fees (i.e., more equal educational opportunity) will not significantly
promote the equal distribution of income.
Checchi, Ichino, and Rustichini (1999) found that family background is an important factor of the performance in the labour
market. Under such circumstances, the equal opportunity of higher education will bring many benefits to children from poor
families. The explanation by Hendel, Shapiro, and Willen (2005) is similar to signal theory. If high school graduates are able to
go to college through acquiring loans, those who do not attend college will indicate one of two potential attributes to potential
employers: that they have low levels of ability, or that they have high levels of ability but are from poor families. If budget con-
straints do not exist; in other words, if all high school graduates can afford to go to college, then only low-ability individuals will
not participate in higher education. Because those who do not participate in higher education will not indicate that they have high
levels of abilities, employers will squeeze the wages of non-skilled workers and enlarge the wage gap between graduates with
higher education and those without higher education. The research by Sylwester (2002) reached different results. Using samples
from 50 countries including those from the OECD, East Asia, Latin America and Africa, he found that increased public education
expenditures relieved income inequality represented by Gini coefficients. The results are quite robust in different countries. In
addition, the effect of reducing income inequality in high income countries is more significant.
Several economists have researched the relationship between education and intergenerational mobility. Becker and Tomes
(1979) constructed a basic theoretical model of intergenerational mobility to explain the effect of education on income inequality
and intergenerational income mobility. Based on this model, many scholars have used empirical evidence to analyse the relation-
ship between educational expansion (an increase in expenditures on education) and income inequality and intergenerational mo-
bility. For example, Restuccia and Urrutia (2004) estimated the impact of innate ability, compulsory education and higher
education on income gaps and income persistence across generations using US data-adjusted parameters. They found that parents'
early educational investment could explain approximately 50% of intergenerational mobility, and income inequality was mainly
due to higher education. Solon (2004) points out three fundamental institutions determining generational mobility—the family,
the labour market and the state by constructing a standard intergenerational mobility model. Families differing in their capacities
and resources to invest in their children, results their children have different school achievements. Nakamura and Murayama
(2011) demonstrated that income distribution and intergenerational mobility were mainly attributable to the proportion of edu-
cation costs on income through setting up a model about intergenerational mobility and economic growth.
112 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125

Therefore, when we estimate the impact of education on income inequality and intergenerational mobility, we should be very
careful about the assumption on the relationship between education costs and income. Many studies assume that the proportion
of educational costs to income will decrease with the increase of income when they estimate intergenerational mobility with the
general equilibrium model (Maoz & Moav, 1999), while these conclusions will not be accurate if the proportion of education costs
increases with an increase in income. Zhu and Vural (2013) showed that wealthy parents will invest more in their children's ed-
ucation, and therefore, we assume a positive relationship between educational investment and parents' income in this paper.
These researches motivate us to investigate the relations between education and income inequality in China. There are quite a
few literature discuss the intergenerational income mobility in China (Deng et al., 2013; Eriksson & Zhang, 2012; Gong et al.,
2012) and all of them found the intergenerational persistence are very high. However they did not investigate the reasons behind
high persistence. China's education system is composed by compulsory education and non-compulsory education. Compulsory
education includes primary education and lower secondary education. Non-compulsory education refers to upper secondary
education and higher education, which is quite different to other OECD countries. Students have to pay comparatively expensive
tuition fees in upper secondary education and not all the students are able to read upper secondary education. In other words,
students and parents will make their education choice in grade 9 instead of grade 12. In addition, China's higher education is com-
paratively competitive than OECD counties, more than 30% senior high school graduates cannot be enrolled in any colleges and
about 50% senior high school graduates cannot be enrolled in four-year universities. Students' education choice in grade 9 has
to consider their probability to attend universities three years later, not only because the competitiveness of higher education,
but the rate of return to senior high school is very low. Our model is mainly based on Restuccia and Urrutia (2004) by considering
China education system. Based on this information, individuals' lifetime was divided into two periods: in the first period, individ-
uals finish compulsory education and in the second period, individuals attend non-compulsory education or go to work. For those
attend non-compulsory education will face the probability not to be enrolled in universities.

3. The model

The economy is populated by individuals who each live for four periods: child, juvenile, young parent, and old parent. An
individual is born as a child and lives two periods–child and juvenile–with his or her parents. In the model early education is de-
fined as the nine-year compulsory education. For the juveniles, they are in the stage of non-compulsory education which include
senior high school and college education. At the end of the second period, a juvenile starts his or her own family and lives the
next two periods as a parent. As a child, the individual inherits innate abilities from his or her parents and does not attend family
economic decisions. Parents select child's education levels for compulsory and non-compulsory schooling. As a result, there are
only two types of decision units in the economy: young parents who live a young child and old parents who live with a juvenile.
For convenience, we denote the first type as a young family and the second as an old family and specify their respective decision
problems accordingly.

3.1. Young family

Early education is paid jointly by government and family expenditure. Government is responsible for tuition fees and textbook
fees, while parents pay miscellaneous fees, such as after-class tutoring fees, sponsor fees, etc. The better the quality of the educa-
tion, the higher the private costs. We assume that parents are altruistic and care only about next generation's quality of life, which
is measured by their children's future Value function. We further assume that families cannot borrow and have no savings; in
other words, all of their expenditures on their children's education affect only current consumption. This assumption may be
very strong, but Carneiro and Heckman (2002) found it a very reasonable one for young parents. Based on the altruism assump-
tion, although investments in their children's education will limit their consumption, young parents will still make the optimal
early-education expenditures based on their children's innate abilities (π) and the government's education expenditures (g).
Specifically, given wage rate w, the value function of young parents Vy(hy, π)can be represented as:


  0 
^
V y hy ; π ¼ max u cy þ β  V o ho ; π; π ð1Þ
c;e

subject to the following constraints:

cy þ e ¼ ð1−τ Þwhy ; e≥0 ð2Þ

0
ho ¼ η  hy ;η ≥1 ð3Þ

^ ¼ Gðe; π; g Þ
π ð4Þ

In Eq. (1), cy is consumption of young family and u(cy) is current utility. hy is human capital level of young parents, π is their
0
^ Þ is the value function of next period. In Eq. (2), e de-
children's innate ability, β is the subjective discount factor and V o ðho ; π; π
notes young parents' early educational investment in their children,τ is tax rate they have to pay for public education expenditure.
 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125 113

Eq. (3) is the law of human capital accumulation for the young parents, which shows that young parents' human capital level ho′
in next period is proportion to their current level hy. In the model, we impose η N 1, which allows human capital to increase with
work experience. Eq. (4) describes how children's acquired human capital: acquired ability (π ^) in the next period is a function of
children's innate ability π, public and private human capital investment g and e. As discussed by Restuccia and Urrutia (2004),
G(e,π,g) take the following form:

^ ¼ Gðπ; e; g Þ ¼ π ðe þ g Þγ ; 0 b γ b 1
π ð5Þ

where γ is the elasticity of acquired ability to education investment which controls the curvature of the Eq. (5).

3.2. Old family

The old parents choose whether to send their juveniles to senior high school in order to have the opportunity to receive a
college education. This decision depends on the juvenile's innate ability, child's educational achievement during compulsory
education and parents' income. Tuition fees for senior high schools and college are also paid jointly by the government and indi-
viduals, however, during the stage of non-compulsory education, government expenditure(subsidy) tilts towards children from poor
family. As not all senior high school students can enrol in universities, the probability for a senior high school enrolled juvenile receiv-
ing college education is determined by the juvenile's innate ability and the quality of his or her compulsory education.
Old parents' utility depends on old parents' current human capital ho, the juvenile's innate ability π and his acquired human
^. Let s= 1 denotes that old parents sent their juvenile to senior high school and s =0 be the otherwise, old parents choose
capital π
^ Þ as follows:
s to maximize their value function V o ðho ; π; π
n h io
^Þ ¼
V o ðho ; π; π  max
 1 V~ 0 ðh ; π; π
^ Þ; E V~ 1 ðh ; π; π
^Þ ð6Þ
o o o o
~
s¼I E V ~0
NV
o o

where V
0
~ ðho ; π; π ^ Þ represents utility if the juvenile enters the workforce immediately after completing compulsory education, and
o
~ 1
^ Þ is the old parents' expected utility if they send the juvenile to senior high school. I(⋅) is indicator function which
E½V o ðho ; π; π
equals to one if E½V ~ 1 N V
~ 0 g. V
~ 0 ðho ; π; π
^ Þ is value function that solves the following problem:
o o o
h
 i
~ 0 ðh ; π; π
V ^ Þ ¼ max uðco Þ þ βE V y h0y ; π0 jπ ð7Þ
o o
co N 0

subject to the following constraints:


0
co ¼ ð1−τÞ who þ why ð8Þ

0
^:
hy ¼ π ð9Þ

In above equations, co is consumption of old family, hy′ denotes the juvenile's human capital level at the end of this period.1 For
child who enters the workforce immediately after compulsory school, π ^ equals the acquired ability π'. Furthermore, π' is the innate
ability of a descendant of the juvenile in the next period, which we will explain later in this section.
Because of the college enrolment test in China, not all children whose parents send them to senior high school will be able to
enrol in college. From the standpoint that the senior high school attendance decision is made, the juvenile's college enrolment
(θ = 1 for attendance, and 0 otherwise) is a random variable. We assume that its probability distribution, p, depends only on
the juvenile's acquire human capital, π^:

pðπ ^ Þμ 1 ; 1 :
^ Þ ¼ min μ 0 ð1 þ π ð10Þ

In Eq. (10), μ0 and μ1 control the level and curvature of the juvenile's probability of enrolling in college. Let V~ 1 ðho ; π; π
^ ; θÞ
o
denote the utility of old parents who send their children to senior high school, then,E½V ~ ðho ; π; π
1
^ Þ can be represented as
o
follows:
h i
E V~ 1 ðh ; π; π ~ 1 ðh ; π; π
^Þ ¼ V ^ ; θ ¼ 1Þ  pðπ ~ 1 ðh ; π; π
^Þ þ V ^ ; θ ¼ 0Þ  ½1−pðπ
^ Þ ð11Þ
o o o o o o

and the utility of old parents who send their children to senior high school is:
h
 i
~ 1 ðh; π; π
V ^ ; θÞ ¼ max uðco Þ þ βEt V y h0y ; π0 jπ ð12Þ
o

1
Note that this is also the human capital level when this child becomes to parents at the beginning of next period.
114 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125

subject to the following constraints:

h
i
0
co þ f high nhigh þθ  f college ncollege ¼ ð1−τ Þ who þwhy 1−nhigh −θ  ncollege þ κ high þθ  κ college ð13Þ

0
^ þ θqcollege π
hy ¼ ð1−θÞqhigh π ^: ð14Þ

In Eqs. (13) and (14), nhigh and ncollege denote the proportion of time spend in senior high school and college over this period.
fhigh and fcollege represent education costs (including opportunity costs) in senior high school and college, qhigh and qcollege are the
rate of return for senior high school education and college. khigh and kcollege denote government subsidies to senior high schools
and colleges, respectively. In the model, we assume that the government subsidies for a juvenile's senior high school and college
education depend on the tuition fee and the background of the juvenile's family, i.e.:



κ high f high nhigh ; who ¼ f high nhigh  max 1−bhigh who ; 0 ð15Þ



κcollege f college ncollege ; who ¼ f college ncollege  max 1−bcollege who ; 0 : ð16Þ

In this model, innate ability π is an important source of cross section inequality and human capital correlation between parents
and children. It is only through π that compulsory and non-compulsory (senior high school and college) increase/decrease the
inequality between and across generation. We assume the logarithm of π' and π follow a first-order autoregressive (AR(1))
process:
 0

2
log π ¼ ρ logðπÞ þ ε; ε  N 0; σ : ð17Þ

Eqs. (8) to (17) describe the trade-off between sending juveniles to senior high school and letting them enter the workforce
immediately. For those parents who send their children to senior high school, the cost is not only the tuition they pay for non-
compulsory school but also the time—and therefore foregone income—that the juvenile could earn during this period. Moreover,
there is always a risk that the child will not pass the college enrolment test (“GaoKao”) and therefore will not attend college.
However, let children to pursue better education is also profitable: child who receives better education can achieve a higher
human capital level and thus higher (discounted) utility in the future as parents.

3.3. Production

For production, we assume that human capital HD and technology A are the only two input factors, the production function of
the constant to scale firms is, i.e.:

D
F ðH Þ ¼ AH : ð18Þ

The first order condition of companies' profit maximization problem is:


0 D
w¼ F H ¼A ð19Þ

which indicates that wage rate w equals the production technology A.

3.4. Recursive competitive equilibrium

Given the tax rate τ, the subsidy function κhigh and κhigh, a recursive competitive equilibrium is a wage rate w, young and old
^ Þ, policy function for consumption cy(hy, π), co ðho ; π; π
family's value function Vy(hy, π), V o ðho ; π; π ^ Þ, education investment in com-
pulsory education e(hy, π), senior high school attendance decision sðho ; π; π ^ Þ and the distribution of young and old family
^ Þ, such that:
μy(hy, π) and μ o ðho ; π; π
1. Household utility maximization. Policy function maximizes the utility maximization problem for young and old family.
2. Firm profit maximization.
3. Accordance condition. Labour supply Hs, total public expenditure in compulsory education GKc, total public expenditure in se-
nior high school education GKhigh, total public expenditure in college education GKcollege, total private education in compulsory
education E, total senior high school expenditure (include private and public) Khigh, total college expenditure Kcollege, total
consumption of young and old family Cy and Co are summation from the micro-level.
4. Market clear.
 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125 115

1. Labour market:

HS ¼ HD ¼ H ð20Þ

2. Goods market clear:

Y ¼ F ðH Þ ¼ C y þ C o þ E þ g þ K high þ K college ð21Þ

5. Balance budget of government.

GK high þ GK college þ g ¼ τY ð22Þ

6. The distribution of young and old family satisfy:

Z n

  o n o

^Þ ¼
μ o ðho ; π; π ^  I η  hy ¼ ho dμ h hy;π
I G e hy ; π ; π; g ¼ π ð23Þ
hy;π

 Z n o
0 0  
μ y hy ; π ¼ ^ Þ  I π
½1−sðho ; π; π ^ ¼ h0y  prob π0 jπ dμ o ðho ; π; π
^ Þþ
^
h ;;π;π
Z o
hn o n o i ð24Þ
 
^ ¼ h0y pðπ
^ Þ I qcollege π
sðho ; π; π ^ ¼ h0y ð1−pðπ
^ Þ þ I qhigh π ^ ÞÞ prob π 0 jπ dμ o ðho ; π; π
^Þ
^
ho ;π;π

7. Some additional key variables in the model:

(1) Wage premium for senior high school graduates:

Z ,Z
^ Þwqhigh ð1−pðπ
sðho ; π; π ^ ÞÞπ
^ dμ o ðho ; π; π
^Þ ^ ÞÞwπ
ð1−sðho ; π; π ^ dμ o ðho ; π; π
^Þ
^
ho ;π;π ^
ho ;π;π
premhigh ¼ Z Z ð25Þ
^ Þð1−pðπ
sðho ; π; π ^ ÞÞdμ o ðho ; π; π
^Þ ^ ÞÞdμ o ðho ; π; π
ð1−sðho ; π; π ^Þ
^
ho ;π;π ^
ho ;π;π

(2) Wage premium for college graduates:

Z ,Z
^ Þwqcollege pðπ
sðho ; π; π ^ Þπ
^ dμ o ðho ; π; π
^Þ ^ ÞÞwπ
ð1−sðho ; π; π ^ dμ o ðho ; π; π
^Þ
^
ho ;π;π ^
ho ;π;π
premcollege ¼ Z Z ð26Þ
^ Þpðπ
sðho ; π; π ^ Þdμ o ðho ; π; π
^Þ ^ ÞÞdμ o ðho ; π; π
ð1−sðho ; π; π ^Þ
^
ho ;π;π ^
ho ;π;π

(3) As an indicator of cross-section inequality, we define the standard deviation of logged young parents' wage income, as:

2 2 32 312
Z
 Z


6 6 7 7
σ hy ¼ 4 4 log hy − log hy dμ h hy ; π 5 5 ð27Þ
hy ;π hy ;π
116 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125

(4) As an indicator of intergeneration correlation, we define correlation coefficient of wage income between old parents and
their (old) children, ρincome, as:


cov logðho Þ; log h0y
ρincome ¼

 ð28Þ
stdð logðho ÞÞstd log h0y

Before we move on to the next section, three things are worth to mention. First, literally innate ability is defined as ability
determined by genetic factors, however, because we focus only on education in the model, it is worth to be notice that the innate
ability π may not only contain genes, but also cover those factors that reflect non-educational investments. Thus ρ and σπ may
also reflect the correlation of factors such as family background, health status, social networks, and other factors that are not
related to educational investments but that may contribute to wage inequality. Second, the above model implicates children
who enrol in college will not be dropped out. As a matter of fact that college drop-out rate in China is very low (approximately
1%), we think this is an acceptable assumption; third, following Hanushek and Welch (2006) we define wage premium for senior
high school and college graduates as the ratio of mean wage income between these two groups and non-high-school attendance
group (See Eqs. (25) and (26). Because of selection bias, wage premium (premhigh and premcollege) does not always equal the re-
turn of education (qhigh and qcollege).2

4. Calibration and model performance

In Section 3, we analyse the mechanism of human capital accumulation, transmission and the role of government expenditure.
However, in such a model with heterogeneity, we cannot reach an analytical solution. To quantitatively asset the behaviour of this
model and finally conduct the policy experiments, in this section, we calibration parameters in the model and test model's
performance.
Each period in the model corresponds to 15 years, so individual get their compulsory education in childhood and non-
compulsory education during juvenile period. For family's utility function u(c), we assume it takes a constant relative risk aversion
(CRRA) form:

c1−σ −1
uðcÞ ¼ ð29Þ
1−σ

where σ denotes the elasticity of inter-temporal substitution. Following Restuccia and Urrutia (2004), we set σ equals to 1.5.
Some parameters can be matched with real data: we set nhigh = 3/15 and ncollege = 4/15 to match their respective years of educa-
tion in senior high school and college; as η represents old parents' mark-up of human capital that they accumulate from working
experience, we set η equals to 1.09 by calculating the ratio of wage mean of males between 46 and 60 and 30–45 from Chinese
Household Income Project 2002 and 2007.3 Under the assumption of balanced government finances, τ represents the proportion
of total education expenditures over GDP(including the total compulsory education expenditures g and total post-compulsory ed-
ucation expenditures F). Based on the mean value of total education expenditures over GDP between 1990 and 1995,4 τ is set to
3.03%. We unified production technology A as well as wage rates w to 1 and set the family subjective discount factor β to 0.395.5
On the other hand, the other 11 parameters (the left column of Table 1) are difficult to calibrate as it is hard to find corre-
sponding values from data of micro-survey or public finance. We calibrate them to match the following observations (target)
for China6:
(1) Mean value of senior high school enrolment rate during 1990–1995 is 53% (Education Statistical Yearbook, 1996).
(2) Mean value of college enrolment rate during 1990–1995 is 42% (Education Statistical Yearbook, 1996).
(3) Mean value of government compulsory education expenditures in GDP was 1.9% during 1990–1995 (Education Expenditure
Statistics Yearbook and China Statistics Yearbook, 1991–1996).
(4) The average government senior high school expenditures' share of GDP was 0.56% during 1990–1995 (Education Expendi-
ture Statistics Yearbook and China Statistics Yearbook, 1991–1996).
(5) The average government higher education expenditures' share of GDP was 0.57% during 1990–1995 (Education Expendi-
ture Statistics Yearbook and China Statistics Yearbook, 1991–1996).
(6) The government's share of senior high school expenditures was 71% during 1990–1995 (Education Expenditure Statistics
Yearbook and China Statistics Yearbook, 1991–1996).

2
See Heckman and Li (2004), Heckman and Vytlacil (2000, 2001) for more detail.
3
The value in CHIP 2002 is 1.13 and the value in CHIP 2007 is 1.05; we take the average.
4
In order to match the cohorts with wages in CHIP 2002 and 2007, we did not use the latest education expenditures and school enrolment data, but selected the
average values between 1990 and 1995. Junior high school or senior high school graduates at the beginning of 1990s were around 30 years old in 2002 and 2007. Their
wages could reflect the rate of return on the corresponding qualifications.
5
Corresponding to a 0.95 of annual discount rate.
6
Find more details in Appendices A and B.
 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125 117

Table 1
Model parameters.

Parameter Value Description

μ0 0.292 Probability of enrolling in college

μ1 0.978 Probability of enrolling in college
bhigh 0.366 Parameter in government senior high school subsidy function
bcollege 0.253 Parameter in government college subsidy function
fhigh 0.153 Education cost in senior high school
fcollege 0.225 Education cost in college
qhigh 1.018 Rate of return for senior high school
qcollege 1.498 Rate of return for college
nhigh 0.2 Time spend in senior high school
ncollege 0.2667 Time spend in college
γ 0.388 Elasticity of acquired ability to education investment
τ 0.0303 Tax rate
β 0.395 Subjective discount factors
σ 1.5 Elasticity of inter-temporal substitution
η 1.09 Human capital mark-up from work experience
A 1 Technology
σπ 0.53 Variance of innovation in AR(1) process of innate ability
ρ 0.208 Correlation coefficient of innate ability

(7) The government's share of higher education expenditures was 83% during 1990–1995 (Education Expenditure Statistics
Yearbook and China Statistics Yearbook, 1991–1996).
(8) The wage premium for senior high school graduates compared to junior high school graduates is premhigh = 1.19, based on
calculations from CHIP (2002, 2007).
(9) The wage premium for college graduates compared to senior high school graduates is premcollege = 2.14, based on calcula-
tions from CHIP (2002, 2007).
(10) The standard deviation of the logarithm of young parents' wages is σhy = 0.69, measured as the standard deviation of the
labour force aged 31–45 years, according to CHIP (2002, 2007).
(11) The correlation coefficient of wage for parents and children is ρincome = 0.477
Table 1 lists all of the calibrated parameters. The elasticity of acquired human capital to private investment made during
childhood is γ = 0.294, which is consistent with Haley's (1976) estimate; μ0 = 0.292 and μ1 = 0.978 show the probability
of college enrolment is a concave function with respect to acquired ability; qhigh and qcollege take the values of 1.018 and
1.498, implying a relatively low return on senior high school education but a much higher return on college education.

The model parameters and their values are listed in Table 1 while Table 2 reports model's performance compared to the real
data. Generally speaking, the results are quite good. All of the indicators generated by the model have values similar to the real
data. Although senior high school and college wage premiums generated by the model are slightly higher than the real data, we
think the result is acceptable because real data is computed from micro-data.
In addition to the first moment indicators shown in Table 2, we can also examine model's performance based on the second
moment indicators. Fig. 1 shows the average enrolment rates for senior high school and college within five income groups ranking
by parent's income level from left to right, where Fig. 1A represents model's results and Fig. 1B is calculated from real data.
Though senior high school and college enrolment rates from the model are lower for the lowest-income families and higher
for the wealthiest families than in the survey data, the model shows a similar increasing trend of senior high school/college
enrolment rate with parents' income increases as real data does.
Though the model behaves fairly well qualitatively, the level between the two differs markedly. Compare to the result calcu-
lated from real data, the enrolment rate for senior high school and college is lower for children from low income group and much
higher for those in high income group. According to these may partly arise from the absence of credit market. Because the rate of
return for senior high school is relatively low(1.018 in our calibration), the expected return for sending child to high school de-
pends crucially on the probability of college enrolment, which is determined by the acquired ability π ^ at the end of compulsory
education. The absence of credit market means that child from low income family may not get enough education investment,
even for those who born with high innate ability.

5. Sources of income inequality and intergenerational income mobility

In this section, we use the calibrated model to study further the sources of income inequality and intergenerational income
mobility. Having found in Section 4, both the correlation coefficient and the standard deviation of innate ability are smaller
than those of income, we want to know further that during which education stage that human capital inequality and mobility
are enlarged. Table 3 reports the correlation coefficient and standard deviation of the logarithm of innate ability, π, acquired

7
For detail please refer to Eq. (28).
118 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125

Table 2
Model performance.

Indicator Real data Model

College enrolment rate 0.53 0.52

Senior high school enrolment rate 0.42 0.43
Proportion of compulsory education expenditures over GDP 0.019 0.019
Proportion of senior high school expenditures over GDP 0.0056 0.0059
College expenditures over GDP 0.0057 0.0053
Public spending in total senior high school expenditure 0.71 0.72
Public spending in total college expenditure 0.83 0.78
Wage premium of senior high school graduates 1.09 1.01
Wage premium of college graduates 2.14 2.11
The standard deviation of young parents log wages. 0.69 0.7
Intergenerational wage mobility 0.47 0.48

Real Data Source: Education Statistical Yearbook(1991–1996) and Chinese Household Income Project (2002,2007).

ability (after compulsory education), π ^ and final human capital level, hy′. These measures represent the human capital level at three
stages of life (child, juvenile, young parent).8
First, we find that innate ability is an important source of income inequality. It explains approximately 76.8% (0.53/0.7) of the
cross-sectional disparity in young parents' human capital level and 48.4% (0.208/0.48) of the intergenerational correlation. Sec-
ondly, the standard deviation of the logarithm of hy′ is 32% higher than that of the logarithm of innate ability, of which 76%9
has finished after compulsory education. Third, the correlation coefficient of π ^ is almost the same as that of hy′, which indicates
that much of the enlargement of inequality within and across generations is completed by the end of compulsory education. In
other words, in addition to innate ability, education received during the compulsory stage is the most important factor in income
inequality enlargement both within and across generations.
The above analysis reveals the process by which human capital is accumulated in stages; however, the process may differ for
people from different backgrounds. Fig. 2 reports the mean of innate ability, acquired ability and final income for children from
five different groups based on their parents' income level. For convenience of comparison, we normalize the levels of these
three variables to equal one for the lowest-income group. It is clear that there is little difference among the groups with regard
to innate ability: on average, a child from one of the wealthiest families is only 1.36 times ‘smarter’ than a child from one of the
poorest families. However, the gap in human capital level increases to 2.35 at the end of compulsory education and 2.89 at the
end of non-compulsory education. This indicates the importance of family background and family educational choices in deter-
mining the inequality of final human capital level and, therefore, income.
Given the government expenditures on compulsory and non-compulsory education, acquired ability, π ^ and final human capital
level, hy′ are determined mainly by the family's private education expenditure and senior high school attendance decision. For this
reason, in Table 4 we tabulate and calculate their mean value by different innate ability and parents' income level.
Table 4A reports private human capital investment during compulsory education for children with various levels of innate abil-
ity and family background. For the same level of children's innate ability, it indicates that human capital investment increases
with parents' income. For those families whose parents earn the most, early education investment increases with children's innate
ability. However, although children with the highest ability still get the most compare to others within the same income group,
children who come from lowest income group receive more that those in medium income group. This finding is in accordance
with Restuccia and Urrutia (2004).
Table 4B and C indicates the same trend as Table 4A: within the same innate ability group, children's acquired ability and se-
nior high school enrolment rate increase with parent's income; given the family background, children with higher innate ability
always achieve higher acquired ability and have a better chance to receive better education.

6. Model behaviour under alternative parameters

One important contribution of this paper is that we analyse the effect of nature versus nurture human capital on inequality
and intergenerational mobility of income. In this section, we examine the behaviour of the model to some key parameters in
the model. In the interest of saving space, we focus on two sets of parameters: ρ and σπ, which may affect innate ability; γ
and qcollege, which may affect acquired human capital.
Table 5 shows the main indicator of the model under different values of ρ and σπ. In our benchmark setting, education enlarge
the innate ability disparity and decrease the mobility of between adjacent generations. Thus we could expect the disparity and
correlation of income and acquired ability would increase with ρ and σπ. Indeed, we find that when ρ increases from 0.208
(benchmark) to 0.41, cross-sectional disparity and intergenerational persistence of income are greater. The influence of human
capital also is even greater. On the other hand, if ρ reduce to 0.01, the income persistence across generations decreases, but

8
In the model, old parents' human capital mark-up is set to be an exogenous mark-up of young parents' human capital level, determined by the parameter. Therefore,
listing the corresponding statistics for ho is unnecessary.
9
Because the standard deviation of the logarithm of π^̂ is 24.5% higher than, this number is calculated as 24.5%/32%.
 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125 119

Fig. 1. Enrolment rates of senior high school and college(%). Source: One-percent National Sample Census (2005).

the cross-sectional income gap shows no significant change. The cross-sectional disparity does not change, because if ρ is compar-
atively low, the probability of poor families having innately intelligent children is higher, but these children face strong budget
constraints.
Decreasing the standard deviation of σπ reduces cross-sectional income inequality but increases the disparity across genera-
tions. If σπ is reduced from the benchmark model level (0.53) to 0.33, cross-sectional inequality decreases from 0.69 to 0.47,
while intergenerational persistence increases from 0.48 to 0.56. This is because high-ability individuals will attract more educa-
tional investment from their parents, especially during the non-compulsory education period. If the number of high-ability
individuals decreases, the gaps in education investment–and thus the future income of the grown children–across income cohorts
decrease. Intergenerational persistence increases, because poor families may reduce their educational investment, especially their
post-secondary education investment, because there are fewer high-ability children. The parameters ρ and σπalso have significant
effects on the senior high school and college enrolment rates. These results show that the parameters must be calibrated based on
real data; otherwise, the analysis is not valid.
Another two important parameters that decide family's private education investment is γ and qcollege. They describe the impor-
tance of compulsory education and higher education to income. Table 6 shows the model behaviour for different value of these
two parameters. If γ reduces from the benchmark model level (0.388) to 0.188, the contribution of compulsory education to in-
tergenerational income persistence declines while cross-sectional income disparity increases. The effects of return on non-
compulsory education on cross-sectional income disparity and intergenerational income mobility are both augmented. This result
suggests that if γ decreases, parents decrease their investment in compulsory education and increase their investment in non-
compulsory education. On the contrary, when γ raise to 0.488, the contribution of compulsory education to cross-sectional income
disparity and intergenerational income mobility are both enhanced, but the effects of higher education shrink.
Wage premiums are highly correlated to the senior high school enrolment rate. If qcollege raise from 1.498 to 1.598, the senior
high school enrolment rate increases by approximately 30%, while college progression rate is unchanged. Table 6 also shows that
increasing qcollege strengthens the contribution of compulsory education to income inequality and reduces the explanatory power
of higher education with respect to income persistence across generations. If qcollege reduces to 1.398, the contributions of compul-
sory education and higher education to intergenerational income mobility do not experience significant changes, but the influence
of higher education on income disparity declines. In addition, the senior high school and college enrolment rates also decline. The
results reported in Tables 5 and 6 may also provide a traceable framework to compare the role of education related indicator in
income disparity across countries (Björklund & Jäntti, 1997).

Table 3
Decomposition of income inequality and intergenerational income mobility.

Innate ability (π) ^)

Acquired ability (π Human capital (h′)
y

Cross-sectional disparity (Standard deviation of log) 0.53 • 0.66 • 0.70

Intergenerational correlation (Correlation coefficient) 0.208 • 0.47 • 0.48

Note: The boundary for cross-sectional disparity is [0,+∞] and [0,1] for intergenerational correlation.
120 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125

Fig. 2. Decomposition of income inequality for various income groups.

The level of ability and income of each group reported in Fig. 2 is their relative size compare to those in the lowest income level.

7. Policy experiments

In this section, we focus on the design of public education investment policy based on the calibrated model. We assume that
the government education expenditures' share of GDP increase to 3.15% from the original 3%, and observe the changes in enrol-
ment rates and income inequality for various strategies of allocating education expenditures among the stage of compulsory,
senior high school and college education. We show all the result in Table 7 and explain each policy experiment as follows:

Policy experiment 1: The entire incremental expenditure is invested in compulsory education, for all children, and other alloca-
tion strategies are unchanged.
Policy experiment 2: The entire incremental government education expenditure is invested in compulsory education, but
targeted for children from low-income families. In contrast, in the benchmark model, education expenditures are equal for
all children in a given grade level. In this experiment, the allocation strategy for compulsory education is changed such that

Table 4
Main education indicator within various levels of innate ability and family background.

Parents' income

Child's innate ability Low Middle High

A. Compulsory-education investment
Low 0.0005 0.0043 0.0395
Middle 0.0001 0.0013 0.0525
High 0.0056 0.0096 0.0849

B. Child's human capital at the end of compulsory education

Low 0.18409 0.19014 0.21968
Middle 0.30914 0.31076 0.36706
High 0.68476 0.72090 0.91217

C. Senior high school enrolment rate(%)

Low 5.72 19.65 76.68
Middle 24.66 54.27 72.28
High 58.77 100.00 100.00

Note: For Table 4A and B, the numerical value solely is meaningless since it may vary according to computational method. However, it is meaningful to compare
the level between different parents income and ability group.
 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125 121

Table 5
Model behaviour for different ρ and σπ.

Benchmark Set1 Set2 Set3 Set4

ρ 0.208 0.008 0.408 0.208 0.208

σπ 0.53 0.53 0.53 0.33 0.73

Intergenerational correlation
Innate ability 0.208 0.008 0.408 0.208 0.208
Acquired ability 0.45 0.327 0.585 0.483 0.391
Income 0.48 0.385 0.625 0.559 0.421

Cross-sectional variance
Innate ability 0.53 0.53 0.53 0.33 0.73
Acquired ability 0.592 0.624 0.734 0.43 0.876
Income 0.689 0.666 0.803 0.469 0.937

Other variables: enrolment rate

Senior high school 52% 60.7% 56% 54% 66%
College 43% 42% 45% 40% 46%

Note: The boundary for cross-sectional disparity is [0,+∞] and [0,1] for intergenerational correlation.

the government invests more in children from low-income families. Specifically, the government investment function is
changed as follows:

^ ¼ Gðπ; e; g Þ ¼ πðkgðhÞe þ g Þγ ; 0 b γ b 1
π ð30Þ

in which kg is the subsidy function that satisfies the following constraint:

(



Tkg::::::F hy b 20%
kg hy ¼ ð31Þ
0::::::::Otherwise

in which F is a cumulative function of ability. The government expenditures on compulsory education under this type of tax rate
can be written as follows:
Z h
i

g þ kg hy f hy dhy : ð32Þ

Eqs. (30) through (32) indicate that low-income families (representing the bottom quantile of families, based on income)
whose children benefit from 1 unit of addition government education expenditure obtain τkg units of subsidy. We calibrate the
model parameters by assuming that government expenditures on senior high school and higher education, as shares of GDP,
are unchanged and that all of the incremental government education expenditure is devoted to compulsory education for children
from low-income families.

Table 6
Model behaviour for different γ and qcollege.

Benchmark Set1 Set2 Set3 Set4

γ 0.388 0.188 0.488 0.388 0.388

qcollege 1.498 1.498 1.498 1.398 1.598

Intergenerational correlation
Innate ability 0.208 0.208 0.208 0.208 0.208
Acquired ability 0.45 0.269 0.522 0.426 0.509
Income 0.48 0.328 0.565 0.456 0.543

Cross-sectional variance
Innate ability 0.53 0.53 0.53 0.53 0.53
Acquired ability 0.592 0.566 0.712 0.639 0.684
Income 0.7 0.645 0.752 0.662 0.766

Other variables: enrolment rate

Senior high school 52% 69% 49% 27% 86%
College 43% 52% 39% 43% 43%

Note: The boundary for cross-sectional disparity is [0,+∞] and [0,1] for intergenerational correlation.
122 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125

Table 7
Decomposition of income gaps across sections and generations.

Policy Indicator Innate ability Acquired ability Income High school enrolment rate College enrolment rate

Bechmark σπ 0.53 0.66 0.7 0.528 0.421

ρ 0.285 0.308 0.354
Policy experiment 1 σπ 0.53 0.7 0.72 0.537 0.424
ρ 0.288 0.308 0.353
Policy experiment 2 σπ 0.53 0.68 0.69 0.543 0.421
ρ 0.288 0.25 0.26
Policy experiment 3 σπ 0.53 0.7 0.72 0.527 0.421
ρ 0.288 0.311 0.349
Policy experiment 4 σπ 0.53 0.69 0.72 0.528 0.421
ρ 0.288 0.311 0.35

Note: Policy experiment 1: Increase public educational expenditure in compulsory education and target all children. Policy experiment 2: Increase public educa-
tional expenditure in compulsory education and only target children from low-income families. Policy experiment 3: Increase public educational investment in
post-compulsory education and only target children from low-income families. Policy experiment 4: Increase public educational expenditure on post-compulsory
education and only target children from low-income families so as their children were admitted to a university.

Policy experiment 3: The incremental government education expenditure is devoted to juveniles from low-income families who
attend senior high school. That is, as long as low-income parents send their juveniles to senior high school, they receive the
corresponding subsidy regardless of whether their juveniles attend college. Because the government senior high school expen-
ditures' share of GDP in the benchmark model is rather low, we assume that the increased subsidy effectively increases low-
income families' incomes, thereby mitigating their budget constraints, as long as they persuade their juveniles to attend senior
high school.
The budget constraint for old parents becomes the following:




0 0 0
co þ f high nhigh þ θ  f college ncollege ¼ ð1−τÞfwho þ why ½θ 1−nhigh −ncollege þ ð1−θÞ 1−nhigh g þ kg hy  w  hy ð33Þ

in which kg represents the subsidy for low-income families that send their juveniles to senior high school; its exact specification is
similar to Eq. (31) but for old families.

Policy experiment 4: The incremental government education expenditure is devoted to higher education of juveniles from low-
income families. As with the senior high school subsidy, this college subsidy is obtained only if the parents send their children
to senior high school and their children are admitted to a university. Because government higher education expenditures'
share of GDP is rather low, the increased expenditure effectively subsidizes low-income families' income. That is, the new bud-
get constraint for old parents is Eq. (33), and kg satisfies Eq. (31).
We find that if the incremental government education expenditure is consumed in grades 1–9, the effect of early education on
cross-sectional income disparity increases significantly (from 0.66 to 0.70) and may slightly reduce income persistence across gen-
erations but increases cross-sectional income gaps to some extent. The senior high school progression rate increases slightly, but
the college progression rate is unchanged. However, if the incremental government expenditure on compulsory education is de-
voted to children from low-income families (Policy experiment 2), the cross-sectional income gaps change little and the intergen-
erational income persistence significantly decreases (the intergenerational correlation shrinks from 0.354 to 0.260). In addition,
the senior high school progression rate rises significantly. The results of Policy experiment 3 show that investing all of the incre-
mental government education expenditure in senior high school attendance of children from low-income families slightly reduces
intergenerational income mobility but sharply reduces income inequality. Furthermore, its effects on the senior high school and
college progression rates are limited. The effects of Policy experiment 4 are similar to those of Policy experiment 3: They slightly
reduce intergenerational income correlation but increase cross-sectional income gaps.
By comparing four types of policy experiments—increasing government expenditures on compulsory education for all children,
increasing government expenditures on compulsory education for children from low-income families, increasing government ex-
penditures on senior high school for children from low-income families, and increasing government expenditures on higher ed-
ucation for children from low-income families—we find that the most effective and efficient way to reduce income inequality is to
subsidize compulsory education for children from low-income families (Policy experiment 2). This subsidy would enhance low-
income families' educational investment in their children's compulsory education and increase the quality of that education,
which would likely result in their children receiving higher levels of schooling and, therefore, earning higher wages as adults.
Such a subsidy would significantly relax intergenerational income mobility.

8. Concluding remarks

In this paper, we have analysed the roles of innate ability, compulsory education, and non-compulsory education with respect
to cross-sectional income gaps and intergenerational income persistence, by constructing a four-period overlapping-generations
 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125 123

equilibrium model and using it to conduct policy experiments. We have found that, in addition to exogenous factors such as in-
nate ability, human capital investment is a rather important factor in determining cross-sectional and intergenerational income
inequality. By further decomposing the income inequality, by dividing families into five groups based on income level, we have
found that differential investment in children's early education is the most important driver of cross-sectional income inequality
and intergenerational income persistence. The difference in innate ability between children from families in the top income
quantile and children from families in the bottom income quantile is approximately 1.36, but that gap increases to 2.35 at the
end of compulsory education and further increases to 2.89 at the end of higher education. Because high-income families invest
more heavily in their children's early education, their children receive more schooling than do children with the same innate abil-
ity who are from low-income families. Therefore, this gap in early education investment results in greater income inequality and
less intergenerational income mobility than do differences in innate ability. Among different subsidy policy, we found that the
most effective and efficient way to reduce income inequality is to subsidize low-income families' early-education investment,
which could mitigate young parents' budgetary concerns.
Though this paper emphasises the role of education, we should also be aware that the mechanism in forming income inequal-
ity is too complex to explain by only one factor like education. In one aspect, education could affect income inequality through
other mechanism, such as marriage matching, health, children's quantity, etc.; on the other hand, other mechanism will affect ed-
ucation as well. For example, parents' nutrition will affect children's health and children’ health will further affect their education
achievement and labour productive. In other words, health may be another important factor to income inequality. Future work
may include health in innate ability to further develop the model.
Finally, throughout this paper we limit ourselves to steady state analysis. This is due to both computational costs and the fact
that we see understanding steady state inequality as a necessary step that comes before studying the transitions and evolution of
inequality over time. However, as many parameter we take as constant(such as rate of return to education) in the model have
been changing over time in China and will in all likelihood continue to do so, it is more interesting to see how will these change
affect income inequality and intergenerational income mobility from a dynamic perspective. Further work requires breakthrough
both in theory and computation method.

Acknowledgement

The research was supported by Nature Science foundation of China, project No. 71203014 and the fundamental research funds
for the central universities.

Appendix A. Model-Solving Process

A.1. Initialization

1.1 Use a 15-states Markov process to simulate Eq. (18), denoting the minimum and maximum values of π as πmin and πmax,
respectively.
^.
1.2 Set the state space of the young family's and old family's value functions as discrete. The states for old family are ho ; π; π
a. Based on Eq. (6) for acquired ability, set the minimum and maximum values of π ^ as G(πmin, 0, g) and G(πmax, 0, 5 ⋅ g),
respectively.
b. Set the minimum and maximum values of hy as G(πmin, 0,g) and G(πmax, 0,5 ⋅ g), respectively.
Given these parameters, the process of solving the model can be explained as the following three steps:

A.2. Value function iteration

2.1 Set an initial value for the government's compulsory education expenditure, g0.
^ Þ.
2.2 Set an initial value for the old family's value function, vo ðho ; π; π
^ Þ, do the following:
2.3 Value function iteration: For the ith guess of gj and the jth guess of voj ðho ; π; π
a. Obtain the young family's value function,vjy+1(hy, π), by solving Eqs. (1) through (5).
b. Update the old family's value function, vojþ1 ðho ; π; π ^ Þ, by solving Eqs. (8) through (14).
^ Þ and vojþ1 ðho ; π; π
c. Compare voj ðho ; π; π ^ Þ. If max jjVecðvoj ðho ; π; π
^ Þ−vojþ1 ðho ; π; π
^ ÞÞjjbε; ε ¼ 0:01, stop the iteration; if not, pro-
ceed to step 3.

A.3. Calculate the invariant distribution

3.1 Guess the initial distribution of the old family,μo. Choose a uniform distribution for all states.
3.2 For the kth iteration, for the invariant distribution, μko calculate the distribution of the young family, μk+1 y based on Eq. (25).
3.3 Update μk+1
o based on Eq. (24) and μk+1y .
3.4 Compare μk+1
o and μko until they converge. If max ||Vec(μk+1 o − μko)||b ε , ε = 0.01, stop the iteration; if not, return to step
3.2.
124 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125

A.4. Update gi+1

Based on the invariant distribution, calculate total output; government expenditure on compulsory, senior high school and col-
lege education; and private expenditure on compulsory, senior high school, and college education. Then update the government
expenditure on compulsory education, gi+1, consistent with the government's balanced budget, as follows:

GK high þ GK college þ g ¼ F þ g ¼ τY ð1Þ

Return to step 2, and continue iterations until |gi+1 − gi | b 0.0001.

Appendix B. Calibration process

We use a Modified Newton Method to calibration the parameters in the model, and all of the calculations are written in
Fortran 90 language.
Let x = [μ0, μ1, γ, bhigh, bcollege, fhigh, fcollege, qhigh, qcollege, σπ, ρ]′ ∈ X ⊂ R11, in which R11 denotes the 11 parameters that must be cali-
brated in the model. Given x, denotes the vector that contains the 11 results of the model that are listed in Table 1. Thus, we
can consider the whole model as a function that takes x as input and generates y as output. Let yreal ∈ Y ⊂R11 be the corresponding
indicator calculated from real data. The calibration process is to solve the following non-linear function:

real real
GðxÞ ¼ y −y ¼ y − F ðxÞ ¼ 0 ð1Þ

Let x⁎ be the solution to the above equation. We can impose the following first-order Taylor expansion of G(x)
around x⁎:
   
GðxÞ≈G x þ J x ðx−xÞ ð2Þ

Following the approach of Heer and Maussner (2009), the numerical solution can be reached by the following iterative
procedure:

sþ1 s  s −1  s 
x ¼ x −J x G x ð3Þ

in which xs is the sth input vector of the model, xs+1 is the s + 1th input vector, and J(⋅) is the Jacobian matrix of G(⋅).
In order to speed up the programme, we use a Modified Broyden method to update the Jacobian matrix, J(⋅) (see Heer &
Maussner, 2009). We let ‖⋅‖ be the module of 11-dimensional Euclidean space and stop the iteration when ‖G(x)‖ b ε , ε = 0.05.
In order to reach a global solution for the parameters, we set their initial values as follows: The parameter qcollege is set to
1.403, as estimated by Heckman and Li (2004), ρ = 0.32 as estimated by Black, Devereux, and Salvanes (2009) based on US
data, σπ = 0.49 and γ = 0.32 as assumed by Restuccia and Urrutia (2004). For the initial value of the remaining parameters,
reliable estimates are difficult to obtain; therefore, we try various initial values in a relatively loose interval and find that the
result is robust.

References

Bai, X. (2004). Education and income inequality: An empirical study from China. Management World, N0., 6 (In Chinese Edition).
Becker, G. S., & Tomes, N. (1979). “An equilibrium theory of the distribution of income and intergenerational mobility”. The Journal of Political Economy, 1153–1189.
Björklund, A., & Jäntti, M. (1997). “Intergenerational income mobility in Sweden compared to the United States”. The American Economic Review, 1009–1018.
Björklund, A., Lindahl, M., & Plug, E. (2006). The origins of intergenerational associations: Lessons from swedish adoption data. The Quarterly Journal of Economics,
999–1028.
Black, S. E., Devereux, P. J., & Salvanes, K. G. (2009). “Like father, like son? A note on the intergenerational transmission of IQ scores”. Economics Letters, 105(1),
138–140.
Carneiro, P., & Heckman, J. J. (2002). The evidence on credit constraints in post‐secondary schooling. The Economic Journal, 112(482), 705–734.
Checchi, D., Ichino, A., & Rustichini, A. (1999). More equal but less mobile?: Education financing and intergenerational mobility in Italy and in the US”. Journal of Public
Economics, 74(3), 351–393.
Deng, Q., Gustafsson, B., & Li, S. (2013). Intergenerational income persistence in urban China”. Review of Income and Wealth, 59(3), 416–436.
Eriksson, T., & Zhang, Y. (2012). The role of family background for earnings in rural China. Frontiers of Economics in China, vol. 7(3). (pp. 465–477). Higher Education
Press.
Gong, H., Leigh, A., & Meng, X. (2012). Intergenerational income mobility in urban China. Review of Income and Wealth, 58(3), 481–503.
Haley, W. J. (1976). Estimation of the earnings profile from optimal human capital accumulation. Econometrica: Journal of the Econometric Society, 1223–1238.
Hanushek, E. A., & Welch, F. (2006). Handbook of the economics of education. Vol. 1, (ERIC).
Heckman, J. J., & Li, X. (2004). Selection bias, comparative advantage and heterogeneous returns to education: Evidence from China in 2000. Pacific Economic Review,
9(3), 155–171.
Heckman, J. J., & Vytlacil, E. J. (2000). The relationship between treatment parameters within a latent variable framework. Economics letters, 66(1), 33–39.
Heckman, J. J., & Vytlacil, E. J. (2001). Identifying the role of cognitive ability in explaining the level of and change in the return to schooling. Review of Economics and
Statistics, 83(1), 1–12.
Heer, B., & Maussner, A. (2009). Dynamic general equilibrium modeling: Computational methods and applications[M]. Springer Science& Business Media.
Hendel, I., Shapiro, J., & Willen, P. (2005). Educational opportunity and income inequality. Journal of Public Economics, 89(5), 841–870.
Lai, D. (2004). “The expansion of education and income inequality”. Economic Research Journal, No., 10 (In Chinese Edition).
Li, S., & Luo, C. (2011). How unequal is China? Economic Research Journal, 4.
 J. Yang, M. Qiu / China Economic Review 37 (2016) 110–125 125

Maoz, Y. D., & Moav, O. (1999). Intergenerational mobility and the process of development. The Economic Journal, 109(458), 677–697.
Meng, X., Shen, K., & Xue, S. (2013). Economic reform, education expansion, and earnings inequality for urban males in China, 1988–2009. Journal of Comparative
Economics, 41(1), 227–244.
Mincer, J. (1974). Schooling, experience, and earnings. Human Behavior & Social Institutions No., 2.
Nakamura, T., & Murayama, Y. (2011). Education cost, intergenerational mobility, and income inequality. Economics Letters, 112(3), 266–269.
Plug, E. (2004). Estimating the effect of mother's schooling on children's schooling using a sample of adoptees. American Economic Review, 358-68.
Ravallion, M., & Chen, S. (2007). China's (Uneven) progress against poverty. Journal of Development Economics, 82(1), 1–42.
Restuccia, D., & Urrutia, C. (2004). Intergenerational persistence of earnings: The role of early and college education. American Economic Review, 1354–1378.
Solon, G. (2004). A model of intergenerational mobility variation over time and place. Generational income mobility in North America and Europe (pp. 38–47).
Sylwester, K. (2002). Can education expenditures reduce income inequality? Economics of Education Review, 21(1), 43–52.
Zhu, G., & Vural, G. (2013). Inter-generational effect of parental time and its policy implications. Journal of Economic Dynamics and Control, 37(9), 1833–1851.

View publication stats