Migration Networks, Export Shocks, and Human Capital Acquisition: Evidence from China

Abstract

We examine the impact of export shocks proxied by destination countries’ tariffs on the postmiddle school enrollment of the rural population in China. We complement the literature by examining across-region spillover effects of export shocks through initial migration networks. We find that the reduction of export tariffs at both the local and migration-destination prefectures significantly decreases postmiddle school enrollment of the 16-18 years old cohort, but the latter is stronger. Further analysis suggests that employment in secondary industry rise significantly with the reduction of export tariffs, which improves job opportunities and thus increases the postmiddle school dropout rate.

Full text

1
Migration Networks, Export Shocks, and Human Capital Acquisition:
Evidence from China
Shu Cai, Xinzheng Shi, and Zhufeng Xu*
May 2022
Abstract
We examine the impact of export shocks proxied by destination countries’ tariffs on the
postmiddle school enrollment of the rural population in China. We complement the literature
by examining across-region spillover effects of export shocks through initial migration
networks. We find that the reduction of export tariffs at both the local and
migration-destination prefectures significantly decreases postmiddle school enrollment of the
16-18 years old cohort, but the latter is stronger. Further analysis suggests that employment in
secondary industry rise significantly with the reduction of export tariffs, which improves job
opportunities and thus increases the postmiddle school dropout rate.
JEL Codes: F16, J24, O12, O14
Keywords: Tariff shocks, Educational choice, High school enrollment, Migration networks
* Shu Cai: Associate Professor in Jinan University, email: shucai.ccer@gmail.com; Xinzheng
Shi: Associate Professor in Tsinghua University, email: shixzh@sem.tsinghua.edu.cn;
Zhufeng Xu: Assistant Professor in the School of Economics in Central University of Finance
and Economics, email: zhufeng@zhufengxu.com. We acknowledge helpful comments from
participants in different conferences and seminars. Xinzheng Shi acknowledges financial
support from China National Natural Science Foundation (Project 71673155) and Tsinghua
University Initiative Scientific Research Program (Project 2021THZWJC14). Shu Cai
acknowledges financial support from China National Natural Science Foundation (Project
72173056). All remaining errors are our own.
Electronic copy available at: https://ssrn.com/abstract=4101730

1
1. Introduction
Recent literature has investigated how local export expansion affects
the school attainment of children at their key education age (e.g., Atkin,
2016; Li, 2018).
1
Local export expansion can increase the opportunity costs
of schooling and the returns to schooling, with the former and latter
decreasing and increasing school attainment, respectively. However, when
individuals make decisions about whether to continue their schooling, they
consider not only local economic opportunities but also economic
opportunities in other regions because they can migrate to areas with more
job opportunities. This consideration is particularly true in developing
countries where economic development is unbalanced across regions. This
issue remains unexplored in the current literature. Our study fills this gap by
investigating how export expansion in local and nonlocal regions affects
local children’s educational attainment.
We focus on China from 2000-2015 because China provides a good
context for studying this issue. China has experienced waves of export
expansion. In the 1990s, exports increased fourfold from 62 billion dollars in
1990 to 249 billion dollars in 2000 (China Statistical Yearbook, 1991, 2001).
After China entered the WTO in 2001, exports accelerated, increasing from
249 billion dollars in 2000 to 2,273 billion dollars in 2015 (China Statistical
1
Li et al. (2019) focus on long term impacts and find that export exposure leads to
decreased completed years of schooling, cognitive abilities, wages, and noncognitive
outcomes.
Electronic copy available at: https://ssrn.com/abstract=4101730

2
Yearbook, 2001, 2016). In the meantime, another feature of the Chinese
economy provides additional opportunities for studying this issue. Since the
1990s, the population of migrants moving to work in other cities has grown
rapidly. Given the existence of a migrant network, export expansion in
destination cities affects the educational decisions of individuals in their
hometowns. With the high incomes of migrants in destination cities, their
remittances can ease credit constraints and increase education investment for
their children in their hometowns (Yang, 2008). The migrant network also
makes it easier for children in hometowns to migrate to work when more
work opportunities are brought by export expansion in destination cities (de
Brauw and Giles, 2017). Thus, we examine the impact of export expansion
locally and in other prefectures on the educational attainment of children in
the local prefecture.
The critical difficulty of investigating the impact of export expansion
on education attainment lies in the potential endogeneity problem. For
example, export expansion could occur because of supply-side factors, such
as an increase in infrastructure investment, which may also affect individuals’
education decisions. In this study, we rely on the initial across-prefecture
variation in export structures and the overtime variation in export tariffs to
construct an index of export expansion for each prefecture in each year.
Given that export tariffs reflect demands in foreign countries and initial
export structures are predetermined, our measurement of export expansion is
Electronic copy available at: https://ssrn.com/abstract=4101730

3
likely to be exogenous. We then utilize a weighted average of export
expansion in other prefectures by using the share of migrations from the
local prefecture to other prefectures as weights to investigate the impact of
export expansion in other prefectures.
China has a nine-year compulsory education system, and students’
decision to continue schooling is most likely to be affected by outside shocks
after they finish middle school. Given that we do not have information about
the age of each person finishing middle school, we focus on the cohort
between 16 and 18 years of age, when they are most likely to finish middle
school. We find that export expansion in both local and other prefectures
significantly reduces ordinary and vocational high school attendance (we use
high school to denote ordinary and vocational high school hereafter) of the
cohort between 16 and 18 years old in rural areas, but the effect of export
expansion in other prefectures is stronger. This finding is robust to different
checks. We further investigate channels through which export expansion
affects high school enrollment. We first find that the spillover effect from
prefectures with more workers in the secondary industry is stronger,
suggesting that emigrants of local prefectures are more likely to work in the
secondary industry.
2
We then find that export expansion significantly
increases the employment growth rates of local residents and emigrants in
the secondary industry, suggesting that improving job opportunities is indeed
2
The secondary industry includes industries of manufacturing; construction; mining; and
the production and supply of electricity, heat, gas and water
Electronic copy available at: https://ssrn.com/abstract=4101730

4
the underlying mechanism. We also provide suggestive evidence showing
that the opportunity cost effect of export expansion is stronger than the
income effect.
Our study contributes to the literature in three ways. First, existing
studies mainly focus on the impact of local export expansion. For example,
Atkin (2016) shows that the growth of export manufacturing industries
causes students to drop out of school in grade nine in Mexico. Li (2018) uses
Chinese data and shows that high-skill export shocks increase high school
and college enrollment, but low-skill export shocks decrease both.
3
Li (2018)
also analyzes the impact of export expansion in nonlocal cities on local
education attainment. However, she only considers nonlocal cities in the
sense of geographic distance, but geographically close cities are not
necessarily destinations for migrants. Our study extends the literature by
investigating the impact of export expansion in other prefectures through
migration networks.
Second, our study sheds light on the literature investigating the direct
impact of international or domestic migration on household education
investment in hometowns. For example, Yang (2008) examines how
remittances sent by overseas Filipinos affect educational investment in their
3
Studies have used data from different countries such as the US (Greenland and Lopresti,
2016), India (Oster and Steinberg, 2013; Shastry, 2012, Edmonds, Pavcnik, and Topalova,
2010; Jensen, 2010), Mexico (Majlesi, 2014; Caselli, 2014), Vietnam (Edmonds and
Pavcnik, 2005), and Bangladesh (Heath and Mobarak, 2015). Olney (2013) uses
cross-country data to show that exporting unskilled-intensive goods depresses average
years of schooling, whereas exporting skill-intensive goods increases years of schooling.
Electronic copy available at: https://ssrn.com/abstract=4101730

5
home country. de Brauw and Giles (2017) study how migration affects the
decision of middle school graduates to attend high school in rural China. In
contrast, our study shows that migrant networks can act as an intermediate
channel that transmits the impact of export expansion in other prefectures to
home prefectures, enhancing our understanding of the roles played by
migration.
Third, the literature studying the education issue in developing
countries mostly focuses on supply-side determinants, such as school quality,
teacher attendance, and textbook provision.
4
However, our study shows the
importance of demand-side factors. This has implications for governments
seeking to develop effective educational policies.
The remainder of this paper is organized as follows. Section 2
provides background knowledge on China’s export expansion, migration,
and education system. Section 3 introduces the data and measurements.
Section 4 describes the empirical strategies employed. Section 5 presents the
main results and various robustness checks. Section 6 investigates the
mechanism. Section 7 concludes.
2. Background
2.1. China’s Export Expansion
4
See, for example, Park et al. (2015); Duflo, Dupas, and Kremer (2011); Duflo, Hanna,
and Ryan (2012); and Glewwe, Kremer, and Moulin (2009).
Electronic copy available at: https://ssrn.com/abstract=4101730

6
Before the economic reform of 1978, Chinese trade took place within
a central-planning framework. The State Planning Commission made plans
for almost all of China’s export and import activities, and a few foreign trade
companies controlled by the Ministry of Foreign Trade were responsible for
carrying out export and import plans. In 1977, Chinese trade accounted for
only 0.6% of world trade by volume (Lardy, 1994) with 7.6 billion dollars of
exports and 7.2 billion dollars of imports (China Statistical Yearbook, 1978).
China gradually reformed its trade regime during the 1980s and
1990s. The Chinese government took many steps to provide economic
incentives for exports. Such incentives include a realistic exchange rate, a
rebate of indirect taxes on processed exports, a duty drawback on inputs used
in processed exports, a reduction of barriers to exports, and a growing role of
foreign-funded firms as exporters. These incentives led to an explosion of
Chinese exports, increasing from 7.6 billion dollars in 1977 to 249 billion
dollars in 2000 (China Statistical Yearbook, 1978, 2001).
China joined the WTO at the end of 2001. This represents China’s
other wave of trade liberalization. Chinese exports increased from 249
billion dollars in 2000 to 2,273 billion dollars in 2015 (China Statistical
Yearbook, 2001, 2016). Xu et al. (2020) found that destination countries’
tariffs had significantly negative effects on China’s exports in 2002 and
2013.
Electronic copy available at: https://ssrn.com/abstract=4101730

7
2.2. Precollege Education System in China
China’s education system before college includes primary school (6
years), middle school (3 years), and high school (3 years). Primary and
middle school education are compulsory for all children. Middle school
graduates need to take an entrance examination before they can be admitted
to high schools. A prefecture-wide uniform high school entrance
examination is administered to all middle school graduates by the prefecture
education bureau. To gain admission to ordinary high schools, students need
to achieve examination scores above the cutoff levels set by ordinary high
schools. If students’ test scores are lower than any cutoff, they can attend
vocational high schools, which typically have no cutoff, or they can exit
schooling. In 2000, approximately 29% of middle school graduates were
enrolled in ordinary high schools, and 11% were enrolled in vocational
schools (China Statistical Yearbook, 2001). The former and latter levels
respectively increased to 56% and 34% in 2015 (China Statistical Yearbook,
2016).
2.3. Migration
Although the household registration system, the hukou system, still
restricted labor mobility, a dramatic increase is observed in the number of
migrants in the 1990s.
5
Liang and Ma (2004) show that the number of
5
Hukou is an individual registration of a household registration system adopted in
Electronic copy available at: https://ssrn.com/abstract=4101730

8
intercounty migrants increased from 20 million in 1990 to 79 million in 2000
using a 1% sample of the population census. Cai, Park, and Zhao (2008)
estimate that the number of rural workers migrating to other provinces
exceeded 40 million by 2003. The authors also show that the overwhelming
trend is increasing migration to coastal provinces. Among interprovincial
migrants, the percentages going to eastern provinces from western, central,
and eastern provinces were 68.3%, 84.3%, and 64.4% in 2000, respectively
(Cai, Park, and Zhao, 2008).
Information and referral through earlier migrants have been found in
other countries to be important determinants of potential migration decisions
(e.g., Montgomery, 1991; Carrington, Detragiache and Vishwanath, 1996;
Munshi, 2003). Some studies on the Chinese context present similar findings.
Rozelle et al. (1999) find that villages, where more people migrated out in
1988, experienced rapid growth in migration later in the year. Meng (2000)
shows that the size of current migrants in destinations can explain the
variation in migrants from their hometowns. Zhao (2003) presents evidence
that an individual’s probability of migrating is positively correlated with the
number of earlier migrants from the same village.
3. Data and Measurements
3.1. Data
mainland China. It is divided into agricultural and nonagricultural residency status (often
referred to as rural and urban hukou). See Song (2014) for a more detailed introduction.
Electronic copy available at: https://ssrn.com/abstract=4101730

9
We use different data sources in this paper: (1) we use China
population census data in 2010 and population survey data in 2005 and 2015
to construct prefecture level high school enrollment rate from 2000 to 2015;
(2) we use China population census data in 2000 to construct migration
networks; (3) we combine tariffs imposed on products at the 6-digit level of
the harmonized system (HS) exported from China to other countries from
2000 to 2015 and China custom data in 2000 to construct measurement of
prefecture level export shocks; (4) we obtain prefecture level variables from
statistical yearbooks (2000-2016).
We first rely on data drawn from two waves of population census
conducted in 2000 and 2010 as well as two waves of a 1% population sample
survey conducted in 2005 and 2015. The Chinese government conducted the
first three waves of the population census in 1953, 1964 and 1982. Starting
from the fourth wave of the population census in 1990, the population census
was conducted every 10 years, with the two waves in 2000 and 2010 being
the fifth and sixth waves, respectively.
To complement the population census, the Chinese government has
conducted a population sample survey of a 1% sample of the whole
population since 1987. Starting from the second wave in 1995, the Chinese
government conducted this survey every 10 years, with each wave conducted
in the middle of two population censuses. The 2005 and 2015 population
surveys used in this paper are the third and fourth waves, respectively.
Electronic copy available at: https://ssrn.com/abstract=4101730

10
The population census and population sample survey include detailed
individual-level information. Particularly important for our paper, they
collected individuals’ education and employment status, facilitating the
construction of a high school enrollment rate at the prefecture level.
Additionally, the 2000 population census collected detailed information on
an individual’s migration status, from which we can define migrants and
obtain information on migrants from each prefecture, including the
destination and the number of migrants to each destination. In our paper,
migrants in 2000 are defined as those (excluding students) aged between
16-60 and moving from other prefectures over the past five years.
6
The samples we obtain access to include a 0.95% subsample of the
2000 census, a 0.35% subsample of the 2010 census, a 15% subsample of the
2005 population survey, and a 9.5% subsample of the 2015 population
survey. In most analyses, we focus on rural samples, i.e., individuals with
agricultural hukou.
We obtain tariff data from the World Integrated Trade Solution
(WITS) maintained by the World Bank. The WITS collects information on
tariffs imposed on export products from each country to all other countries in
the world. We focus on tariffs imposed on products at the 6-digit level of the
HS exported from China to other countries from 2000 to 2015.
6
Only for individuals moving to the local area in the past five years, the 2000 population
census includes information on their home prefectures.
Electronic copy available at: https://ssrn.com/abstract=4101730

11
We also use China custom data from 2000. These data include
information on export products at the HS 6-digit level from Chinese
prefectures to other destination countries. Additionally, we obtain prefectural
characteristics from the China City Statistical Yearbooks (2000-2016).
3.2. Measurements
The outcome variable of most interest in our paper is the high school
enrollment rate measured at the prefecture level. The high school enrollment
rate is defined as the share of individuals who have completed high school or
are attending high school in the 16- to 18-year-old cohort who completed
middle school in each prefecture in each year. We focus on the 16- to
18-year-old cohort because these individuals are at the age of attending high
school. We use the 2005 population survey to calculate the high school
enrollment rate for 2000 to 2005, the 2010 population census for 2006 to
2010 and the 2015 population survey for 2011 to 2015. Taking the 2005
population survey as an example, to calculate the high school enrollment rate
in 2004, we use a cohort aged 17-19 in 2005 (aged 16-18 years in 2004).
Education statuses recorded include (1) attending middle school or below; (2)
attending high school or above; (3) graduating from high school or above
and not in school currently; and (4) graduating from middle school or below
and not in school currently. We exclude individuals choosing option (1) since
they are unlikely to have completed middle school in 2004. The high school
Electronic copy available at: https://ssrn.com/abstract=4101730

12
enrollment rate is therefore defined as the sum of individuals choosing
options (2) and (3) over the sum of individuals choosing options (2), (3) and
(4). This measurement could be upward biased if the cohort aged 17-19 was
attending the first year of high school in 2005 (which means they graduated
from middle school in 2004), but this is rarely the case since Chinese
children usually start primary school at 6-7 years old and finish middle
school at 15-16 years old. By doing so, we construct high school enrollment
rate for prefectures in 2000-2015.
7
We construct two variables to capture tariff shocks faced by
prefectures: Ta r if f and Ta rif f spi llo ver . The former represents tariff shocks
imposed on local exports, while the latter represents tariff shocks imposed on
exports in other prefectures and passing through to local prefectures by
migrants. The first variable Tar if f is defined as follows:
!"#$%%!" &
' '
#$%&'(!"#,%&'(((
#$%&'(!,%&'(((
)* !"#$%%)*"
, (1)
where
(
represents the prefecture,
)
represents the 6-digit HS product,
*
represents the destination country, and
+
represents the year.
!"#$%%)*"
is
the tariff imposed on product
)
by destination country
*
in year
+
.
,-./#0!)*+",-...
is the export value of product
)
from prefecture
(
to
destination county
*
in 2000.
,-./#0!+",-...
is the total export value of
prefecture
(
in 2000. We use the year 2000 because it is the earliest year of
our sample period. Intuitively,
!"#$%%!"
is a weighted average of tariffs
7
Although Beijing, Tianjin, Shanghai and Chongqing are provincial level cities, we still
include them in our sample and call them prefectures as well.
Electronic copy available at: https://ssrn.com/abstract=4101730

13
imposed on product k by destination d in year t, where the weight is the share
of product k exported to destination d by prefecture c in 2000 of the total
exports by prefecture c in the same year.
The second variable Tariff spillover is defined as follows:
!"#$%%!"
/&
'
0!,%&'(((
!
′
0!,%&'(((
!"#$%%!
′
"
!
′
1!
, (2)
where
!"#$%%!"
/
is the tariff spillover faced by prefecture c in year t.
!"#$%%!
′
"
is the tariff faced by prefecture
(
′
(other than c) in year t
(constructed according to Equation (1)),
1!+",-...
!
′
is the number of
migrants from prefecture c to prefecture
(
′
in 2000, and
1!+",-...
is the
total number of migrants in prefecture c in 2000. Therefore,
!"#$%%!"
/
is a
weighted average of tariffs imposed on all other prefectures, and the weight
is the share of migrants from prefecture c to each of the other prefectures.
Table 1 presents the summary statistics of the main variables, where
the observation unit is the prefecture by year. As shown, the mean high
school enrollment rate among the rural population was 44% in all prefectures
in China from 2000 to 2015. This value increased dramatically from the
average value of 18% in 2000 to approximately 77% in 2015. The average
export tariff level for 2000-2015 is 8%. The standard variation of the tariff
levels across prefectures and years over the period is large at approximately
three times the mean value. The average tariff spillover is approximately the
same value as the local tariff levels. However, its standard deviation is much
smaller than that of the local tariff.
Electronic copy available at: https://ssrn.com/abstract=4101730

14
Table 2 presents the migration rates among the rural population aged
16–60 years old. As shown in Panel A, the rate of migration of the rural
labor force increases steadily from 3.8% in 2010 to 11.6% in 2010 and then
declines to 8.3% in 2015. Panels B and C report the migration rates of the
destination prefectures, which are categorized according to their spatial
distributions (Panel B) and degrees of economic development (Panel C).
Specifically, Panel B shows that 2.3% of rural laborers migrated outside
their home province in 2000, which is higher than that of within-province
migration by 0.8 percentage points. The difference widened in 2005 and
2010. For instance, the rate of migration outside the home province
increased to 8.5% in 2010, which is 5.4 percentage points more than the rate
of migration within the province. Both rates of migration within the home
province and outside the home province declined from 2010 to 2015, and the
gap in the two rates narrowed as well. Overall, these statistics indicate that
across-province migration is more common than within-province migration
among the rural population in China from 2000-2015. Panel C compares the
rates of migration to the first-tier cities (i.e., Beijing, Shanghai, Guangzhou,
and Shenzhen) and those of migration to the other prefectures. The results
indicate that the rate of migration to the first-tier prefectures in 2005 was
approximately four times greater than that in 2000. The rate of migration to
other prefectures shows a similar pattern over this period, increasing from
0.027 in 2000 to 0.066 in 2005. However, the rate of migration to the
Electronic copy available at: https://ssrn.com/abstract=4101730

15
first-tier cities gradually decreased afterward, whereas the rate of migration
to other prefectures continuously increased until 2010 and then decreased to
a level similar to that in 2005.
4. Empirical Specification
For the main analyses, we use the following empirical specification:
,234!" & 5.6 52!"#$%%!" 6 5-!"#$%%!"
/6 7"
′
8!6 9!3 6 :4" 6 ;!"
, (3)
where
,234!"
is the high school enrollment rate of rural youth between 16
and 18 years old from prefecture
(
in year
+
;
!"#$%%!"
is the level of the
export tariff of prefecture
(
in year
+
;
!"#$%%!"
/
is the spillovers of export
tariffs for prefecture
(
in year
+
;
7"
′
8!
are prefecture-level initial
characteristics in 1999 interacted with year fixed effects;
9!3
are the
prefecture-sample fixed effects (the sample refers to the waves of the
population census and sample survey used in our paper), controlling for any
prefecture-sample-level time-invariance factors;
:4"
are the province-year
fixed effects, controlling for any provincial-level year-varying factors; and
;!"
is the error term. The regressions are estimated using the size of the
population aged 16-18 in rural areas in the corresponding prefecture and year
as weight. We calculate standard errors by clustering over prefectures.
8
Parameters
52
and
5-
are of interest. Specifically,
52
represents the
effect of the local export tariff on the high school enrollment rate in the local
8
Our sample includes 294 prefectures.
Electronic copy available at: https://ssrn.com/abstract=4101730

16
prefecture, whereas
5-
represents the spillover effect of export tariffs in
migration destinations on the local high school enrollment rate.
5. Results
5.1. Descriptive Results
Figure 1 illustrates the high school enrollment rate of rural youth
aged 16–18 years old in each prefecture in China in 2000. The figure
demonstrates a stark geographic difference in the high school enrollment rate
in this year. As shown, the rate is lower in the western and inner prefectures
than in the eastern coastal prefectures. Figure 2 depicts the spatial
distribution of changes in high school enrollment rates and changes in tariff
shocks across prefectures in China from 2000 to 2015. As shown in Panel A,
the high school enrollment rate increases in almost all prefectures, but the
change varies by region. The increase in high school enrollment is much
lower in prefectures in northeastern and southwestern China than in
prefectures in other regions. Panels B and C show that most prefectures
witnessed a decrease in the levels of export tariff and tariff spillovers from
2000–2015. In addition, both levels of tariff and tariff spillover decrease
more in northeastern China than in other regions. This suggests a potential
positive relation between the high school enrollment rate and tariffs and
tariff spillovers. We explore their relationship below through regression
analyses.
Electronic copy available at: https://ssrn.com/abstract=4101730

17
5.2. Baseline Results
We start by presenting the baseline results. Table 3 reports the results
of the impact of local export tariff shocks and export tariff shocks in
destination prefectures of migration on the high school enrollment rate of
rural youth between 16 and 18 years old. Column (1) presents the benchmark
estimates of
52
and
5-
from Equation (3). They are respectively equal to
0.012 and 0.454, with the former being significant at the 10% level and the
latter being significant at the 1% level. The estimated coefficients show that
the high school enrollment rate of rural youth aged 16–18 increases
significantly with both local export tariff shocks and export tariff shocks in
migration-destination prefectures. The magnitude of the coefficients
indicates that a 1 percentage point decrease in export tariffs in
migration-destination prefectures reduces the high school enrollment rate by
0.454 percentage points, whereas a 1 percentage point decrease in local
tariffs reduces the high school enrollment rate by 0.012 percentage points.
Alternatively, the coefficients show that a 1 standard deviation reduction in
export tariffs in destinations leads to a decrease in the local high school
enrollment rate by 2.63 percentage points, and a 1 standard deviation
reduction in local export tariffs leads to a decrease in the local high school
enrollment rate by 0.26 percentage points.
Electronic copy available at: https://ssrn.com/abstract=4101730

18
One main concern regarding the potential endogeneity of the tariff
spillover in Equation (3) is that the initial structure of the migration networks
used to construct the variable tariff spillover may be correlated with the local
high school enrollment rate in later years. To address this concern, we
control for some prefecture-level initial conditions interacted with year fixed
effects. Specifically, in Column (2), we control for the yearly specific effects
of factors that may affect the initial migration structure, including the
difference in the economic conditions (measured by log GDP per capita) of
the local prefecture and those of the prefecture with the highest value of per
capita GDP in 1999 (i.e., Shenzhen) and the log of geographic distance
between the local prefecture and Shenzhen. In Columns (3) and (4), we
further control for the interaction terms of year dummies with a battery of
baseline variables selected according to the ordinary least squares (OLS)
regression and the least absolute shrinkage and selection operator (LASSO),
respectively. The criteria for the selection of these variables are to choose
significant factors in determining the initial destination structure of
migration networks.
9
The estimates based on these alternative specifications are similar to
those reported in Column (1). For instance, the magnitude of the coefficients
in Column (4) indicates that a 1 percentage point decrease in export tariffs in
migration-destination prefectures reduces the high school enrollment rate by
9
See Section A of the Appendix for more details about the selection of the
prefecture-level characteristics.
Electronic copy available at: https://ssrn.com/abstract=4101730

19
0.219 percentage points, whereas a 1 percentage point decrease in local
tariffs reduces the high school enrollment rate by 0.011 percentage points.
Alternatively, the coefficients show that a 1 standard deviation reduction in
export tariffs in destinations leads to a decrease in the local high school
enrollment rate by 1.27 percentage points, and a 1 standard deviation
reduction in local export tariffs leads to a decrease in the local high school
enrollment rate by 0.24 percentage points. These results should alleviate
concerns about the endogeneity of the initial network structure of migration.
We take the results shown in Column (4) of Table 3 as a benchmark.
To address further concerns regarding the potential endogeneity of
the initial network structure of migration and export structure in the
construction of tariff and tariff spillover, we perform a placebo test in Table
B1 in the Appendix by examining the impact on high school enrollment rates
among urban youth aged 16 to 18 years. The export expansion created job
opportunities mostly among foreign direct investment and private firms,
whereas urban residents mostly worked at state-owned enterprises.
Furthermore, urban youth already have a high enrollment rate of high school.
Thus, we expect to find few effects of export tariffs on high school
enrollment among urban youth. As shown by the statistics at the bottom of
the table, the mean of the high school enrollment rate of urban youth is
86.6%, which is approximately double that of their rural counterparts,
demonstrating substantial rural-urban inequality in human capital investment
Electronic copy available at: https://ssrn.com/abstract=4101730

20
in China. The regression results show that neither the tariff nor the tariff
spillover has a significant impact on the high school enrollment rate for
urban youth. These results greatly support our identification assumption.
5.3. Robustness Checks
Table 4 checks the robustness of the results. First, one might be
concerned that the export tariffs imposed by destination countries could be
related to import tariffs imposed by the Chinese government on products
from these countries. Import tariffs might affect competition in domestic
markets, which affects demand for migrant workers by domestic producers
and therefore individuals’ decisions to continue high school. To address this
concern, Column (1) further controls for the local import tariff and import
tariff spillovers in migration destination prefectures, which are, respectively
defined by Equations (1) and (2) with export tariffs replaced by import tariffs.
The coefficients of local export tariff and tariff spillovers are quite similar to
the benchmark estimates in Table 3, while the coefficients of import shocks
per se are not statistically significantly different from 0. These results
indicate that changes in import tariffs are unlikely to confound the estimates
of the impact of export tariff shocks.
One might also be concerned that the change in destination countries’
tariffs may combine with the change in trade uncertainty, affecting China’s
exports to these countries (e.g., Feng, Li and Swenson, 2017) and therefore
Electronic copy available at: https://ssrn.com/abstract=4101730

21
confounding our main results. Following Handley (2014), we measure trade
uncertainty using the log difference between the bound and effectively
applied tariff rates. This log difference is at the product-destination
country-year level. We then aggregate product-destination country-year level
trade uncertainty to the city level using Equation (1) but while replacing
tariffs with trade uncertainty. We also construct the trade uncertainty
spillover by the method given in Equation (2). We include these two
variables in the main regression. To account for the decline in trade
uncertainty after China’s WTO accession in 2001 (Feng, Li and Swenson,
2017), we also include the interactions of trade uncertainty and trade
uncertainty spillover and a dummy for years prior to 2002. The results are
shown in Column (2) of Table 4. We find that the coefficients of tariffs and
tariff spillover are similar to the main results, demonstrating the robustness
of our results.
To address other concerns about possible confounding factors that
are related to the geographic location of the prefecture, we control for the
latitude and longitude of the administrative center of the prefecture
interacted with year dummies and the distance to the nearest main port of
China in 1999 interacted with year dummies. The results remain robust
regardless of whether we add these to the benchmark specification separately
or together (see Columns (3) to (5)).
Electronic copy available at: https://ssrn.com/abstract=4101730

22
Last, one may argue that the estimates could be driven by the global
financial crisis of 2008 and by possible heterogeneity across prefectures in
response to the crisis because of differences in their reliance on trade. To
address this concern, we remove the observations from 2008. The estimated
coefficients of both the local export tariff and the export tariff spillovers are
slightly greater than the benchmark estimates (see Column (6)). To address
concerns of possible structural changes in the local economy after the
financial crisis, we only keep observations for the years prior to 2008 in
Column (7) of Table 4. The estimates of the tariff spillover do not change
significantly, while the coefficient of the local export tariff is negative,
although it is not statistically significantly different from 0 and the
magnitude is small. Overall, these results suggest that the estimates on the
impact of tariff shocks are unlikely to be driven by the effect of the global
financial crisis of 2008.
5.4. Heterogeneity
Table 5 examines the heterogeneity of the effects of export tariff
shocks by gender and region. The first two columns present regressions
similar to those of Column (4) in Table 3 using high school enrollment rate
of females and males, respectively. As demonstrated, the high school
enrollment rate of both female and male youth in China’s rural areas
increases with local export tariffs. Specifically, when the level of local
Electronic copy available at: https://ssrn.com/abstract=4101730

23
export tariff increased by 1 standard deviation, the high school enrollment
rate of male and female students in rural China increased by 0.33 and 0.20
percentage points, respectively. Although the impacts of local export tariffs
among male youth are statistically significant, the difference between males
and females is not statistically significant. Consistent with the estimates of
the average effect, the impact of spillover effects on the high school
enrollment rate is more substantial than the effect of local export tariffs for
both males and females. Specifically, the high school enrollment rate of male
and female students in rural China increased by 1.11 and 1.50 percentage
points, respectively, when the export tariffs in migration-destination
prefectures increased by 1 standard deviation. The difference between males
and females is not significant as well.
Columns (3) to (5) examine the heterogeneity by the initial economic
conditions of the prefecture. The sample is categorized into three regions, i.e.,
eastern, central and western regions, according to classification of the
National Bureau of Statistics of China in 2000, which was based on
economic conditions of the province.
10
As shown, the effects of both local
tariff and tariff spillover on the high school enrollment rate are positive and
significant among prefectures in the central region. For prefectures in eastern
10
The eastern region includes Beijing, Tianjin, Hebei, Liaoning, Shanghai, Jiangsu,
Zhejiang, Fujian, Shandong, Guangdong, Guangxi and Hainan. The central region
includes Shanxi, Inner Mongolia, Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hubei and
Hunan. The western region includes Sichuan, Chongqing, Guizhou, Yunnan, Tibet,
Shaanxi, Gansu, Ningxia, Qinghai and Xinjiang.
Electronic copy available at: https://ssrn.com/abstract=4101730

24
region, the impact of local tariff is greater than that of prefectures in other
regions, although it is statistically insignificant. The impact of tariff spillover
for prefectures in the eastern region is about the same in magnitude as that in
the central region, and is statistically significant. For prefectures in western
region, the impacts of local tariff and tariff spillovers are economically and
statistically insignificant. The results are consistent with the conjecture that
the spillover effects are concentrated in the main areas from which migrants
originate (i.e., the eastern and central regions).
6. Mechanisms
This section investigates the potential channels through which export
tariff shocks may affect the high school enrollment of young people. Given
that rural-to-urban migrants are mainly concentrated in the secondary
industry, the effects of export shocks from destination prefectures with more
workers in the secondary industry are likely salient for educational decisions.
Thus, we first explore the potential heterogeneous responses of high school
enrollment to export tariff shocks from destination prefectures with different
workers in the secondary industry. Second, export shocks in the secondary
industry likely affect job opportunities, which may influence the educational
decisions of households in the original prefecture. We thus examine the
impact of export shocks on the employment growth rate of the local residents
and emigrants of a prefecture in the secondary industry. Last, as discussed in
Electronic copy available at: https://ssrn.com/abstract=4101730

25
the introduction, the effect of export shocks can be attributed to two effects:
the opportunity cost effect and the income effect. The former reduces the
likelihood of high school enrollment, while the latter increases it. We also
provide some suggestive evidence to disentangle the opportunity cost effect
and income effect.
6.1. Role of the Secondary Industry
Given that Chinese rural-to-urban migrants mainly work in the
secondary industry, we expect export tariff shocks to the destination
prefectures with more workers in the secondary industry to be more
pronounced than shocks to other prefectures in affecting the high school
enrollment rate of youth in the original prefecture. To test this hypothesis,
we decompose the variable tariff spillover in Equation (3) into two variables,
namely, the tariff spillover from prefectures with a high volume (above the
median) of second industry workers and that from prefectures with a low
volume (below the median) of secondary industry workers in 1999. Doing so
allows for heterogeneous responses of high school enrollment to export tariff
shocks from destination prefectures with large and small numbers of second
industry workers.
The results are reported in Table 6. As shown in Column (1), both
spillover effects of export tariffs on the high school enrollment rate are
positive and significant. However, the spillover effect of export tariffs in
Electronic copy available at: https://ssrn.com/abstract=4101730

26
prefectures with a high volume of migrant workers in the secondary industry
is much greater than that of export tariffs in prefectures with a low volume of
migrant workers in the secondary industry. Columns (2) to (4) show that the
results are robust when we further control for the interaction terms between
year fixed effects and the initial conditions.
Overall, the results suggest that the positive spillover effects of
export tariffs in migration destinations on high school enrollment are
primarily driven by export tariff shocks from destinations with relatively
abundant secondary industry workers. These results are consistent with the
conjecture that spillover effects on education decisions are mainly caused by
variation in export shocks to the secondary industry where migrants mainly
worked.
6.2. Impact on Employment Growth
We then examine the impact of tariff shocks on the employment
growth rate of local residents and emigrants in the secondary industry of
prefectures. We present the details of constructing employment growth rates
used in this section in Appendix C. Table 7 shows the estimation results.
Column (1) controls for prefecture-sample fixed effects and province-year
fixed effects, whereas Columns (2) to (4) further control for other
year-specific variation in the employment growth rate in the secondary
industry determined by initial conditions selected according to different
Electronic copy available at: https://ssrn.com/abstract=4101730

27
methods. Panel A of Table 7 uses the employment growth rate of local
residents as the outcome variable. We can see that local export tariffs have a
significantly negative effect on the employment growth rate of the residents
in the secondary industry. Panel B of Table 7 uses the employment growth
rate of the secondary industry for emigrants of local prefectures as the
outcome variable. In line with this conjecture, the export tariff of migration
destinations significantly reduces the employment growth rate of emigrants
in the secondary industry.
Overall, the results in Table 7 indicate that a reduction in export
tariffs in local and destination prefectures would significantly raise the
employment growth rates of residents and emigrants in the secondary
industry, respectively. The signal of increases in job opportunity may
circulate around the local prefecture or transmit back to the home prefecture
via migration networks. The decrease in the high school enrollment rate as a
result of the reduction in local export tariffs and tariff spillovers found in the
main results reflects the responses of households’ education decisions to
more employment opportunities or higher expected opportunity costs of
schooling in potential migration destinations.
6.3. Opportunity Cost Effects versus Income Effects
In addition to the channel of the opportunity cost of schooling, export
tariff shocks may have income effects on educational choice, as has been
Electronic copy available at: https://ssrn.com/abstract=4101730

28
well documented in the literature. Remittances decrease when export tariffs
rise. Investment in human capital may be hindered by fewer remittances;
therefore, high school enrollment may decrease when export tariffs increase.
The sign of the income effect on the high school enrollment rate is the
opposite of the sign of the effect of opportunity costs. Thus, the positive
effects of export tariffs on the high school enrollment rate would be a lower
bound for the estimates of opportunity cost effects.
To provide some suggestive evidence about the opportunity cost
effect and income effect, we decompose the spillovers of export tariffs into
spillovers of export tariffs through networks of migrants aged 16-24 years
and those through migrants aged 25-60 years. As an alternative specification,
we further decompose the latter into spillovers of export tariffs through
migrants aged 25-44 years and those aged 45-60 years. Given that older
migrants may be more likely to provide financial support to youth, we may
expect that the spillovers of export tariffs from older migrants may reflect
more of the income effect than the spillovers of export tariffs from younger
migrants. However, one caveat we need to bear in mind is that this exercise
is not feasible for local export shocks such that we cannot distinguish income
effects and opportunity cost effects of the local export shock.
Consistent with this conjecture, Table 8 indicates that the positive
spillover effects are mainly driven by the spillover effects of export tariffs
through younger migrants (16-24 years old), which are slightly greater than
Electronic copy available at: https://ssrn.com/abstract=4101730

29
the estimate of the total effect reported in the benchmark results. Columns (4)
to (6) show that the spillover effects of export tariffs through migrants aged
45-60 years are negative, although they are not statistically significant.
Overall, these results offer some suggestive evidence of a potential income
effect of export tariffs on the high school enrollment rate. However, the
effect is less salient than the effect of opportunity cost in our study context.
7. Conclusion
In this study, we investigate the impact of local export expansion and
export expansion in other prefectures through migration networks on high
school enrollment in rural China. We find that export expansion in
destination prefectures significantly decreases the high school enrollment
rate in rural areas of the original prefecture and with a much greater than the
effect of local export expansion. Beyond the average effects, we find that
export expansion decreases high school enrollment for both boys and girls.
Meanwhile, we find that the effects of tariff spillovers on the high school
enrollment rate are stronger for prefectures with a high volume of migrant
workers in the secondary industry and through networks of younger
migrants.
In terms of policy implications, this study suggests that the demand
for human capital investment is important for educational attainment. The
government needs to pay attention to these factors instead of focusing on the
Electronic copy available at: https://ssrn.com/abstract=4101730

30
supply of education alone. Moreover, while trade liberalization has been
regarded as accelerating economic growth, our study suggests that it may
hinder economic development in the long run when it reduces human capital
acquisition. The overall impact of trade liberalization on economic growth
thus could be reversed in the long run.
Electronic copy available at: https://ssrn.com/abstract=4101730

31
REFERENCES
Atkin, David, (2016). Endogenous Skill Acquisition and Export Manufacturing
in Mexico. American Economic Review, 106(8), 2046-2085.
Cai, Fang, Albert Park, and Yaohui Zhao, (2008). The Chinese Labor Market in
the Reform Era. In China’s Great Economic Transformation, ed. Loren
Brandt and Thomas Rawski, 167-214. Cambridge University Press:
Cambridge.
Carrington, William J., Enrica Detragiache, and Tara Vishwanath, (1996).
Migration with Endogenous Moving Costs. American Economic Review,
86(4), 909-930.
Caselli, Mauro, (2014). Trade, Skill-Biased Technical Change and Wages in
Mexican Manufacturing. Applied Economics, 46(3), 336-348.
China Statistical Yearbook (1978, 1991, 1998, 2001, 2006, 2009, 2016). China
Statistical Press.
De Brauw, Alan, and John Giles, (2017). Migrant Opportunity and the
Educational Attainment of Youth in Rural China. The Journal of Human
Resources, 52(1), 272-311.
Duflo, Esther, Pascaline Dupas, and Michael Kremer, (2011). Peer Effects,
Teacher Incentives, and the Impact of Tracking: Evidence from a
Randomized Evaluation in Kenya. American Economic Review, 101(5),
1739-1774.
Electronic copy available at: https://ssrn.com/abstract=4101730

32
Duflo, Esther, Rema Hanna, and Stephen P. Ryan, (2012). Incentives Work:
Getting Teachers to Come to School. American Economic Review, 102(4),
1241-1278.
Edmonds, Eric V., and Nina Pavcnik, (2005). The Effect of Trade Liberalization
on Child Labor. Journal of International Economics, 65, 401-419.
Edmonds, Eric V., Nina Pavcnik, and Petia Topalova, (2010). Trade Adjustment
and Human Capital Investments: Evidence from Indian Tariff Reform.
American Economic Journal: Applied Economics, 2(4), 42-75.
Feng, Ling, Zhiyuan Li, and Deborah L. Swenson, (2017). Trade Policy
Uncertainty and Exports: Evidence from China’s WTO Accession. Journal
of International Economics, 106, 20-36.
Friedman, Jerome, Trevor Hastie, and Rob Tibshirani, (2010). Regularization
Paths for Generalized Linear Models via Coordinate Descent. Journal of
Statistical Software, 33, 1, 1-22.
Glewwe, Paul, Michael Kremer, and Sylvie Moulin, (2009). Many Children
Left Behind? Textbooks and Test Scores in Kenya. American Economic
Journal: Applied Economics, 1(1), 112-135.
Greenland, Andrew, and John Lopresti, (2016). Import Exposure and Human
Capital Adjustment: Evidence from the U.S. Journal of International
Economics, 100, 50-60.
Handley, Kyle, (2014). Exporting under Trade Policy Uncertainty: Theory and
Evidence. Journal of International Economics, 94 (1), 50-66.
Electronic copy available at: https://ssrn.com/abstract=4101730

33
Heath, Rachel, and A. Mushfiq Mobarak, (2015). Manufacturing Growth and
the Lives of Bangladeshi Women. Journal of Development Economics, 114,
1-15.
Jensen, Robert T., (2010). Economic Opportunities and Gender Differences in
Human Capital: Experimental Evidence from India. NBER Working Paper
16021.
Lardy, Nicholas R. (1994). China in the World Economy. Peterson Institute
Press: All Books.
Liang, Zai, and Zhongdong Ma, (2004). China’s Floating Population: New
Evidence from the 2000 Census. Population and Development Review,
30(3), 467-488.
Li, Bingjing, (2018). Export Expansion, Skill Acquisition and Industry
Specialization: Evidence from China. Journal of International Economics,
114, 346-361.
Li, Jie, Yi Lu, Hong Song, and Huihua Xie, (2019). Long-term Impact of Trade
Liberalization on Human Capital Formation. Journal of Comparative
Economics, 47(4): 946-961.
Majlesi, Kaveh, (2014). Demand for Low-Skilled Labor and Parental
Investment in Children’s Education: Evidence from Mexico. Working
Paper.
Electronic copy available at: https://ssrn.com/abstract=4101730

34
Meng, Xin, (2000). Regional Wage Gap, Information Flow, and Rural-Urban
Migration. In Rural Labor Flows in China, ed. Yaohui Zhao and Loraine
West, 251-277. University of California Press: Berkeley.
Montgomery, James D., (1991). Social Networks and Labor-Market Outcomes:
Toward an Economic Analysis. American Economic Review, 81(5),
1407-1418.
Munshi, Kaivan, (2003). Networks in the Modern Economy: Mexican Migrants
in the U.S. Labor Market, Quarterly Journal of Economics, 118(2),
549-599.
Olney, William W., (2013). The Composition of Exports and Human Capital
Acquisition. Working Paper.
Oster, Emily, and Bryce Millett Steinberg, (2013). Do IT Service Centers
Promote School Enrollment? Evidence from India. Journal of
Development Economics, 104, 123-135.
Park, Albert, Xinzheng Shi, Chang-tai Hsieh, and Xuehui An, (2015). Magnet
High Schools and Academic Performance in China: A Regression
Discontinuity Design. Journal of Comparative Economics, 43, 825-843.
Rozelle, Scott, Guo Li, Minggao Shen, Amelia Hughart, and John Giles, (1999).
Leaving China’s Farms: Survey Results of New Paths and Remaining
Hurdles to Rural Migration. The China Quarterly, 158, 367-393.
Electronic copy available at: https://ssrn.com/abstract=4101730

35
Shastry, Gauri Kartini, (2012). Human Capital Response to Globalization:
Education and Information Technology in India. The Journal of Human
Resources, 47(2), 287-330.
Song, Yang, (2014). What Should Economists Know about the Current Chinese
hukou System? China Economic Review, 29, 200–212.
Xu, Tongsheng, Weiwei Li, Yulian Jiang, and Ya Chen, (2020). The Impact of
Destination Tariffs on China’s Exports: Country, Firm, and Product
Perspectives. Journal of Asian Economics, 71, 101246.
Yang, Dean, (2008). International Migration, Remittances, and Household
Investment: Evidence from Philippine Migrants’ Exchange Rate Shocks.
Economic Journal, 118, 591-630.
Zhao, Yaohui, (2003). The Role of Migrant Networks in Labor Migration: The
Case of China. Contemporary Economic Policy, 21(4), 500-511.
Electronic copy available at: https://ssrn.com/abstract=4101730

36
TABLES AND FIGURES
Figure 1. High school enrollment rate (16–18 years old) in 2000
Notes: The figure illustrates the high school enrollment rate of the rural population aged 16 to
18 in each prefecture in China in 2000. The data are from the China Population Census of
2005.
Electronic copy available at: https://ssrn.com/abstract=4101730

37
Figure 2. Spatial distribution of changes in high school enrollment rates and tariff shocks (2000–2015)
Panel A. Change in high school enrollment rates (2000–2015)
Panel B. Change in tariffs (2000–2015)
Panel C. Change in tariff spillovers (2000–2015)
Notes: Panel A illustrates the change in the high school enrollment rate of rural residents aged 16–18 years for each prefecture in China between
2000 and 2015. The data are from the China Population Censuses of 2005 and 2015. Panels B and C illustrate the change in tariffs and tariff
spillovers for each prefecture in China between 2000 and 2015, respectively. The data are drawn from customs data, the WITS, and the China
Population Censuses of 2000 and 2015.
Electronic copy available at: https://ssrn.com/abstract=4101730

38
Table 1. Summary statistics of the main variables
Notes: The table illustrates the summary statistics of the main variables. The observation unit
of the variables is the prefecture-year. The summary statistics are estimated from the size of
the population aged 16-18 in rural areas in the corresponding prefecture and year as weight.
Obs . Mean S.D. Median Min Max
(1) (2) (3) (4) (5) (6)
Panel A: Full sample
High s chool en ro llment rate (16-18 y ears old) 4704 0.442 0.178 0.442 0.000 1.000
Tariff (Local) 4704 0.080 0.220 0.041 0.000 3.060
Tariff (Spillover, 16-60 years old migrants as weights) 4704 0.060 0.058 0.043 0.012 0.725
Tariff (Spillover, 16-24 years old migrants as weights) 4704 0.055 0.052 0.040 0.008 0.573
Panle B: Year 2000
High s chool en ro llment rate (16-18 y ears old) 294 0.178 0.092 0.162 0.000 0.608
Tariff (Local) 294 0.080 0.239 0.040 0.000 3.042
Tariff (Spillover, 16-60 years old migrants as weights) 294 0.062 0.066 0.042 0.024 0.718
Tariff (Spillover, 16-24 years old migrants as weights) 294 0.057 0.059 0.039 0.017 0.566
Panle C: Year 2005
High s chool en ro llment rate (16-18 y ears old) 294 0.414 0.154 0.399 0.000 0.858
Tariff (Local) 294 0.079 0.221 0.045 0.000 2.931
Tariff (Spillover, 16-60 years old migrants as weights) 294 0.061 0.059 0.042 0.024 0.692
Tariff (Spillover, 16-24 years old migrants as weights) 294 0.057 0.054 0.041 0.022 0.545
Panle D: Year 2010
High s chool en ro llment rate (16-18 y ears old) 294 0.594 0.127 0.614 0.158 0.895
Tariff (Local) 294 0.084 0.217 0.045 0.000 2.941
Tariff (Spillover, 16-60 years old migrants as weights) 294 0.062 0.055 0.048 0.023 0.690
Tariff (Spillover, 16-24 years old migrants as weights) 294 0.056 0.049 0.044 0.015 0.543
Panle E: Year 2015
High s chool en ro llment rate (16-18 y ears old) 294 0.767 0.116 0.779 0.000 1.000
Tariff (Local) 294 0.033 0.027 0.030 0.000 0.387
Tariff (Spillover, 16-60 years old migrants as weights) 294 0.029 0.006 0.029 0.012 0.135
Tariff (Spillover, 16-24 years old migrants as weights) 294 0.028 0.006 0.027 0.008 0.135
Va r ia b l e
Electronic copy available at: https://ssrn.com/abstract=4101730

39
Table 2. Migration rates by destination
Notes: The table reports the mean and standard deviation of migration rates (Panel A) and the
migration rates by destination (Panels B and C) for the rural population between 16 and 60
years old in 2000, 2005, 2010 and 2015. The first-tier prefectures include Beijing, Shanghai,
Guangzhou, and Shenzhen. All summary statistics are estimated from the size of the
population aged 16-18 in rural areas in the corresponding prefecture and year as weight.
(1) (2) (3) (4) (5)
Pane l A
Rat e o f migration 0.038 0.109 0.116 0.083 0.094
(0.029) (0.092) (0.106) (0.058) (0.093)
Pane l B
Rat e o f migration within p rovince 0.015 0.023 0.031 0.024 0.025
(0.024) (0.034) (0.044) (0.025) (0.037)
Rat e o f migration o ut sid e p rov ince 0.023 0.086 0.085 0.060 0.069
(0.024) (0.094) (0.097) (0.056) (0.086)
Pane l C
Rat e o f migration t o t he firs t tier p refe cture s 0.011 0.043 0.031 0.020 0.028
(0.014) (0.046) (0.038) (0.021) (0.037)
Rat e o f migration t o o th er p refe cture s 0.027 0.066 0.085 0.063 0.066
(0.018) (0.054) (0.075) (0.044) (0.063)
Ye a r
2000
Ye a r
2005
Ye a r
2010
All
Ye a r s
Ye a r
2015
Electronic copy available at: https://ssrn.com/abstract=4101730

40
Table 3. The impact of tariff shocks on the high school enrollment rate of rural youth
Notes: The difference in the log of GDP per capita is defined as the difference between the
log GDP per capita of the local prefecture and that of the prefecture with the highest per
capita GDP in 1999 (i.e., Shenzhen). The log of distance is defined as the logarithm of the
geographic distance between the local prefecture and Shenzhen. The regressions are
estimated from the size of the population aged 16-18 in rural areas in the corresponding
prefecture and year as weight. Standard errors in parentheses are clustered at the prefecture
level. ***, **, and * represent the 1%, 5%, and 10% significance levels, respectively.
(1) (2) (3) (4)
Tariff (Local)
0.012* 0.016*** 0.013** 0.011*
(0.007) (0.006) (0.006) (0.006)
Tariff (Spillover)
0.454*** 0.369*** 0.222** 0.219**
(0.105) (0.115) (0.090) (0.088)
Prefecture-s ample fixed effects Ye s Ye s Ye s Ye s
Province-year fixed effects Ye s Ye s Ye s Ye s
Difference in log GDP per cap ita in 1999 × Year FE No Ye s Ye s Ye s
Lo g d is ta nc e × Ye ar FE No Ye s Ye s Ye s
Prefecture characteristics in 1999 (OLS) × Year FE No No Ye s No
Prefectures characterist ics in 1999 (LASSO) × Year FE No No No Ye s
Mean of high school enrollment rate 0.442 0.442 0.442 0.442
Adjusted R-squared
0.962 0.969 0.972 0.972
Number o f p refec tures
294 294 294 294
Obs erv atio ns
4,704 4,704 4,704 4,704
High s chool en ro llment rate (16-18 y ears old)
Electronic copy available at: https://ssrn.com/abstract=4101730

41
Table 4. Robustness checks
Notes: The difference in the log of GDP per capita is defined as the difference between the
log GDP per capita of the local prefecture and that of the prefecture with the highest per
capita GDP in 1999 (i.e., Shenzhen). The log of distance is defined as the logarithm of the
geographic distance between the local prefecture and Shenzhen. The regressions are
estimated from the size of the population aged 16-18 in rural areas in the corresponding
prefecture and year as weight. Standard errors in parentheses are clustered at the city level.
***, **, and * represent the 1%, 5%, and 10% significance levels, respectively.
Longitud
e and
latitude
Distance
to p ort
Both
Drop
year
2008
Keep
years ≤
2007
(1) (2) (3) (4) (5) (6) (7)
Tariff (Local)
0.011* 0.013** 0.011* 0.011* 0.011* 0.017** -0.006
(0.006) (0.006) (0.006) (0.006) (0.006) (0.008) (0.009)
Tariff (Spillover)
0.221** 0.233** 0.225*** 0.207** 0.217** 0.252** 0.301*
(0.089) (0.105) (0.086) (0.090) (0.087) (0.099) (0.162)
Import tariff (Local)
0.032
(0.052)
Import tariff (Spillover)
0.280
(0.380)
ΔlogT ariff (Local) × 1(Year < 2002)
-0.005
(0.014)
ΔlogT ariff (Spillover) × 1(Year < 2002)
0.036
(0.211)
ΔlogT ariff (Local)
0.016
(0.028)
ΔlogT ariff (Spillover)
0.361
(0.509)
Prefecture-sample fixed effects Yes Yes Yes Yes Yes Yes Yes
Province-year fixed effects Yes Yes Yes Yes Yes Yes Yes
Diff. in log GDP per capita in 1999 × Year FE Yes Yes Yes Yes Yes Yes Yes
Log distance × Year FE Yes Yes Yes Yes Yes Yes Yes
Prefecture char. in 1999 (LASSO) × Year FE Yes Yes Yes Yes Yes Yes Yes
Longitude × Year FE No No Yes No Yes No No
Latitude × Year FE No No Yes No Yes No No
Log distance to port × Year FE No No No Yes Yes No No
Mean of high school enrollment rate 0.442 0.447 0.442 0.442 0.442 0.437 0.349
Adjusted R-squared
0.972 0.972 0.972 0.972 0.972 0.971 0.968
Number of prefectures
294 268 294 294 294 294 294
Observations
4,704 4,277 4,704 4,704 4,704 4,410 2,352
Trade
uncertain
ty
Financial Crisis
Import
tariff
Geograp hic variation
High school enrollment rate (16-18 years old)
Electronic copy available at: https://ssrn.com/abstract=4101730

42
Table 5. Heterogeneity in educational responses to the tariff shocks
Notes: The difference in the log of GDP per capita is defined as the difference between the
log GDP per capita of the local prefecture and that of the prefecture with the highest per
capita GDP in 1999 (i.e., Shenzhen). The log of distance is defined as the logarithm of the
geographic distance between the local prefecture and Shenzhen. The regressions in Columns
(1) and (2) are estimated using the size of female and male populations aged 16-18 in rural
areas in the corresponding prefecture and year as weight, respectively, while the regressions
in Columns (3) to (5) are estimated using the size of the population aged 16-18 in rural areas
in the corresponding prefecture and year as weight. Standard errors in parentheses are
clustered at the city level. ***, **, and * represent the 1%, 5%, and 10% significance levels,
respectively.
Female Male Ea s t Cen tral West
(1) (2) (3) (4) (5)
Tariff (Local)
0.009 0.015* 0.088 0.013*** -0.016
(0.008) (0.008) (0.093) (0.005) (0.031)
Tariff (Spillover)
0.258** 0.192** 0.343** 0.362* * 0.101
(0.108) (0.093) (0.172) (0.147) (0.200)
Prefecture-s ample fixed effects Ye s Ye s Ye s Ye s Ye s
Province-year fixed effects Ye s Ye s Ye s Ye s Ye s
Difference in log GDP per cap ita in 1999 × Year FE Ye s Ye s Yes Ye s Ye s
Lo g d is ta nc e × Ye ar FE Ye s Ye s Ye s Ye s Ye s
Prefecture characteristics in 1999 (LASSO) × Year FE Ye s Ye s Ye s Ye s Ye s
Cofficients of Tariff (Local) are equal (P value)
Cofficients of Tariff (Spillover) are equal (P value)
Mean of high school enrollment rate 0.431 0.453 0.464 0.446 0.401
Adjusted R-squared
0.963 0.955 0.972 0.972 0.971
Number o f p refec tures
292 294 100 100 94
Obs erv atio ns
4,672 4,704 1,600 1,600 1,504
By g ender
By re gio n
Out come va riab le:
High s chool en ro llment rate (16-18 y ears old)
0.66
0.48
0.47
0.52
Electronic copy available at: https://ssrn.com/abstract=4101730

43
Table 6. Heterogeneity in the impact of tariff spillovers
Notes: The variables of tariff spillover include export tariff shocks from migration destination
prefectures with below- and above-median migrant workers in the secondary industries in
1999. The difference in the log of GDP per capita is defined as the difference between the log
GDP per capita of the local prefecture and that of the prefecture with the highest per capita
GDP in 1999 (i.e., Shenzhen). The log of distance is defined as the logarithm of the
geographic distance between the local prefecture and Shenzhen. The regressions are
estimated using the size of the population aged 16-18 in rural areas in the corresponding
prefecture and year as weight. Standard errors in parentheses are clustered at the prefecture
level. ***, **, and * represent the 1%, 5%, and 10% significance levels, respectively.
(1) (2) (3) (4)
Tariff (Local)
0.012** 0.016*** 0.012** 0.011*
(0.006) (0.005) (0.006) (0.006)
Tariff (Spillover,
from prefectures with b elow median migrant workers
0.077** 0.082* * 0.075* * 0.072*
in s eco nd ary ind us tries )
(0.035) (0.037) (0.036) (0.037)
Tariff (Spillover,
from prefectures with a bove median mig rant wo rkers
0.320*** 0.227*** 0.120** 0.121**
in s eco nd ary ind us tries )
(0.088) (0.083) (0.052) (0.051)
Prefecture-s ample fixed effects Ye s Ye s Ye s Ye s
Province-year fixed effects Ye s Ye s Ye s Ye s
Difference in log GDP per cap ita in 1999 × Year FE No Ye s Ye s Ye s
Lo g d is ta nc e × Ye ar FE No Ye s Ye s Ye s
Prefecture characteristics in 1999 (OLS) × Year FE No No Ye s No
Prefecture characteristics in 1999 (LASSO) × Year FE No No No Ye s
Mean of high school enrollment rate 0.442 0.442 0.442 0.442
Adjusted R-squared
0.962 0.969 0.972 0.972
Number o f p refec tures
294 294 294 294
Obs erv atio ns
4,704 4,704 4,704 4,704
High s chool en ro llment rate (16-18 y ears old)
Electronic copy available at: https://ssrn.com/abstract=4101730

44
Table 7. The impact of tariff shocks on employment growth rate in the secondary
industries
Notes: Panel A reports the impact on employment growth rate of people who resided in
certain prefecture (including both local residents and immigrants) in certain year, whereas
Panel B reports the impact on employment growth rate of emigrants from certain prefecture
in certain year. The difference in the log of GDP per capita is defined as the difference
between the log GDP per capita of the local prefecture and that of the prefecture with the
highest per capita GDP in 1999 (i.e., Shenzhen). The log of distance is defined as the
logarithm of the geographic distance between the local prefecture and Shenzhen. The
regressions are estimated from the size of the population aged 16-18 in rural areas in the
corresponding prefecture and year as weight. Standard errors in parentheses are clustered at
the prefecture level. ***, **, and * represent the 1%, 5%, and 10% significance levels,
respectively.
(1) (2) (3) (4)
Tariff (Local)
-0.101** -0.097** -0.094** -0.092**
(0.040) (0.040) (0.047) (0.042)
Tariff (Spillover)
-0.223 -0.205 -0.021 0.042
(0.279) (0.287) (0.296) (0.269)
Mean of employment growth rate 0.021 0.021 0.021 0.021
Adjusted R-squared 0.227 0.244 0.348 0.303
Number o f p refec tures
294 294 294 294
Obs erv atio ns
4,704 4,704 4,704 4,704
Tariff (Local)
-0.005 -0.006 -0.005 -0.004
(0.007) (0.006) (0.006) (0.007)
Tariff (Spillover)
-0.419** -0.359** -0.177* -0.209**
(0.189) (0.181) (0.099) (0.101)
Mean of employment growth rate 0.023 0.023 0.023 0.023
Adjusted R-squared 0.847 0.859 0.868 0.868
Number o f p refec tures
294 294 294 294
Obs erv atio ns
4,704 4,704 4,704 4,704
Prefecture-s ample fixed effects Ye s Ye s Ye s Ye s
Province-year fixed effects Ye s Ye s Ye s Ye s
Difference in log GDP per cap ita in 1999 × Year FE No Ye s Ye s Ye s
Lo g d is ta nc e × Ye ar FE No Ye s Ye s Ye s
Prefecture characteristics in 1999 (OLS) × Year FE No No Ye s No
Prefecture characteristics in 1999 (LASSO) × Year FE No No No Ye s
Emplo ymen t g ro wth ra te in t he s ec on dary in d us t rie s
Panel A: Local Residents
Panel B: Emigrants
Electronic copy available at: https://ssrn.com/abstract=4101730

45
Table 8. The impact of tariff shocks on the high school enrollment rate among rural
youth
Notes: The difference in the log of GDP per capita is defined as the difference between the
log GDP per capita of the local prefecture and that of the prefecture with the highest per
capita GDP in 1999 (i.e., Shenzhen). The log of distance is defined as the logarithm of the
geographic distance between the local prefecture and Shenzhen. The regressions are
estimated from the size of the population aged 16-18 in rural areas in the corresponding
prefecture and year as weight. Standard errors in parentheses are clustered at the prefecture
level. ***, **, and * represent the 1%, 5%, and 10% significance levels, respectively.
(1) (2) (3) (4) (5) (6)
Tariff (Local)
0.016*** 0.017*** 0.013** 0.016** * 0.017*** 0.013* *
(0.005) (0.005) (0.005) (0.006) (0.005) (0.005)
Tariff (Spillover,
using 16-24 years old migrants as weights)
0.570*** 0.270* * 0.315* * * 0.573* * * 0.276** 0.322** *
(0.184) (0.126) (0.120) (0.182) (0.126) (0.119)
Tariff (Spillover,
using 25-60 years old migrants as weights)
0.064 0.167 0.019
(0.107) (0.112) (0.080)
Tariff (Spillover,
using 25-44 years old migrants as weights)
0.075 0.155 0.021
(0.107) (0.103) (0.081)
Tariff (Spillover,
using 45-60 years old migrants as weights)
-0.011 0.008 -0.008
(0.021) (0.027) (0.019)
Prefecture-s ample fixed effects Ye s Ye s Yes Ye s Ye s Yes
Province-year fixed effects Ye s Ye s Ye s Ye s Ye s Yes
Diff. in lo g GDP per cap ita in 1999 × Year FE No Ye s Ye s No Yes Ye s
Lo g d is ta nc e × Ye ar FE No Ye s Ye s No Ye s Ye s
Prefecture char. in 1999 (LA SSO) × Year FE No No Ye s No No Ye s
Mean of high school enrollment rate 0.442 0.442 0.442 0.442 0.442 0.442
Adjusted R-squared
0.963 0.969 0.972 0.963 0.969 0.972
Number o f p refectures
294 294 294 294 294 294
Obs erv atio ns
4,704 4,704 4,704 4,704 4,704 4,704
High s chool en ro llment rate (16-18 y ears old)
Electronic copy available at: https://ssrn.com/abstract=4101730

46
APPENDIX
A. Selection of Prefecture Characteristics
To control for the potential endogeneity of the initial migration networks,
we incorporate a range of baseline prefecture characteristics that might affect
migration rates.
Specifically, the dependent variable is prefecture level migration rates in
2000, and the explanatory variables are a range of prefecture characteristics in
1999 (obtained from the China City Statistical Yearbook (2000)). We use two
methods to find variables that are significantly correlated with migration rates,
as shown in the following.
A1. Selection by the Ordinary Least Squares (OLS)
Traditionally, we could construct the objective function
!"#
!
$
%&
'
(")*"+,"
′
-.#
$
"%&
/
where
*"
is the migration rates from prefecture
0
in 2000;
,"
are a vector of
characteristics of prefecture
0
in 1999 (including an intercept term);
("
is the
size of the population aged 16-18 in rural areas from prefecture
0
in 2000, and
("
is used as weight; parameters
-
are chosen to minimize the objective
function; and
&
is the number of the prefectures in our study. The estimation
results are shown in Table A1.
We include a prefecture characteristic in Equation (3) in the main text if it
is significant at least at the 10% level.
A2. Selection by the Least Absolute Shrinkage and Selection Operator
(LASSO)
We also use LASSO to select prefecture level variables. The objective
function of LASSO is as following:
Electronic copy available at: https://ssrn.com/abstract=4101730

47
!"#
!
$
%&
'
(")*"+,"
′
-.#1 23-3&
$
"%&
/
where
3-3&
is the
4&
-norm of parameters
-
(i.e., 5
3-'3
'
), and
2
is a tuning
parameter to control the overall penalty of parameters
-
.
Once the tuning parameter
2
is chosen, this objective function yields a set
of deterministic parameter estimates. In our study, we use the cross-validation
to determine the value of tuning parameter
2
. To reduce arbitrariness, we use
the function cv.glmnet from the R Package glmnet. Specifically, it adopts the
most regularized model within one standard error of the minimized
cross-validated error (Friedman et al., 2010). We include all prefecture
characteristics
,"
in Equation (3) if their estimates are not equal to zero.
Electronic copy available at: https://ssrn.com/abstract=4101730

48
Table A1. Results from OLS and LASSO
Migration Rate in 2000
OLS
LASSO
Ln(GDP)
-0.015*
(0.009)
Share of primary industry in GDP
0.013
(0.023)
Share of secondary industry in GDP
0.014
(0.023)
Share of tertiary industry in GDP
0.013
(0.023)
Ln(Population)
0.010*
(0.006)
Ln(Non-agricultural population)
0.001
(0.004)
Population growth rate
0.001**
0.00016
0.000
Ln(Population density)
0.041**
(0.017)
Ln(Total # of workers)
0.012*
(0.006)
Ln(Self-employed workers in urban areas)
0.002
(0.003)
Share of workers in primary industry
0.004
(0.025)
Share of workers in secondary industry
0.004
(0.025)
Share of workers in tertiary industry
0.005
(0.025)
Ln(Employed workers)
0.031
(0.057)
Ln(Total wage)
-0.005
(0.057)
Ln(Average wage)
-0.002
(0.055)
Ln(Total fixed assets)
-0.008*
-0.00114
(0.004)
Ln(Investment in housing)
0.002
(0.002)
Ln(Investment in real estate)
-0.004*
(0.002)
Ln(# of hospitals)
0.000
Electronic copy available at: https://ssrn.com/abstract=4101730

49
(0.003)
Ln(# of inpatient beds)
-0.007
(0.005)
Ln(# of doctors)
-0.006
(0.005)
Ln(# of books in libraries)
0.005
(0.004)
Ln(# of books per 100 persons)
-0.005
(0.004)
Ln(Total output in primary industry)
0.005
(0.007)
Ln(Area)
0.047***
(0.016)
Ln(Area of built-up)
-0.002
(0.004)
Ln(Area of arable lands)
-0.02
(0.018)
Ln(Area of arable lands per capita)
0.004
-0.00768
(0.019)
Ln(Meat output)
-0.003
(0.004)
Ln(Aquatic output)
0.001
0.00142
(0.005)
Ln(Vegetable output)
-0.044***
(0.016)
Ln(Fruit output)
-0.002*
(0.001)
Ln(Milk output per capita)
-0.003*
-0.00293
(0.002)
Ln(Poultry output per capita)
0.006**
(0.003)
Ln(Egg output per capita)
-0.004
-0.00198
(0.003)
Ln(Aquatic output per capita)
0.004
0.00255
(0.006)
Ln(Vegetable output per capita)
0.045***
(0.016)
Ln(Output of above-scale enterprises)
-0.020
-0.00246
(0.021)
Ln(# of above-scale enterprises)
0.001
(0.005)
Ln(Sales of above-scale enterprises)
0.009
Electronic copy available at: https://ssrn.com/abstract=4101730

50
(0.021)
Ln(Profits of above-scale enterprises)
0.000
(0.003)
Ln(VAT payable of above-scale enterprises)
0.002
-0.00455
(0.008)
Ln(Net fixed assets of above-scale enterprises)
-0.007
-0.00247
(0.006)
Ln(Current assets of above-scale enterprises)
-0.006
-0.00269
(0.007)
Ln(Passengers)
0.046***
0.00361
(0.015)
Ln(Passengers by railway)
0.002
(0.002)
Ln(Passengers by road)
-0.036***
(0.013)
Ln(Freight)
0.006
(0.008)
Ln(Freight by railway)
-0.003
(0.002)
Ln(Freight by road)
-0.002
(0.006)
Ln(# of post office)
0.005**
0.00097
(0.002)
Ln(# of FDI contracts)
0.000
(0.002)
Ln(Contract value of FDI)
0.001
(0.001)
Ln(Real value of FDI)
0.001
(0.001)
Constant
-2.134
0.17530
(3.442)
Adjusted R-squared
0.686
Observations
294
294
Notes: For prefectures with missing share of primary industry in GDP, share
of secondary industry in GDP, share of tertiary industry in GDP, share of
workers in primary industry, share of workers in secondary industry, and share
of workers in tertiary industry, we use provincial level average value to impute
them. Therefore, the sum of the first three variables and the sum of the latter
three variables are not necessarily equal to one such that we can include all of
them in the regression.
Electronic copy available at: https://ssrn.com/abstract=4101730

51
B. Supplementary Table
Table B1. The impact of tariff shocks on the high school enrollment rate among urban
youth
Notes: The difference in the log of GDP per capita is defined as the difference between the
log GDP per capita of the local prefecture and that of the prefecture with the highest per
capita GDP in 1999 (i.e., Shenzhen). The log of distance is defined as the logarithm of the
geographic distance between the local prefecture and Shenzhen. The regressions are
estimated from the size of the population aged 16-18 in urban areas in the corresponding
prefecture and year as weight. Standard errors in parentheses are clustered at the prefecture
level. ***, **, and * represent the 1%, 5%, and 10% significance levels, respectively.
(1) (2) (3) (4)
Tariff (Local)
-0.015 -0.011 -0.013 -0.015
(0.018) (0.019) (0.020) (0.019)
Tariff (Spillover)
0.086 0.110 0.101 0.083
(0.083) (0.087) (0.096) (0.095)
Prefecture-s ample fixed effects Ye s Ye s Ye s Ye s
Province-year fixed effects Ye s Ye s Ye s Ye s
Difference in log GDP per cap ita in 1999 × Year FE No Ye s Ye s Ye s
Lo g d is ta nc e × Ye ar FE No Ye s Ye s Ye s
Prefecture characteristics in 1999 (OLS) × Year FE No No Ye s No
Prefectures characterist ics in 1999 (LASSO) × Year FE No No No Ye s
Mean of high school enrollment rate 0.866 0.866 0.866 0.866
Adjusted R-squared
0.900 0.909 0.912 0.912
Number o f p refec tures
294 294 294 294
Obs erv atio ns
4,704 4,704 4,704 4,704
High s chool en ro llment rate (16-18 y ears old)
Electronic copy available at: https://ssrn.com/abstract=4101730

52
C. Construction of Employment Growth Rates
We show how we construct employment growth rates of secondary
industry for local residents and emigrants of each prefecture. This variable is
used in Section 6.2.
We first obtain the total number of workers in the secondary industry
(including industries of manufacturing; construction; mining; and the
production and supply of electricity, heat, gas and water) from the China City
Statistical Yearbook for each prefecture in each year, denoted as
6789:8"(
'
with c representing prefecture and t representing year. We calculate the
employment growth rate of residents using the midpoint method as follows:
6789:8"(
'+6789:8")(*&
'
)6789:8"(
'16789:8")(*&
'.;<;%=
In order to calculate employment growth rate of emigrants in the second
industry, we calculate the total number of emigrants in the second industry
using the following formula:
6789:8"(
+>
'
?"
′
)(%#,,-
"
@"
′
)(%#,,-
"
′
." 6789:8"
′
(
'/
where
6789:8"(
+
is the number of emigrant workers in the second industry
from prefecture c in year t.
6789:8"
′
(
'
is the number of residential workers in
the secondary industry in prefecture
0
′
in year t,
?"
′
)(%#,,-
"
is the number of
emigrants from prefecture c to prefecture
0
′
in 2005, and
@"
′
)(%#,,-
is the
number of residents in prefecture
0
′
in 2005. Then, we calculate the
employment growth rate of emigrants using the midpoint method as follows
6789:8"(
++6789:8")(*&
+
)6789:8"(
+16789:8")(*&
+.;<;%=
Electronic copy available at: https://ssrn.com/abstract=4101730

53
The average annual growth rate of the employment in the secondary
industry among the residents and the emigrants in all prefectures in China from
2000-2015 is 2.1% and 2.8%, respectively. The employment growth rates
fluctuated over the period. While the employment in the secondary industries on
average decreased in years 2000 and 2015 for both residential people and
emigrants, the employment increased in 2005 and 2010—the years that
witnessed the most rapid expansion of the secondary industry of China,
particularly among the emigrant population.
Electronic copy available at: https://ssrn.com/abstract=4101730