Journal of Development Economics 101 (2013) 133–147

Contents lists available at SciVerse ScienceDirect

Journal of Development Economics

journal homepage: www.elsevier.com/locate/devec

The economic impact of Special Economic Zones: Evidence from

Chinese municipalities☆
Jin Wang ⁎
Division of Social Science, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong

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

Article history: The paper exploits a unique Chinese municipal dataset to assess the impact of Special Economic Zones on the
Received 4 October 2010 local economy. Comparing the changes between the municipalities that created a SEZ in earlier rounds and
Received in revised form 1 September 2012 those in later waves, I find that the SEZ program increases foreign direct investment not merely through
Accepted 31 October 2012
firm relocation, and does not crowd out domestic investment. With dense investment in the targeted
municipality the SEZ achieves agglomeration economies and generates wage increases for workers more
JEL classification:
O16
than the increase in the local cost of living. The effects are heterogeneous: for zones created later the benefits
O47 are smaller while the distortions in firm location behavior are larger than those for the early zones.
F21 Municipalities with multiple SEZs experience larger effects than those with only one SEZ.
R10 © 2012 Elsevier B.V. All rights reserved.

Keywords:
Special Economic Zone
Foreign direct investment
TFP growth
Factor price

1. Introduction
Economists have long debated the potential benefits and distortions
associated with the spatially targeted programs.1 More recently, the
agglomeration economies have been rigorously identified that explain
productivity advantages for firms located in denser areas (Combes et
al., forthcoming; Greenstone et al., 2010; Kline and Moretti, 2011),
while the efficiency losses from mobile workers and firms relocating
across the boundaries of targeted areas are found to be modest in the
case of US Federal Empowerment Zones (Busso et al., forthcoming).
Despite the increasingly sophisticated work on place-based policies,
there is a tremendous lack of empirical evidence for evaluating such
programs in the context of developing countries. 2
☆ I am indebted to Oriana Bandiera and Timothy Besley for their guidance. I thank Eric
Verhoogen, two anonymous referees, Joshua Angrist, Robin Burgess, Rajeev Dehejia, Greg
Fischer, Maitreesh Ghatak, Henrik Kleven, Guy Michaels, Gerard Padró i Miquel, Albert
Park, Steve Pischke, Mark Schankerman, Cheng-gang Xu and Alwyn Young for their com-
ments. Seminars at the LSE, the RES UK Conference, the University of Oxford, the HKUST,
and the Nanyang Technological University Singapore made helpful comments to improve
the work. I am also grateful to Junxin Feng for support with the data collection, and to the
LSE STICERD members as well as Ruixue Jia for helpful discussions. All errors remain my own.
⁎ Tel.: +852 2358 7834; fax: +852 2335 0014.
E-mail address: sojinwang@ust.hk.
1
See Glaeser and Gottlieb (2008), Glaeser et al. (2010), Greenstone and Looney
(2010), and Moretti (forthcoming).
2 4
See Peters and Fisher (2002) for reviews on the UK enterprise zones and more re-
cent US empowerment zones and regional development initiatives.
To fill that gap this study takes advantage of the gradual establish-
ment of Special Economic Zones (SEZs) across Chinese municipalities
since 1979, which constitutes a unique laboratory for a large-scale
study of SEZs. Special Economic Zones are contained geographic re-
gions within a country with more liberal laws and economic policies
to encourage foreign-invested manufacturing and services for export
(Shah, 2008). Fig. 1 displays the significant correlation between the
SEZ experiment and FDI outcome in China. 3 Worldwide there were
approximately 3000 SEZs in 135 countries in 2008, accounting for
over 68 million direct jobs and over US$ 500 billion of direct trade-
related value added within the zones (World Bank, 2008). Like
many place-based programs, the SEZs attempt to foster agglomera-
tion economies – they promote firm interactions that increase
productivity in dense areas – by building clusters or attracting tech-
nologically advanced industrial facilities (Combes et al., 2011).
The question of whether SEZs have meaningful effects on the local
economy therefore has great policy relevance, and yet previous
research on SEZs consists mainly of case and theoretical studies. 4
My main objective in this paper is to quantify the impact of the SEZ
programs and explore the mechanisms through which the effects
work. Kline (2010) and Busso et al. (forthcoming) are the two closest
predecessors to my investigation in framework and method. In the
3
See Prasad and Wei (2007) and Feenstra and Wei (2010).
See Willmore (1996), Kung (1985), Ge (1999), Park (1997), Rolfe et al. (2004),
Aggarwal et al. (2008), Aggarwal (2005) and Litwack and Qian (1998).

0304-3878/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jdeveco.2012.10.009
134 J. Wang / Journal of Development Economics 101 (2013) 133–147

Fig. 1. SEZs, FDI and trade outcome: national aggregate statistics. Notes: the graph displays the significant correlation between the SEZ experiment and FDI related outcome
including Foreign Direct Investment, Exports, Imports and the proportion of foreign invested enterprises' industrial output in China.
 J. Wang / Journal of Development Economics 101 (2013) 133–147
context of U.S. place based programs they developed a spatial equilib-
rium model with landlords, firms, and mobile workers and show that
the incidence and efficiency of local subsidies depend critically on
“the degree of preference heterogeneity in the population and the
structure of any agglomeration economies” (Kline, 2010, p.383).
Building on their work, this study proceeds to examine the effect of
Chinese SEZs on agglomeration economies as well as on firm and
worker behavior. In particular, after showing that investment in the
subsidized municipality increases, I estimate the elasticity of the mu-
nicipal total factor productivity (TFP) growth with respect to the SEZ
program. I further distinguish the scenario that the SEZ program sim-
ply shifts firms from one municipality to another from the case that it
creates new activity. I also assess through the local price change
whether the SEZ program's benefits were arbitraged away by
workers' migration. Finally, I evaluate whether the effects are hetero-
geneous across municipalities belonging to different granting waves
as well as receiving different SEZ treatment intensities.
For the analyses I use a large number of official sources to
construct a novel dataset covering 321 Chinese prefecture-level
municipalities between 1978 and 2008. Detailed data allows for an
examination of China's municipal economies before, during and
after the expansion of SEZs, and the empirical analysis compares
changes between municipalities establishing SEZs earlier and later,
as well as between municipalities with multiple SEZs and only one
SEZ, conditional on a rich set of control variables. To the best of my
knowledge, Wei (1995) and Alder et al. (2012) have been the only ex-
ceptions to use city-level data to assess the effect of the SEZ policies.
Wei (1995) examines the relationship between the SEZ policy and
growth, but the limited data from 1980 to 1990 prevents him from
conducting a comprehensive evaluation. A recent paper by Alder et
al. (2012) exploits a panel dataset covering 270 Chinese cities for
23 years to examine the impact of SEZs on city GDP growth. That
work, however, examines the state-level zones approved by the cen-
tral government, which are a subset of the SEZs considered in my
study. Moreover, their work focuses on GDP growth while this
paper investigates a rich set of welfare indicators such as investment,
TFP growth and factor prices.
The validity of the standard Difference in Difference (DID) strategy
and the causal interpretation given to the results relies on the as-
sumption that the later moving municipalities are a valid counterfac-
tual for what would have happened to the earlier adopters in the
absence of a SEZ program. However, the timing of the SEZ programs
is not likely to be random. The earlier group might differ substantially
from the later adopters in terms of a set of factors such as geographic
location, industrial conditions, human capital, infrastructure, financial
development, factor prices and other unobserved characteristics. To
tackle the identification challenge, the arbitrary permanent heteroge-
neity between earlier and later adopting municipalities in their
unobserved characteristics is addressed by allowing for municipality
fixed effects. I also include flexible province-year fixed effects to
control for differential province-specific time effects such as macro
shocks. Beyond that, I use detailed information on municipal charac-
teristics to construct meaningful comparable groups. I match each
municipality which established SEZs in the first wave with its closest
counterparts that established SEZs in the second wave in terms of
various characteristics and, more importantly, historical trends of
main outcomes. A similar approach is then adopted for the later
rounds. The municipalities of adjacent rounds are similar in terms of
observable characteristics, and so perhaps more likely to be similar
in unobserved traits as well.
The analysis reveals several important findings. First, the SEZ
program has an overall positive effect on investment. The introduc-
tion of a SEZ program on average significantly increases the level of
per capita FDI by 21.7% and the growth rate of FDI by 6.9 percentage
points. On the other hand the SEZ program neither crowds in nor
crowds out domestic investment. Moreover, there exists a sizable
135
creation effect by the SEZ in the municipality and a partial diversion
effect that increases with the number of neighboring SEZs. By 2008
the municipality's own SEZ program had on average increased per
capita FDI by 112% while adjacent SEZs had diverted away per capita
FDI by 33%.
Second, the SEZ program generates significant agglomeration
economies. It increases the technological progress of the earlier treat-
ed municipalities by 1.6 percentage points compared to the counter-
parts that carried out the program later, an effect which builds up
gradually after a SEZ program is in place.
Third, the average wage of workers in the treatment group in-
creases by 8% more than in the control group while there is a 5%
rise in the cost of living. The estimates of factor prices suggest that
the benefits of the SEZ program are not completely arbitraged away
by worker mobility.
Fourth, the early zones overall experience a larger increase in in-
vestment, TFP growth and factor prices. The early zones' FDI creation
effect in particular is much larger than the diversion effect. In con-
trast, the newly attracted FDI of zones established later is not much
larger than the reallocation of activities. These results imply that
once the whole country becomes liberalized, SEZs might generate
large distortions in firm behavior as later zones tend to be closer
substitutes for each other (Busso et al., forthcoming).
Finally, municipalities with multiple SEZs exhibit greater FDI at-
traction, agglomeration economies and factor price changes relative
to those with only one SEZ.
In addition to the research on place-based policies, my paper adds
to several other strands of literature. First, the paper evaluates the
FDI's impact of the SEZ program at the municipal level and so builds
a bridge between country-level and firm-level studies of FDI. 5 The de-
sign balances the inferential validity and the understanding of FDI's
macro-level impact on the local economy. The findings that FDI
brought by the SEZs enhances the municipal investment and techno-
logical progress is consistent with the results of studies of Chinese
manufacturing and exporting firms (Fernandes and Tang, 2012; Liu,
2008). Second, methodologically this study follows the lead of
Persson and Tabellini (2008), Abadie (2005), Heckman et al. (1997)
and Blundell et al. (2004) in combining matching and difference in
differences to analyze the effects of reform in a macroeconomic con-
text. The matching technique controls for selection into the SEZ pro-
gram at different times and allows for estimating the treatment
effects in a non-parametric way.
The next section starts with the historical background of China's
SEZ experiment. Section 3 provides a brief description of the dataset.
Section 4 estimates the average impact of the SEZ experiment on local
economic outcomes. Section 5 further explores the heterogeneity in
the SEZ granting sequence as well as the program intensity to
examine the different impacts of the SEZs. Section 6 offers concluding
remarks.
2. Background
In this section I discuss some essential features of the SEZ experi-
ment. China's administrative system has five hierarchical levels of
government: (1) central; (2) provincial; (3) prefecture; (4) county;
and (5) township. This paper focuses on the prefecture level where
the SEZ experiments have been carried out.
In the late 1970s, China's State Council approved small-scale SEZ
experiments in four remote southern cities: Shenzhen, Zhuhai and
Shantou in Guangdong Province, as well as Xiamen in Fujian Province.
5
Some scholars view FDI as an important source of capital using country level data,
including Whalley and Xin (2006), McGrattan and Prescott (2009), Desai et al. (2009).
Others focus on FDI as an important source of technology spillover, for example Coe et
al. (2009) [cross-country study], Liu (2008), Hale and Long (2007) and Abraham et al.
(2010) [firm-level study].
136 J. Wang / Journal of Development Economics 101 (2013) 133–147
China started with virtually zero foreign direct investment and almost
negligible foreign trade before 1978, so those zones were considered
a test base for liberalization of trade, tax and other policies
nationwide. In August 1980 the People's Congress passed the first
Regulation for Guangdong SEZs. This regional law was the first of its
kind to be tested, drafted with the help of legal experts sent from
the central government (Cai, 2008; Xu, 2011). When the experiment
was later expanded into other provinces, this was the law they
adopted and modified. 6 The law explicitly provides the following
policy package for foreign investors:
1) Private Property Rights Protection: the SEZs encourage foreign cit-
izens, overseas Chinese, and compatriots from Hong Kong and
Macau to set up enterprises and other establishments on their
own or in joint ventures with Chinese partners. The SEZs guaran-
tee to protect their assets, accrued profits and other rights. This is
a very important commitment by the Chinese government, since
there was no constitutional protection of private property rights
outside the SEZs until a constitutional amendment in 2004.
2) Tax incentives: foreign investors are promised a reduced corporate
income tax rate of 15–24% further differentiated on the basis of the
technological content of their products, compared to the 33% paid
by domestic firms. They are to bear virtually zero custom duties
and enjoy duty free allowances for production materials. There are
income tax exemptions for foreigners working in the SEZs as well.
3) Land use policy 7: under Chinese law, all land is under state own-
ership. In the SEZs foreign investors may lawfully obtain the rights
for land development and business use. They may also transfer or
lease land rights, or mortgage them for stipulated purposes and
terms of use. When foreigners invest in projects encouraged by
the state for an operational term of more than 15 years they are
exempt from land use fees for five years and pay half the usual
fee for the following five years. The land use right is guaranteed
for projects that have a total investment of at least US
$10 million, or that are considered to be technologically advanced
with a major influence on local economic development.
A distinctive feature of the SEZ program is its decentralized imple-
mentation (Huang, 1998; Xu, 2011). An administrative committee,
commonly selected by the local government, oversees the economic
and social management of the zone, including approving the FDI pro-
jects up to a certain limit, building and improving the infrastructure,
and regulating the land use on behalf of the local administration
(Zeng, 2011). The government made clear the targets of SEZs in terms
of four principles: “Construction primarily relies on attracting and uti-
lizing foreign capital; primary economic forms are Sino-foreign joint
ventures and partnerships as well as wholly foreign-owned enterprises;
products are primarily export-oriented; economic activities are primar-
ily driven by market forces.”
Supported by the initial achievements of the first group of SEZs,
the central government expanded the SEZ experiment in 1984. Four-
teen other coastal cities were opened to foreign investment. 8 From
1985 to 1988 the central government included even more coastal mu-
nicipalities in the SEZ experiment. 9 In 1990 the Pudong New Zone in
Shanghai was opened to foreign investment along with other cities in
the Yangtze River valley. The pattern of granting SEZ status in those
earlier years is not random. According to State Council documents, 10
the central government authorized municipalities to establish the
SEZs based on their better geographical location, industrial condition
6
See Zhongfa (1979).
7
Source: the government website of Zhejiang Province.
8 13
Listed north to south: Dalian, Qinhuangdao, Tianjin, Yantai, Qingdao, Lianyungang,
Nantong, Shanghai, Ningbo, Wenzhou, Fuzhou, Guangzhou, Zhanjiang, and Beihai.
9
Listed north to south: Liaodong Peninsula, Hebei Province (which surrounds
Beijing and Tianjin), Shandong Peninsula, Yangtze River Delta, Xiamen-Zhangzhou-
Quanzhou Triangle in southern Fujian Province, Pearl River Delta, and Guangxi.
10
See various State Council documents issued in 1984,1985,1987,1988,1991,1992,1993.
and human capital. 11 From 1992 to 1994, the State Council opened a
number of border cities and all the capital cities of the inland prov-
inces and autonomous regions. In addition, 222 state-level economic
zones and 1346 province-level economic zones were gradually
established within municipalities to provide better infrastructure
and achieve agglomeration of foreign investors. 12 As a result, a
multi-level and diversified pattern of opening coastal areas and inte-
grating them with river, border, and inland areas took shape in China.
The Chinese SEZ program is described by the World Bank as a unique
zone-within-zone case because large opened economic zones (the
whole municipality) hosted smaller zones (state-level and province-
level economic zones) within their territory.
According to the laws and regulations, open economic areas,
state-level and province-level economic zones maintain no systematic
tax policy differences towards foreign investors. There are, however,
some differences in SEZ administration.13 Open economic areas and
state-level economic zones have higher level administrative committees
than provincial level SEZs and their committees enjoy more authority in
managing the zones. For example, they are allowed to approve FDI pro-
jects up to a higher threshold. They are also given priority access to
state-owned land and SEZ-related infrastructure loans.14
Fig. 2 displays the geographic evolution of the SEZ experiment.
Table 1 presents descriptive characteristics of the four big SEZ granting
waves: 1978–1985, 1986–1990, 1991–1995, and 1996–2008. The pro-
portion of municipalities with SEZs was 0% in 1978, 9% in 1985, 24% in
1990, 69% in 1995 and 92% in 2008. The SEZ experiment was expanded
from coastal, industrially more developed locations to inland, industri-
ally less-developed areas. Municipalities that implemented the SEZ
program earlier had higher land prices than those in later rounds.
They also had better infrastructure including highway density, airports,
ports, telecommunications, and better financial development. But there
were no significant differences in human capital across the four groups
of municipalities that established the SEZs in different periods.
3. The data
In order to evaluate the impact of SEZs, I constructed a new panel
dataset on 321 Chinese prefecture-level municipalities. 15 Detailed
data contains information on GDP, investment, employment, exports
and factor prices, as well as a digital GIS map of Chinese municipali-
ties that is coded with the year the SEZ was created. Data Appendix
I contains more details on the construction of these variables.
3.1. Special Economic Zone Index
The dataset comprises the SEZ information detailed in Appendix II:
1. Lists of coastal and inland municipalities which have been granted
open economic area status and the year of granting;
2. Lists of state-level Economic and Technological Development
Zones, New and High-technology Industrial Development Zones,
Export Processing Zones, and Border Economic Cooperative
Zones, the size of the zones and the year when the municipalities
were authorized to establish them;
3. Lists of province-level Economic and Technological Development
Zones, and New and High-technology Industrial Development
11
China's development strategy based on location is discussed in Démurger et al.
(2002).
12
State-level SEZs are authorized by the central government; Province-level SEZs are
authorized by provincial governments. These zones are typically located in the subur-
ban regions of a major city (Zeng, 2011).
The industrial preference of some zones is discussed in Alder et al. (2012).
14
See Guofa (2005, 2010), Shangzifa (2006, 2008, 2009), Caijian (2010) and
Fagaidiqu (2010).
15
I drop provincial-level municipalities including Beijing, Shanghai, Tianjin and
Chongqing from the sample to address the concern that they are not comparable with
prefecture-level municipalities.
 J. Wang / Journal of Development Economics 101 (2013) 133–147 137

Fig. 2. The geographic evolution of the Special Economic Zone experiment. Notes: if the whole municipality was granted the status of open economic area; or within the munici-
pality, only a certain geographical area was allowed to establish state-level economic zones, or province-level economic zones, the municipality was entitled to use preferential
policies (including property rights protection, tax breaks, cheaper land bills, etc.) to attract foreign direct investment. Therefore, I define the municipality to be a Special Economic
Zone (SEZ) from a general prospective.

Zones, the size of each zone and the year when the municipality
was authorized to establish it.
Being granted open economic area status means that the whole area
of the municipality is a large SEZ for foreign investors. Being authorized
to establish a state-level or province-level economic zone means that
within the municipality a certain geographical area can be used as a SEZ
to host foreign investors. In the full sample, some municipalities were
granted open economic area status and then allowed to establish the
state-level and province-level economic zones in later years. A large SEZ
can thus contain multiple specific zones within its boundaries. Today,
some coastal municipalities such as Shenzhen, Dalian and Guangzhou
have such nested zones. In contrast, most inland municipalities were
not granted open economic area status. They have relatively fewer and
smaller economic zones within their city areas. The fact that the coastal
municipalities were allowed to establish more and larger SEZs is highly
correlated with their potential for attracting foreign direct investment.
I set a general SEZ dummy variable, Sipt, to one if an entire municipal-
ity was granted open economic area status, or if a state-level or
province-level economic zone was authorized within the municipality,
and zero otherwise.
Although in the baseline specification the evaluation of the SEZ policy
focuses on the average effects of being or not being in the SEZ program, I
also examine whether or not early zones exhibit different impacts relative
to later zones, along with whether municipalities with multiple zones ex-
perience different effects from those with only one. To do so, I group the
municipalities based on the SEZ granting waves: group 1 [1978–1985] is
composed of 28 municipalities that were exposed to the SEZ reform
138 J. Wang / Journal of Development Economics 101 (2013) 133–147

Table 1
The summary statistics of the SEZ granting sequence.

Timing Group 1 Group 2 Group 3 Group 4

[1978–1985] [1986–1990] [1991–1995] [1996–2008]

SEZs SEZs SEZs SEZs No SEZ

A. Granting sequence
Municipalities newly granted with SEZs 28 49 143 75 26
Municipalities with SEZs 28 77 220 295 295
Ratio of municipalities with SEZs 0.09 0.24 0.69 0.92 0.92
The SEZ program onset year 1984 1988 1993 2002 –
(2.02) (0.57) (1.09) (3.76) –

B. Municipal characteristics in 1978

Per capita industrial output (RMB) 622 603 425 280 271
(487) (626) (487) (303) (634)
Per capita secondary students (person) 0.06 0.06 0.07 0.06 0.05
(0.02) (0.02) (0.02) (0.02) (0.03)
Distance to the coast (100 miles) 0.15 1.34 3.76 4.76 10.12
(0.20) (2.33) (3.11) (3.61) (5.88)
Road density (km/square km) 0.24 0.22 0.22 0.15 0.07
(0.13) (0.10) (0.28) (0.08) (0.07)
Airport (=1 if in municipality) 0.07 0.06 0.13 0.12 0.27
(0.26) (0.24) (0.34) (0.33) (0.45)
Port (=1 if in municipality) 0.96 0.53 0.12 0.03 0.00
(0.19) (0.50) (0.32) (0.16) (0.00)
Per capita post and telecommunications (RMB) 2.8 1.8 2.0 1.6 1.8
(1.6) (1.3) (1.5) (1.2) (1.0)
Per capita deposits in financial institutions (RMB) 131 127 111 94 188
(104) (174) (148) (166) (244)
Per capita loans by financial institutions (RMB) 258 190 176 180 95
(178) (137) (124) (309) (80)
Average wage of workers (RMB) 564.1 573.9 596.1 599.1 770.9
(67.9) (83.8) (102.7) (95.9) (215.9)
Land price level 2.32 4.27 5.04 6.24 7.00
(1–7 highest to lowest) (1.12) (1.55) (1.51) (0.91) (0.00)

Notes: Standard Errors reported in brackets. Based on the timing of experimenting with SEZs, the sample is classified into four groups (group 1 [1978–1985], group 2 [1986–1990],
group 3 [1991–1995], and group 4 [1996–2008]).

between 1978 and 1985; group 2 [1986–1990] is composed of 49 munic- Bureau of Statistics acknowledges the potential bias, and apart from
ipalities that had the SEZ experiment between 1986 and 1990; group 3 annual revisions to the national income and product accounts data,
[1991–1995] is composed of 143 municipalities that were chosen to it has so far conducted two benchmark revisions in 1993 and 2006.
implement the SEZ program between 1991 and 1995; group 4 The first, following the 1993 tertiary (service) sector census, adjusted
[1996–2008] is composed of the remaining 101 municipalities. Moreover, the 1978–93 tertiary sector value added and, by implication, the sum
I define the treatment intensity parameter as one when a municipality
only has one treatment among state-level SEZs, provincial SEZs and
open economic areas. When a municipality has two of the programs,
the treatment intensity is defined as two. When a municipality has carried
out three programs, the treatment intensity is three. According to the
classification, 67 municipalities had been exposed to treatment intensity
3, 53 municipalities had been exposed to treatment intensity 2, and 175
municipalities had been treated with intensity 1 by the end of the sample
period.

3.2. Foreign direct investment

Data on utilized foreign direct investment, exports, and industrial

output by foreign-invested enterprises, are used to quantify the direct
outcomes of the SEZ experiments. Fig. 3 plots the sample mean of the log-
arithm of per capita foreign direct investment by year for the four groups
of municipalities classified based on the timing of their SEZ experiments.
The SEZs increase FDI significantly after each group of municipalities was
authorized to establish their SEZs. However, the effect seems to be much
Fig. 3. The SEZs on FDI outcome by groups. Notes: 321 prefecture level municipalities
stronger for the municipalities which carried out their SEZ experiments
are classified into four groups based on their timing of carrying out the Special
earlier. To prevent biased estimates due to potential selection problems, Economic Zone experiment. Group 1 is composed of municipalities that were exposed
more rigorous methods are used in the main specification. to the SEZ reform between 1978 and 1985 (1978–1985); group 2 is composed of mu-
nicipalities that had the SEZ experiment between 1986 and 1990 (1986–1990); group
3.3. Growth accounting data 3 is composed of municipalities that were chosen to implement the SEZ program be-
tween 1991 and 1995 (1991–1995); group 4 includes municipalities that had not car-
ried out the SEZ reform by 1995. The graph displays the sample mean of natural
Some scholars question the credibility of the data published by logarithm of per capita FDI by year by group without controlling for any municipal
China's statistical office (Holz, 2008; Young, 2003). China's National characteristics and macroeconomic shocks.
 J. Wang / Journal of Development Economics 101 (2013) 133–147
of the sectoral value added, i.e., gross domestic product (GDP). The
second benchmark revision occurred in early 2006 following the
2004 economic census of the secondary sector (industry, construc-
tion) and of the tertiary sector using OECD methods. The dataset for
this study uses the latest municipal statistics after these adjustments.
Following the lead of Caselli (2005) and of Young (2003), I have
constructed series for real GDP, real capital stock, human capital
augmented labor and share of labor income. It is noteworthy that
the municipal statistics only offer the total municipal employment,
which reflects the actual utilization of total labor force during a
certain period of time (National Statistical Bureau, 2002). There are
no disaggregated statistics on the number of employees within the
SEZs, nor on the residents living inside and outside the SEZ.
3.4. Factor prices
Average Wage of Workers refers to the average nominal wage
during a certain period for those who work full-time for state
owned enterprises and also those with collective, private, joint or for-
eign ownership. It reflects the general level of wage income in the
municipality, which however does not distinguish the wage level
within the SEZ from that outside the SEZ.
In terms of the cost of living, the National Statistical Bureau (NSB)
of China uses a consumer price index (CPI) to measure changes over
time in the amount consumers need to spend on a representative
consumption basket. More precisely, the CPI covers the prices of
tobacco and liquor; clothing; household facilities, items and mainte-
nance services; health care and personal articles; transportation and
communication; recreation, education and cultural articles and
services; and residence costs including rent, utilities and mainte-
nance. It is worth noting that home prices are not included in the
CPI because the CPI reflects the price change of households' current
consumption. The acquisition of, payment for, and use of a house do
not occur simultaneously. The volume of housing consumed in a
period may have little relationship to the payments made in the
period. Moreover, houses are an asset that many households pur-
chase and hold as an investment. Their prices reflect both the user
cost and the expectations of future appreciation. The NSB therefore
uses rents, utility bills and maintenance fees to estimate the user
cost of housing as one component of the CPI. As a complement to
the CPI, I also collect municipal housing transaction prices to examine
whether a SEZ program drives up local house prices. 16
4. Average SEZ effects
The empirical analysis relies on the variation in the timing of
when SEZs were created across the sample of municipalities. As
described in Section 2, the timing of the SEZ experiment provides
significant variations both between and within municipalities which
I will exploit in my identification strategy.
4.1. Event study
Before introducing a parametric specification, I conduct an event
study along the lines of Jacobson et al. (1993) for the primary
outcomes.
X
6
k
Y ipt ¼ α i þ γpt þ Dipt δk þ εipt ð1Þ
k≽−6;k≠−1
16
The data available since the 1990s denote the price of newly constructed residen-
tial buildings, and were calculated according to the floor space sold and the sale values
in real estate development reports. Because the transaction records have been reported
only since 2010, year-on-year data with the same scope as in the US cannot be
calculated.
139
Here the Yipts are the outcome variables including foreign direct
investment, domestic investment, TFP growth and factor prices in
k
municipality i of province p at time t. The dummy variables, Dipt ,
jointly represent the SEZ program designation event. In particular, si
denotes the year when municipality i carried out its SEZ experiment.
−6 − k
I further define Dipt = 1, if t − si ≤ − 6, and 0 otherwise; Dipt = 1, if
t − si = k, and 0 otherwise, k = − 5, − 4, − 3, − 2, 0, 1, 2, 3, 4, 5;
6+
Dipt = 1, if t − si ≥ 6, and 0 otherwise.
The dummy for k = − 1 is omitted so that the post-treatment
effects are relative to the period immediately prior to the start of
the program. Therefore, the parameter of interest δk identifies the
causal effect of the SEZ program k years following its occurrence,
assuming setting up a SEZ affects outcomes up to 6 years prior to
the program. αi is a fixed effect that captures permanent differences
in the municipalities' observed and unobserved characteristics, such
as underlying abilities and endowments which might influence FDI
performance. The γpts are the coefficients of a set of dummy variables
for each province in a given year in the sample period that capture the
general time pattern of primary outcomes in the same province. 17 To
deal with potential problems of serial correlation, I adopt a conserva-
tive approach in estimating standard errors and allow the disturbance
terms εipt to be clustered by municipality throughout. 18
Table 2 reports the estimates of Eq. (1) for a set of local economic
outcomes. To help visualize the dynamic effect, Fig. 4 displays the
point estimates of FDI along with the 95% confidence bands. The
municipalities seem to have absorbed substantial foreign direct
investment from the time their SEZ program took hold. Further, the
estimated effect of the SEZ on FDI intensifies following its establish-
ment. As is shown in Column (1), per capita FDI increases by an aver-
age of 6% in the year when the SEZ program is implemented relative
to the previous year. The magnitude becomes 75% six years after the
SEZ experiment begins. Moreover, no substantial increase or decrease
is evident during the periods before the SEZ experiment. Therefore,
my confidence in these results – the FDI attracted is large and
long-term and appears only after municipalities have carried out a
SEZ program – is enhanced. A similar dynamic pattern holds for a
rich set of outcomes including TFP growth, wages and the CPI though
there is no such increase in domestic investment and house prices, as
reported in Table 2. The dynamics of the SEZ's effect seem to be
meaningful since investment, technological progress and price
changes take time to accumulate.
4.2. Baseline specification
The point estimates in the event study suggest that the SEZ
program affects not only the levels but also the trends in primary out-
comes such as FDI outcomes, TFP growth, wages and the CPI. Specifi-
cally, I define the post-SEZ trend to be Fipt = t − si if t ≥ si and
0 otherwise, where si again denotes the year the municipality
launched its SEZ program. I also specify the SEZ to influence only
the level of outcomes such as domestic investment and house prices.
The specifications are as follows:
Y ipt ¼ α i þ γpt þ Sipt δ þ F ipt β þ εipt ; ð2aÞ
Y ipt ¼ α i þ γpt þ Sipt δ þ εipt : ð2bÞ
Sipt is the key variable indicating the SEZ experiment and is defined in
Section 3.1. All other controls are as previously defined. The effect of
17
An alternative specification using the municipality specific trends cannot fully ac-
count for the nonrandom SEZ assignment.
18
Clustering the disturbance terms by province–assume that municipalities in the
same province face common shocks–leads to the standard errors on the estimates be-
ing slightly larger than those reported.
140 J. Wang / Journal of Development Economics 101 (2013) 133–147

Table 2
An event study: the effects of the SEZs on local economy.

Dependent variable Capital investment TFP growth Prices

FDI Exports FIEs output Domestic investment Wage Consumer price index House prices

(1f) (2f) (3f) (4) (5) (6) (7) (8)

≥6 years before 0.122⁎ 0.011 0.289 0.026 −0.000 −0.004 0.012⁎ −0.108
(0.072) (0.113) (0.203) (0.057) (0.0046) (0.012) (0.006) (0.066)
5 years before 0.022 −0.113 0.305⁎ 0.020 −0.005 −0.006 0.003 −0.021
(0.055) (0.081) (0.173) (0.042) (0.005) (0.008) (0.004) (0.052)
4 years before −0.044 −0.097 0.075 0.048 −0.004 −0.003 0.004 −0.094⁎⁎
(0.054) (0.068) (0.149) (0.038) (0.005) (0.007) (0.003) (0.046)
3 years before −0.002 −0.025 0.060 0.078⁎⁎ 0.000 −0.001 0.003 −0.045
(0.047) (0.052) (0.133) (0.032) (0.005) (0.005) (0.002) (0.043)
2 years before 0.009 −0.013 0.003 0.055⁎⁎ 0.000 −0.004 0.002 0.048
(0.038) (0.038) (0.117) (0.028) (0.005) (0.004) (0.002) (0.040)
Year of change 0.064⁎ 0.169⁎⁎⁎ −0.025 0.041 −0.003 −0.010⁎⁎ 0.002 −0.008
(0.038) (0.041) (0.106) (0.027) (0.004) (0.005) (0.002) (0.036)
1 year later 0.278⁎⁎⁎ 0.312⁎⁎⁎ 0.235⁎ 0.031 −0.002 −0.002 0.005⁎⁎ 0.018
(0.058) (0.063) (0.129) (0.034) (0.004) (0.006) (0.003) (0.035)
2 years later 0.318⁎⁎⁎ 0.440⁎⁎⁎ −0.002 0.027 0.005 −0.004 0.006 0.038
(0.065) (0.078) (0.174) (0.039) (0.005) (0.007) (0.004) (0.043)
3 years later 0.386⁎⁎⁎ 0.591⁎⁎⁎ 0.260⁎ 0.035 −0.001 0.009 0.008⁎ −0.000
(0.071) (0.092) (0.143) (0.046) (0.005) (0.008) (0.005) (0.043)
4 years later 0.434⁎⁎⁎ 0.619⁎⁎⁎ 0.367⁎⁎ 0.018 0.005 0.008 0.008 −0.018
(0.078) (0.102) (0.162) (0.050) (0.005) (0.010) (0.006) (0.047)
5 years later 0.509⁎⁎⁎ 0.766⁎⁎⁎ 0.274 0.023 0.005 0.009 0.009 −0.018
(0.083) (0.108) (0.176) (0.054) (0.005) (0.011) (0.007) (0.046)
≥6 years later 0.749⁎⁎⁎ 1.110⁎⁎⁎ 0.473⁎⁎ −0.046 0.011⁎⁎ 0.038⁎⁎⁎ 0.019⁎ −0.048
(0.095) (0.144) (0.211) (0.073) (0.005) (0.015) (0.010) (0.052)
Obs. 9947 9946 3929 9933 9614 9933 9944 5170
R -sq 0.87 0.89 0.93 0.93 0.26 0.99 0.99 0.85

Notes: Observations are at the municipality-province-year level. “Year of change” is an indicator variable that equals one in the year of a SEZ program onset and zero otherwise. The variable
“≥6 years before” is an indicator variable that equals one if an observation is at least six years before the SEZ program starts. The variables “2 years before” to “5 years before” are indicator
variables that equals one if an observation is two to five years before the SEZ program starts. The variables “1 year later” to “5 years later” are indicator variables that equal one if an ob-
servation is one to five years after the SEZ program starts. The variable “≥6 years later” is an indicator variable that equals one in all other post-SEZ program years. The indicator variable
“1 year before” is left out so that the post-treatment effects are relative to the period immediately prior to the start of the program. In columns 1–3, the dependent variables are the natural
log of the measure of per capita Foreign Direct Investment related outcomes that are reported in the column heading. In column 4, the dependent variable is the natural log of per capita
domestic investment. In column 5, the dependent variable is the total factor productivity growth. In columns 6–8, the dependent variables are the natural log of the average worker's wage,
CPI and house prices. Robust standard errors are reported in parentheses, clustered by municipality. All regressions control for province-year fixed effects and municipality fixed effects.
⁎⁎⁎ Significant at the 1% level.
⁎⁎ Significant at the 5% level.
⁎ Significant at the 10% level.

the SEZ experiment on the level of the outcomes is identified by δ,
and the effect on the trend of the outcomes is identified by β.
Table 3 presents the results. To begin with, Column (1f) provides
the estimates of Eq. (2a), indicating that the SEZ program increases
the level of per capita FDI by 21.7% and the growth rate of FDI by
6.9 percentage points. Columns (2f)–(3f) confirm the contribution
of the SEZs in attracting vertical FDI, which takes advantage of
low-cost production in China for products to be exported and is
fueled mostly by China's Asian neighbors. 19 Column (4) investigates
the effect of SEZs on domestic capital formation. The estimate
shows that a SEZ program neither crowds in nor crowds out domestic
investment. Hence the SEZs increase a municipality's investment
overall. In aid of quantifying any agglomeration economies I apply
the growth accounting approach (Caselli, 2005; Young, 2003) to cal-
culate total factor productivity growth. A key step is to estimate
labor and capital shares. The most disaggregated GDP data Chinese of-
ficial statistics provide using the income approach is at the provincial
level. 20 I use the provincial capital share as a proxy for the municipal
capital share. By 2008 the average post-treatment period is
13.45 years. Column (5) therefore implies that the SEZ program
increases total productivity growth by 1.6 percentage points. To com-
pare this contribution with average TFP growth at municipality level,
2.6%, during the sample period, the treated municipalities experience
a 62% increase in TFP growth relative to the not-yet treated ones. For
comparison purposes I also use national capital share, θk = 0.4,

Fig. 4. The dynamic effect of SEZs on FDI. Notes: the horizontal axis measures the
number of years since the SEZ program took place. The plots connected by the solid
line indicate changes in ln(per capita FDI) compared to the period immediately before
the SEZ experiment conditional on municipality fixed effects and province-year fixed
effects. See Table 2 for the exact numbers of these point estimates. The dotted lines
indicate the 95% confidence intervals where standard errors are clustered at the
municipality level.
19
See Whalley and Xin (2006) and Ekholm et al. (2007). Horizontal FDI, which in-
volves the transfer of production (mainly from North America and Western Europe)
to service the Chinese internal market is not the main form of FDI in China especially
during the 1980s and 1990s when the SEZs were widely established (National Statisti-
cal Bureau, 2009 and other various issues).
20
See Hsueh and Li (1999) for 1978–95, and the National Statistical Bureau (2007)
for 1993–2004.
 J. Wang / Journal of Development Economics 101 (2013) 133–147 141

Table 3
The effects of the SEZs on local economy.

Dependent variable Capital investment TFP growth Factor prices

FDI Exports FIEs output Domestic investment Wage Consumer price index House prices

(1f) (2f) (3f) (4) (5) (6) (7) (8)

SEZ 0.217⁎⁎⁎ 0.402⁎⁎⁎ 0.159 −0.018 0.0014 −0.002 −0.003 0.038

(0.061) (0.087) (0.141) (0.044) (0.0027) (0.009) (0.005) (0.031)
PostSEZ trend 0.069⁎⁎⁎ 0.091⁎⁎⁎ 0.035⁎ 0.0011⁎⁎⁎ 0.006⁎⁎⁎ 0.004⁎⁎⁎
(0.010) (0.015) (0.021) (0.0003) (0.002) (0.001)
R-sq 9947 9946 3929 9933 9614 9933 9944 5170
Obs. 0.88 0.89 0.93 0.93 0.26 0.99 0.99 0.85

Notes: All observations are at the municipality-province-year level. SEZ is an indicator variable that equals one if an observation is after the SEZ program starts and zero otherwise.
PostSEZ trend is an indicator variable that denotes a linear trend after the SEZ program onset. In columns 1f–3f, the dependent variables are the natural log of the measure of per
capita Foreign Direct Investment related outcomes that are reported in the column heading. In column 4, the dependent variable is the natural log of per capita domestic invest-
ment. In column 5, the dependent variable is the total factor productivity growth. In columns 6–8, the dependent variables are the natural log of the average worker's wage, CPI and
house prices. Robust standard errors are reported in parentheses, clustered by municipality. All regressions control for province-year fixed effects and municipality fixed effects.
⁎⁎⁎ Significant at the 1% level.
⁎⁎ Significant at the 5% level.
⁎ Significant at the 10% level.

reported by Young (2003) and the international benchmark of Caselli specifications:

(2005), θk = 1/3, as a proxy for municipal capital share. Both results
are very similar. Since the estimates are not sensitive to using the pro-  
vincial or national average share, this mitigates any concern that Y ipt ¼ α i þ γpt þ Sipt δ þ F ipt β þ Oipt ϕ þ Oipt  Nipt θ þ εipt ; ð3aÞ
using an upper level capital share would cause a large measurement
error.
 
The findings on capital formation and technological progress are
Y ipt ¼ α i þ γpt þ Sipt δ þ Oipt ϕ þ Oipt  N ipt θ þ εipt : ð3bÞ
consistent with the results of previous research showing that firms
are more productive when they cluster (Greenstone et al., 2010).
The agglomeration benefits justify policies encouraging new
where Oit is equal to one if there is any other SEZ in the same province
business investment in a targeted area (Greenstone and Looney,
or nearby provinces that borders municipality i, and zero otherwise.
2010).
Nit denotes the number of neighboring municipalities with SEZs. All
Columns (6)–(8) examine the impact of the SEZ program on factor
other controls are as previously defined. We would expect positive
prices. Column (6) shows that, driven by the capital inflow, the
coefficients δ and β of the municipal SEZ indicators to capture any
municipalities in which a SEZ program took place experienced a 0.6
creation effect and a negative coefficient ϕ of the dummy variable
percentage point increase in the growth rate of the average worker
indicating nearby SEZs for any diversion effect. The parameter of the
wage compared to those in cities without a SEZ. The income increase
interaction term (Oit × Nit) denotes the effect of having more
is likely to drive up consumption in the municipality, and indeed the
neighboring municipalities with SEZs on the municipal economic
growth rate of living costs measured by the CPI rose by 0.4 percentage
outcomes. 22
points, as suggested by Column (7). Since the average post-treatment
Table 4 reports the estimated coefficients of Eqs. (3a) and (3b).
period is 13.45 years the estimates imply an 8% increase in the aver-
Columns (1f)–(3f) present the results on FDI related outcomes. The
age wage and a 5% increase in the CPI. Column (8) shows that house
coefficients of the SEZ indicators are all positive and the magnitudes
prices do not exhibit any increase because of a SEZ program. The find-
are similar to those in Table 3. These results confirm that there is a
ings are consistent with the fact that non-trivial barriers prevent
significant creation of a SEZ program on municipal FDI outcomes.
workers from moving from one municipality to another in China
The coefficients of the other nearby SEZ indicator are negative and
(Au and Henderson, 2006a,b) 21 and that land markets in China are
significant, suggesting that there is indeed a sizable diversion. As
poorly developed (Deng et al., 2008).
more geographically contiguous cites create SEZs, a city's FDI
decreases more. 23
To further assess the magnitude of the estimates, by 2008 the av-
4.3. Creation versus diversion
erage post-treatment period is 13.45 years and the average number
of neighboring SEZs is 4.82. The average creation for FDI is therefore
In terms of aggregate efficiencies, there are concerns that the for-
112% while the average diversion is 33%. The creation effect is larger
eign direct investment SEZs attract may not be created, but rather
than that of the diversion effect, which provides us with the relative
diverted. When a SEZ program is in place, foreign investors might
importance of those two effects.
simply change their location decision from a neighboring non-SEZ
There are suggestive evidence that more neighboring SEZs
municipality or province to the municipality with the SEZ. If this
increase the municipal own TFP growth. Finally, consistent with the
were the case, SEZs would merely redistribute FDI within China. The
pattern on FDI related outcomes, the increasing number of adjacent
two mechanisms present different welfare consequences and thus
SEZs reduces the municipal factor prices despite a lack of statistical
policy implications (Kline and Moretti, 2011). As a result, I separately
power.
identify the creation and diversion effects by regressing each munic-
ipal economic outcome on its own SEZ program dummy variable
and the indicators of other adjacent SEZs. I use the following 22
Nit is not included in the regression because Oit × Nit = Nit.
23
It is natural to examine whether the diversion or agglomeration effect differs
depending on the city's SEZ status. The ideal test is to use the interaction of the SEZ in-
dicator and the number of neighboring SEZs. However, a large number of cities in the
21
China restricts internal migration of its population between urban and rural areas, sample were granted with SEZs at the same wave. The resulting correlation between
between big and small cities, and between regions. The prime instrument of control is the interaction term and the main effect, 0.84, prevents me from using the fore-
the household registration (hukou system). mentioned specification.
142 J. Wang / Journal of Development Economics 101 (2013) 133–147

Table 4
The effects of the SEZs on local economy: diversion versus creation.

Dependent variable Capital investment TFP growth Prices

FDI Exports FIEs output Domestic investment Wage Consumer price index

(1f) (2f) (3f) (4) (5) (6) (7)

SEZ 0.228⁎⁎⁎ 0.419⁎⁎⁎ 0.190 −0.024 0.0012 −0.0016 −0.0027

(0.060) (0.086) (0.144) (0.043) (0.0027) (0.0093) (0.0052)
PostSEZ trend 0.066⁎⁎⁎ 0.086⁎⁎⁎ 0.033 0.0011⁎⁎⁎ 0.0060⁎⁎⁎ 0.0035⁎⁎⁎
(0.010) (0.015) (0.020) (0.0003) (0.0016) (0.0011)
OtherSEZ −0.048 −0.105 −0.331⁎ −0.135⁎⁎⁎ 0.0019 −0.0017 0.0049
(0.070) (0.093) (0.177) (0.048) (0.0045) (0.0109) (0.0074)
OtherSEZ × Number of neighboring SEZs −0.058⁎⁎ −0.086⁎⁎ −0.036 0.045⁎⁎⁎ 0.0009 −0.0003 −0.0019
(0.025) (0.034) (0.045) (0.015) (0.0009) (0.0039) (0.0029)
Obs. 9947 9946 3929 9933 9614 9933 9944
R-sq 0.88 0.89 0.93 0.93 0.26 0.99 0.99

Notes: All observations are at the municipality-province-year level. SEZ is an indicator variable that equals one if an observation is after the SEZ program starts and zero otherwise.
PostSEZ trend is an indicator variable that denotes a linear trend after the SEZ program onset. The dummy variable “OtherSEZ” equals one if the municipality is adjacent to any other
SEZs and zero otherwise. In columns 1f–3f, the dependent variables are the natural log of the measure of per capita Foreign Direct Investment related outcomes that are reported in
the column heading. In column 4, the dependent variable is the natural log of per capita domestic investment. In column 5, the dependent variable is the total factor productivity
growth. In columns 6–7, the dependent variables are the natural log of the average worker's wage and CPI. Robust standard errors are reported in parentheses, clustered by
municipality. All regressions control for province-year fixed effects and municipality fixed effects.
⁎⁎⁎ Significant at the 1% level.
⁎⁎ Significant at the 5% level.
⁎ Significant at the 10% level.

5. Heterogeneous SEZ effects
The SEZs can affect the local economy differently at different
stages of the program expansion, or the impacts may be heteroge-
neous across different treatment intensities. To shed light on that, I
now exploit the full richness of the SEZ program to explore these
two forms of heterogeneous effect that have not been previously
studied.
5.1. Early zones versus late zones
To see whether earlier SEZs exhibit different impact patterns than
those founded later, I estimate the following OLS specifications for the
municipal outcomes Yipt:
X
4
q
X
4
q To further address the endogeneity of the SEZ granting sequence, I
Y ipt ¼ α i þ γpt þ δ1 Sipt þ δq Gi Sipt þ β1 F ipt þ βq Gi F ipt
 q¼2 q¼2
þ Oipt ϕ þ Oipt  Nipt θ þ εipt ;
X
4   group during the same period in the absence of a SEZ program.
q
Y ipt ¼ α i þ γpt þ δ1 Sipt þ δq Gi Sipt þ Oipt ϕ þ Oipt  Nipt θ þ εipt :
q¼2
Where Gi2 is equal to one if a municipality was granted a SEZ
between 1986 and 1990 (group 2), and zero otherwise; Gi3 is equal
to one if a municipality was granted a SEZ between 1991 and 1995
(group 3), and zero otherwise. Gi4 is equal to one if a municipality
had not been granted with the SEZs by 1995 (group 4), and zero
otherwise. The municipalities that carried out the SEZ program
between 1978 and 1985 (group 1) are the reference group in the
specification. All other controls are as previously defined.
Table 5 presents the results comparing changes on local economic
outcomes in group 2, which carried out the SEZ program in the sec-
ond wave, group 3, which introduces the SEZ program in the third
round, and the final group relative to group 1, the earliest SEZ group.
Column 1 shows that the creation effect for the early zones is
much larger than for the later zones. Moreover, diversion becomes in-
creasingly important for the later zones. To take the 1978–1985 SEZs
for example, by 2008 a SEZ program on average increases per capita
FDI by 254%, while neighboring SEZs decrease municipal FDI by 21%.
In contrast, the creation effect for the 1996–2008 SEZs is 39% while
the diversion effect of surrounding SEZs is 28%. FDI seems to be sen-
sitive to the relative advantages enjoyed by the municipalities.
These results imply that when China becomes more liberalized, the
later SEZs tend to be closer substitutes for each other and therefore
generate more distortions in FDI location choice.
In terms of the domestic investment, the OLS estimates in column
2 indicate that a crowding-in effect disappears for later SEZs. There
are positive, however, no significant difference between the early
groups and late groups in terms of agglomeration economies. Finally,
as reported in columns 4 and 5, in line with the evidence on invest-
ment, the early zones experience the largest increases in local factor
price changes among the four groups. As the SEZ program gradually
expands to more municipalities, its effect on local prices becomes
less bigger.
implement a difference-in-differences matching estimator.24 I establish
that matched earlier treated and later treated municipalities are similar
ð4aÞ in terms of observables, so the performance of the latter can serve as a
counterfactual for what would have been the performance of earlier
Extending the standard DID matching in microeconomic applica-
tions (Abadie, 2005; Heckman et al., 1997; Blundell et al., 2004),
ð4bÞ Persson and Tabellini (2008) develop an approach that copes with
complications such as different treatment dates for different
observations in the treatment group. Their method is highly relevant
for my research setting as the municipalities were authorized to
establish SEZs in different years.
Within their framework, I implement the estimation in four steps.
First, I define a group of treated and a group of control municipalities
and estimate the probability of treatment. In particular, I estimate a
propensity score model over the period 1978–1990 on the sample
treated between 1978 and 1985 using as controls the group with
SEZs created between 1986 and 1990. The same technique is applied
in contrasting the municipalities that were treated between 1986 and
1990 with those treated between 1991 and 1995 used as control
areas over the period between 1978 and 1995, and then the munici-
palities that were treated between 1991 and 1995 with the last
group as controls over the period from 1978 to 2008. Table 1 provides
24
See Smith and Todd (2005), Imbens and Wooldridge (2009), Caliendo and
Kopeinig (2008).
 J. Wang / Journal of Development Economics 101 (2013) 133–147 143

Table 5
The effects of the SEZs on local economy: by the SEZ timing.

Dependent variable Capital investment TFP growth Prices

FDI Domestic investment Wage Consumer price index

(1) (2) (3) (4) (5)

SEZ 1.033⁎⁎⁎ 0.177 0.0130 0.0379 0.0042

(0.192) (0.129) (0.0095) (0.0267) (0.0147)
SEZ × group 2 [1986–1990] −0.654⁎⁎⁎ −0.184 −0.0083 −0.0255 0.0080
(0.197) (0.072) (0.0112) (0.0272) (0.015)
SEZ × group 3 [1991–1995] −0.849⁎⁎⁎ −0.267⁎⁎ −0.0128 −0.0365 −0.0014
(0.195) (0.121) (0.0103) (0.0288) (0.0163)
SEZ × group 4 [1996–2008] −0.678⁎⁎⁎ −0.110 −0.0157 −0.0297 0.0069
(0.231) (0.144) (0.0110) (0.0329) (0.0185)
PostSEZ trend 0.062⁎⁎⁎ 0.0008⁎ 0.0078⁎⁎⁎ 0.0060⁎⁎⁎
(0.014) (0.0004) (0.0024) (0.0015)
PostSEZtrend × group 2 [1986–1990] 0.002 −0.0001 −0.0041⁎⁎ −0.0023⁎⁎
(0.011) (0.0005) (0.0020) (0.0010)
PostSEZtrend × group 3 [1991–1995] −0.012 0.0002 −0.0028 −0.0042⁎⁎⁎
(0.013) (0.0005) (0.0023) (0.0012)
PostSEZtrend × group 4 [1996–2008] −0.057⁎⁎⁎ 0.0006 −0.0060⁎ −0.0051⁎⁎
(0.021) (0.0007) (0.0031) (0.0020)
OtherSEZ −0.011 −0.122⁎⁎ 0.0024 −0.0012 0.0060
(0.021) (0.050) (0.0045) (0.0106) (0.0074)
OtherSEZ × Number of neighboring SEZs −0.051⁎⁎ 0.045⁎⁎⁎ 0.0008 0.0008 −0.0005
(0.024) (0.016) (0.0010) (0.0039) (0.0029)
Obs. 9947 9933 9614 9933 9944
R -sq 0.88 0.93 0.26 0.99 0.99

Notes: All observations are at the municipality-province-year level. SEZ is an indicator variable that equals one if an observation is after the SEZ program starts and zero otherwise.
PostSEZ trend is an indicator variable that denotes a linear trend after the SEZ program onset. Group 2 [1986–1990] is an indicator variable that equals one if a municipality created
the SEZs between 1986 and 1990 and zero otherwise. Group 3 [1991–1995] is an indicator variable that equals one if a municipality created the SEZs between 1991 and 1995 and
zero otherwise. Group 4 [1996–2008] is an indicator variable that equals one if a municipality had not created a SEZ by 1995 and zero otherwise. The dummy variable “OtherSEZ”
equals one if the municipality is adjacent to any other SEZs and zero otherwise. In column 1, the dependent variable is the natural log of the measure of per capita Foreign Direct
Investment. In column 2, the dependent variable is the natural log of per capita domestic investment. In column 3, the dependent variable is the total factor productivity growth. In
columns 4–5, the dependent variables are the natural log of the average worker's wage and CPI. Robust standard errors are reported in parentheses, clustered by municipality. All
regressions control for province-year fixed effects and municipality fixed effects.
⁎⁎⁎ Significant at the 1% level.
⁎⁎ Significant at the 5% level.
⁎ Significant at the 10% level.

evidence of the validity of using the groups in this way. The munici- a k-nearest neighbor approach, a treatment case is matched with k
palities of adjacent rounds are more similar in terms of the observed control cases in an interval.
characteristics which may have played important roles in the selec- Second, for each treated municipality i I compute the difference in
tion process. They might also, therefore, be more similar in terms of the mean outcomes measured before and after the treatment date (si)
unobserved characteristics. and prior to the start of its matched control j's SEZ program (sj):
The variables used in estimating the propensity score include the
pre-treatment per capita industrial output, per capita number of sec-
ondary school students, the distance to the coast, highway density, j 1 1
dY i ¼ ∑ Y − ∑Y : ð6Þ
airport, port, per capita post and telecommunications, per capita T aij si ≤tbsj ipt T bi tbsi ipt
loans by and deposits in the financial institutions, wages, land price
levels and more importantly historical trends in the outcomes
examined to compare municipalities with similar outcomes trends Where Yipt is the municipal outcome in period t, the subscript i de-
prior to being designated a SEZ. For example, for the 1978–1990 SEZ notes the treated municipality and the superscript j refers to a certain
sub-sample, I use the characteristics in 1978 and pre-trends of prima- control municipality j. Tija is the number of years after the treatment
ry outcomes to estimate a logit model. These variables, denoted as X, date and prior to the start of j's SEZ program; Tib is the number of
are likely to affect the propensity for a municipality being selected to years before the treatment date.
experiment with the SEZ earlier and are also likely to be instrumental For each of the not-yet-treated controls j, I compute the difference
in affecting the outcomes. I create a D = 1 if the municipality had in outcomes over the periods before and after the SEZ program onset
implemented a SEZ program by 1985 and D = 0 if the municipality date in the treated municipality i it is matched with and prior to the
carried out the SEZ experiment between 1986 and 1990. start of j's own SEZ program:
 
′
Pr fD ¼ 1jX g ¼ ϕ X β : ð5Þ i 1 1
dY j ¼ ∑ Y − ∑Y : ð7Þ
T aij si ≤tbsj jpt T bi tbsi jpt

For the 1986–1995 SEZ sub-sample, I use the characteristics in

1985 and the pre-treatment trends of the outcomes to estimate a
logit model. The binary treatment variable is defined to be one if
the municipality was granted a SEZ between 1986 and 1990, and
zero if it was granted a SEZ between 1991 and 1995. A similar ap-
proach is implemented with the 1991–2008 SEZ sub-sample. Using
The resulting variable is dYji where the j subscript refers to a
certain municipality j among the controls and the superscript i refers
to the treated municipality it is matched with. This design ensures
that the periods over which the differences for the treated and the
control municipalities are computed coincide exactly.
144 J. Wang / Journal of Development Economics 101 (2013) 133–147

Third, for each treated municipality, I compute the weighted Table 6

average (wij as the weight for each matched control) of the difference DID matching results of the SEZs impacts: by the SEZ timing.

in difference estimator 25: Sub-sample 1978–1990 1986–1995 1991–2008

SEZs SEZs SEZs

j i Treated municipalities Group 1 Group 2 Group 3

α i ¼ ∑ wij dY i −∑ wij dY j : ð8Þ [1978–1985] [1986–1990] [1991–1995]
j j
Control municipalities Group 2 Group 3 Group 4
Finally, I compute the average estimated effect of the SEZ program [1986–1990] [1991–1995] [1996–2008]
on the local economy in the treated municipalities as a simple average PSM matching estimates (1) (2) (3)
of individual estimates:
Investments
Log (per capita FDI) 0.60 0.33 0.20
(0.18)⁎⁎⁎ (0.13)⁎⁎ (0.07)⁎⁎⁎
1
^ ¼ ∑ αi ;
α ð9Þ (0.25)⁎⁎ (0.18)⁎ (0.28)
I i Log(per capita domestic 0.35 0.20 −0.03
investment)
where I is the number of treated municipalities. (0.15)⁎⁎ (0.08)⁎⁎ (0.04)
(0.28) (0.12) (0.21)
I note that such k-nearest neighboring matching with replacement
is likely to use some control municipality multiple times. The same Agglomeration economies
controls might thus be matched with several treated municipalities TFP growth −0.019 0.010 −0.003
and possibly at very different SEZ onset dates. I further follow (0.013) (0.008) (0.003)
(0.021) (0.014) (0.010)
Persson and Tabellini (2008) to adjust the computation of the stan-
dard error of the estimators taking into account possible correlation Prices
(Please refer to Appendix A). The upper bound of the standard error Log (average worker wage) 0.077 0.034 0.012
is estimated by assuming independent observations while the lower (0.027)⁎⁎ (0.013)⁎⁎ (0.016)
bound is computed by assuming perfect correlations among repeated (0.041)⁎ (0.034) (0.095)
Log (consumer price index) −0.022 0.028 0.004
observations in control municipalities. Since the matching section of (0.013) (0.014)⁎⁎ (0.010)
my paper is an empirical application of their approach, I refer the (0.024) (0.026) (0.041)
reader to their paper for technical details. After matching: N. treated 15 33 127
As shown in Appendix A, there is a large overlap in p-scores be- municipalities
N. control municipalities 14 90 94
tween treated and not-yet-treated municipalities, which ensures the
Before matching: N. treated 28 49 143
feasibility of matching. Taking the 1978–1990 SEZ sub-sample as an municipalities
example, there are 15 municipalities of the original 28 in the treat- N. control municipalities 49 143 101
ment group on support, and 28 municipalities in the matched control Sample period 1978–1990 1978–1995 1978–2008
group compared to the original sample of 49 municipalities. For the Notes: Difference-in-differences are the difference in changes between treatment and
1986–1995 SEZ sub-sample, 33 of the 49 treated municipalities and control municipalities.
90 of the 143 municipalities in the control group are matched. For First parenthesis reports standard errors estimated assuming independent observations.
Second parenthesis reports standard errors estimated assuming perfect correlations of
the 1991–2008 SEZ sub-sample, only 16 of the 143 treated municipal-
repeated observations in control municipalities.
ities and 7 of the 101 control municipalities are left unmatched. The For three sub-samples, municipalities of adjacent SEZ rounds are chosen as the treat-
matching procedure is able to balance the distribution of the relevant ment and control group that are reported in the column heading. For the 1978–1990
variables in both the control and treatment groups. T-test and the SEZ sub-sample, the treatment group is the municipalities that created SEZs between
pseudo-R2 (Sianesi, 2004) are both fairly low after matching, which 1978 and 1985 while the control group is the municipalities with SEZs between 1986
and 1990. The same technique is applied in contrasting the municipalities that were
suggests that potentially important selection criteria become not treated between 1986 and 1990 with those treated between 1991 and 1995 used as
significant after matching. This means that there are no systematic control areas over the period between 1978 and 1995, and then the municipalities
differences in the distribution of covariates between the control that were treated between 1991 and 1995 with the last group as controls over the pe-
group and the treatment group. riod from 1978 to 2008. The propensity score matching (k-nearest neighbor matching
with replacement) is performed over covariates including initial per capita industrial
Table 6 shows the estimated average treatment effects for the
output, per capita middle school students, the distance to the coast, highway density,
treated group. The first and second parentheses provide the lower airports, ports, per capita post and telecommunications, per capita deposits in financial
and upper bound estimate of the standard error respectively. The institutions, per capita loans by financial institutions, wages, land price levels and his-
point estimate of FDI for the 1978–1990 sub-sample is 60%, meaning torical trends in main outcomes. The DID results are estimated over the matched
that the average change of per capita FDI for the treatment group is sub-sample.
⁎⁎⁎ Significant at the 1% level.
60% higher than that for the control group. For the 1986–1995 ⁎⁎ Significant at the 5% level.
sub-sample the SEZ program increases per capita FDI by 33% relative ⁎ Significant at the 10% level.
to the matched control municipalities. For the 1991–2008 sub-sample
per capita FDI increases by 20% compared to the matched controls.
The effect of a SEZ treatment seems to have decreased marginally For the 1986–1995 sub-sample, the SEZ program increases the aver-
for later zones, which is consistent with the OLS estimates. Moreover, age wages by 3.4% and the CPI by 2.8%; for the 1991–2008
the SEZ program seems to have crowded in domestic investment for sub-sample (the largest sample), the average wage increases by
the first two sub-samples, but not for the 1991–2008 sub-sample. 1.2% relative to the matched control municipalities, while the CPI in-
There is suggestive evidence indicating a positive effect of the SEZs creases by 0.4%. To sum up, the DID matching estimates are consistent
on TFP growth, though the conclusions are not robust for all three overall with the OLS specifications.
sub-samples. Finally, the matching estimates on factor prices for the
1978–1990 sub-sample indicate that the average worker's wage in- 5.2. Multiple SEZs versus one SEZ
creases by 7.7% and the CPI decreases by 2.2% with a SEZ treatment.
The term treatment used in the paper actually entails treatments
of multiple intensities (state-level SEZs, provincial SEZs and open
25
For a k-nearest neighbor matching, 1/k is the weight for each matched control. economic areas) for different cities. I then assess whether the effect
 J. Wang / Journal of Development Economics 101 (2013) 133–147 145

Table 7
The impact of the SEZ intensity on local economy.

Dependent variable Capital investment TFP growth Prices

FDI Exports FIEs output Domestic investment Wage CPI

(1f) (2f) (3f) (4) (5) (6) (7)

SEZ 0.100 0.011 0.080 0.001 −0.0012 −0.014 0.001

(0.064) (0.084) (0.207) (0.042) (0.0034) (0.009) (0.006)
PostSEZ trend 0.018⁎ 0.051⁎⁎⁎ 0.014 0.0017⁎⁎⁎ 0.001 0.001
(0.010) (0.018) (0.024) (0.0004) (0.001) (0.001)
SEZ × Intensity being 2 0.219 0.426⁎⁎ −0.037 −0.060 0.0048 0.025 −0.000
(0.143) (0.179) (0.299) (0.087) (0.0058) (0.024) (0.013)
PostSEZ trend × Intensity being 2 0.034⁎⁎⁎ 0.010 0.036 −0.0006 0.002 0.002⁎
(0.012) (0.015) (0.022) (0.0004) (0.001) (0.001)
SEZ × Intensity being 3 0.842⁎⁎⁎ 1.483⁎⁎⁎ 0.767⁎⁎⁎ 0.000 0.0070 0.047⁎⁎ 0.025⁎⁎
(0.112) (0.151) (0.281) (0.059) (0.0059) (0.021) (0.012)
PostSEZ trend × Intensity being 3 0.043⁎⁎⁎ −0.002 0.001 −0.0010⁎⁎⁎ 0.005⁎⁎ 0.004⁎⁎⁎
(0.009) (0.014) (0.022) (0.0003) (0.001) (0.001)
Obs. 9141 9140 3852 9127 8840 9129 9138
R-sq 0.89 0.90 0.94 0.93 0.27 0.99 0.99

Notes: All observations are at the municipality-province-year level. 26 municipalities that received no SEZ treatment during the sample period are excluded from the regression. SEZ
is an indicator variable that equals one if an observation is after the SEZ program starts and zero otherwise. PostSEZ trend is an indicator variable that denotes a linear trend after the
SEZ program onset. I define the treatment intensity parameter as one when a municipality only has one treatment among state-level SEZs, provincial SEZs and open economic areas.
When a municipality has two of the programs, the treatment intensity is defined as two. When a municipality has carried out three programs, the treatment intensity is three. In-
tensity being 2 is an indicator variable that equals one if a municipality had been exposed to treatment intensity 2 by 2008 and zero otherwise. Intensity being 3 is an indicator
variable that equals one if a municipality had been exposed to treatment intensity 3 by 2008 and zero otherwise. In columns 1f–3f, the dependent variables are the natural log
of the measure of per capita Foreign Direct Investment related outcomes that are reported in the column heading. In column 4, the dependent variable is the natural log of per capita
domestic investment. In column 5, the dependent variable is the total factor productivity growth. In columns 6–7, the dependent variables are the natural log of the average
worker's wage and CPI. Robust standard errors are reported in parentheses, clustered by municipality. All regressions control for province-year fixed effects and municipality
fixed effects.
⁎⁎⁎ Significant at the 1% level.
⁎⁎ Significant at the 5% level.
⁎ Significant at the 10% level.

of the SEZ program is heterogeneous across different treatment
intensities using the following panel data specification:
X
3
q
X
3
Y ipt ¼ α i þ γpt þ δ1 Sipt þ δq Ii Sipt þ β1 F ipt þ
q¼2 q¼2
X
3
q (1f), δ 2 and β 2 imply a level effect size of 22% and a growth-rate
Y ipt ¼ α i þ γpt þ δ1 Sipt þ δq Ii Sipt þ εipt :
q¼2
Where Iiq is equal to one if the municipality had been exposed to
the treatment intensity q (q = 2, 3) by 2008, and zero otherwise 26;
all other controls are as previously defined. The 26 municipalities
that received no SEZ treatment during the sample period are
excluded for two reasons. First, a city which had not carried out the
SEZ program by 2008 might have been fundamentally different
from the treated municipalities. Second, the sample size of the
untreated group is too small. The group with a treatment intensity
of one is a more suitable reference group. All the other controls are
as previously defined. An important feature of this empirical setting
is that various municipalities could be exposed to very different
treatment intensities, making it feasible to estimate the δq and βq
coefficients using the intensity variation and time variation of
treatments across municipalities.
The result of Table 7 shows there to be a heterogeneous effect of
treatment intensities on local economic outcomes. More precisely,
there is a positive and significant treatment intensity effect when
moving from a single SEZ (treatment intensity being equal to one)
⋏
to the dual treatment (treatment intensity equals two), δ 2 > 0 and
26
Among the municipalities with more than one SEZ program there are rich varia-
tions in the timing of their first, second and third program. However, exploring the
time-varying jump intensity (from 1 to 3) suffers from the endogeneity problem,
which is difficult to control for. Therefore, I only explore the static differences in the
municipalities' treatment intensities in 2008.
⋏
β 2 > 0. There is also a positive and significant effect when moving
from the first to the third treatment intensity (treatment intensity
⋏ ⋏
equals three), so δ 3 > 0 and β 3 > 0. These confirm that the treat-
q
ment intensity effects increase in absolute magnitude when moving
βq Ii F ipt þ εipt ; ð10aÞ from the first to the second intensity, and from the first to the third
intensity.
Moving from the first to the second treatment intensity in Column
⋏ ⋏
ð10bÞ
effect size of 3.4 percentage points on per capita FDI. Moving from
⋏ ⋏
the first to the third treatment intensity, δ 3 and β 3 indicate a level
effect size of 84% and a growth-rate effect size of 4.3 percentage
points on per capita FDI.
There is no significant difference in domestic investment across
the intensities. The empirical results for other outcomes, including
TFP growth, wages and the CPI, leave unchanged the basic implication
that treatment intensity effects are heterogeneous, as shown in
Columns (5)–(7). The municipalities with multiple SEZs experience
larger increases in TFP growth rate, wages and the CPI.
6. Conclusion
Capital as well as advanced technology is typically desirable for de-
velopment. Aiming to attract foreign capital, boost exports and absorb
advanced technology, SEZs have been widely adopted as a place-based
program. This study contributes to the long-standing debate about
their effectiveness by providing some of the first estimates of the im-
pact of the SEZs on the local economy, as measured by investment,
TFP growth and factor prices.
China's SEZ policy package, including private property rights pro-
tection, tax breaks and land use policy, on average increases per capita
foreign direct investment mainly in the form of foreign-invested and
export-oriented industrial enterprises. The FDI inflow does not,
however, crowd out domestic investment. More importantly, the
majority of the FDI attracted by the SEZs is new activity rather than
simply a reallocation from other non-SEZ areas. Due to the
146 J. Wang / Journal of Development Economics 101 (2013) 133–147
agglomeration economies, the SEZs increase the total factor productiv-
ity growth that provides justifications for such spatially-targeted subsi-
dies. Finally, there is a significant increase in local workers' earnings
and a moderate rise in living costs without a significant increase in
house prices.
While the evidence shows that the SEZs on average benefit the
local economy with relatively small distortions, it appears that there
is heterogeneity in terms of the program effects. Later zones tend to
generate larger distortions in FDI location choice relative to early
ones. The increase in wages also is not as large as with earlier
zones. On the other hand, the municipalities that receive multiple
programs exhibit larger effects on local economic outcomes than
those with one SEZ.
The findings of this paper provide an important insight into how a
local economy gains from the place-based SEZ program. Further
research is required to identify separately the contributions of the
three components of the SEZ policy package (Devereux, 2007). More-
over, the positive effects demonstrated should be interpreted with
caution as they might be closely linked to Chinese institutions includ-
ing the decentralized implementation of the SEZ program by local
governments and China's unusual migration control which prevents
large population inflows from arbitraging the benefits. Whether the
results hold empirically in other countries awaits further work.
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://
dx.doi.org/10.1016/j.jdeveco.2012.10.009.
References
Abadie, A., 2005. Semiparametric Difference-in-Difference Estimators. Review of Eco-
nomic Studies 72, 1–19.
Abraham, F., Konings, J., Slootmaekers, V., 2010. FDI spillovers in the Chinese
manufacturing sector. The Economics of Transition 18 (1), 143–182.
Aggarwal, A., 2005. Performance of Export Processing Zones: A Comparative Analysis of
India, Sri Lanka, and Bangladesh. Working Paper No. 155. Indian Council for
Research on International Economic Relations, New Delhi.
Aggarwal, A., Hoppe, M., Walkenhorst, P., 2008. Special Economic Zones and economic
diversification: some evidence from South Asia. In: Newfarmer, Richard, Shaw,
William, Walkenhorst, Peter (Eds.), Breaking into New markets: Emerging Lessons
for Export Diversification. World Bank, Washington DC.
Alder, S., Shao, L., Zilibotti, F., 2012. The Effect of Economic Reform and Industrial Policy
in a Panel of Chinese Cities. Working paper. University of Zurich.
Au, C., Henderson, J.V., 2006a. Are Chinese cities too small? Review of Economic
Studies, vol. 73(3). Wiley Blackwell, pp. 549–576 (07).
Au, C., Henderson, J.V., 2006b. How migration restrictions limit agglomeration and
productivity in China. Journal of Development Economics 80 (2), 350–388.
Bank, World, 2008. Special Economic Zones: Performance, Lessons Learned, and
Implications for Zone Development.
Blundell, R., Costa Dias, M., Meghir, C., Van Reenen, J., 2004. Evaluating the Employ-
ment Impact of a Mandatory Job Search Assistance Program. Journal of the Europe-
an Economic Association 596–606.
Busso, M., Gregory, J., Patrick, K., forthcoming. Assessing the Incidence and Efficiency of
a Prominent Place Based Policy. American Economic Review.
Cai, D. (Ed.), 2008. Xi Zhongxun Governing Guangdong (Xi Zhongxun Zhuzheng
Guangdong). Chinese Communist Party History Press, Beijing.
Caijian, 2010. Ministry of Finance Circular No. 48, 2010.
Caliendo, M., Kopeinig, S., 2008. Some practical guidance for the implementation of
propensity score matching. Journal of Economic Surveys 22 (1), 31–72.
Caselli, F., 2005. Accounting for cross-country income differences. In: Aghion, P., Durlauf,
S. (Eds.), Handbook of Economic Growth, vol. 1. North-Holland Press, pp. 679–741.
Coe, D.T., Helpman, E., Hoffmaister, A.W., 2009. International R&D spillovers and
institutions. European Economic Review 53 (7), 723–741.
Combes, P., Duranton, G., Gobillon, L., 2011. The identification of agglomeration
economies. Journal of Economic Geography 11 (2), 253–266.
Combes, P., Duranton, G., Gobillon, L., Puga, G., Roux, S., forthcoming. The Productivity
Advantages of Large Markets: Distinguishing Agglomeration from Firm Selection.
Econometrica.
Démurger, S., Sachs, J.D., Woo, W.T., Bao, S.M., Chang, G., 2002. The relative contributions
of location and preferential policies in China's regional development: being in the
right place and having the right incentives. China Economic Review 13 (4), 444–465.
Deng, X., Huang, J., Rozelle, S., Uchida, E., 2008. Growth, population and industrializa-
tion, and urban land expansion of China. Journal of Urban Economics 63 (1), 96115.
Desai, M.A., Foley, C.F., Hines Jr., J.R., 2009. Domestic effects of the foreign activities of
US multinationals. American Economic Journal: Economic Policy 1 (1), 181–203.
Devereux, M.P., 2007. The Impact of Taxation on the Location of Capital, Firms and
Profit: A Survey of Empirical Evidence. Working Paper. Oxford University Centre
for Business Taxation.
Ekholm, K., Forslid, R., Markusen, J.R., 2007. Export-platform foreign direct investment.
Journal of the European Economic Association 5 (4), 776–795.
Fagaidiqu, 2010. National Development and Reform Commission Circular No. 827,
2010.
Feenstra, R.C., Wei, S.J., 2010. Introduction to “China's Growing Role in World Trade”,
published as NBER Chapters. China's Growing Role in World Trade, pp. 1–31.
Fernandes, A., Tang, H., 2012. Learning from Neighbors' Export Activities: Evidence
from Exporters' Survival. Working paper. Tufts University.
Ge, W., 1999. Special Economic Zones and the Economic Transition in China. World
Scientific, Singapore and River Edge, NJ.
Glaeser, E.L., Gottlieb, J.D., 2008. The Economics of Place-making Policies. NBER
working paper.
Glaeser, E.L., Rosenthal, S.S., Strange, W.C., 2010. Urban economics and entrepreneurship.
In: Glaeser, Edward, Rosenthal, Stuart, Strange, William (Eds.), Cities and
Entrepreneurship: Journal of Urban Economics, vol. 67(1), pp. 1–14.
Greenstone, M., Looney, A., 2010. An Economic Strategy to Renew American Communi-
ties. Hamilton Project, Brookings Institution.
Greenstone, M., Hornbeck, R., Moretti, E., 2010. Identifying agglomeration spillovers:
evidence from winners and losers of large plant openings. Journal of Political
Economy 118 (3), 536–598.
Guofa, 2005. State-Council Circular No. 15, 2005.
Guofa, 2010. State-Council Circular No. 9, 2010.
Hale, G., Long, C.X., 2007. Are There Productivity Spillovers from Foreign Direct
Investment in China. Federal Bank of San Francisco working paper No.
2006–13.
Heckman, J., Ichimura, H., Smith, J., Todd, P., 1997. Matching as an Econometric Evalu-
ation Estimator Evidence from a Job Training Program. Review of Economic Studies
64, 605–654.
Holz, C.A., 2008. China's 2004 economic census and 2006 benchmark revision of GDP
statistics: more questions than answers. China Quarterly 193, 150–163.
Hsueh, T.T., Li, Q. (Eds.), 1999. China's National Income: 1952–1995. Westview Press,
Boulder.
Huang, Y.S., 1998. FDI in China: An Asian Perspective. Chinese University Press, Hong
Kong.
Imbens, G.W., Wooldridge, J.M., 2009. Recent developments in the econometrics of
program evaluation. Journal of Economic Literature 47 (1), 5–86.
Jacobson, L.S., LaLonde, R.J., Sullivan, D.G., 1993. Earnings losses of displaced workers.
American Economic Review 83 (4), 685–709.
Kline, P., 2010. Place based policies, heterogeneity, and agglomeration. American
Economic Review: Papers and Proceedings 100 (2), 383–387.
Kline, P., Moretti, E., 2011. Local Economic Development, Agglomeration Economies
and the Big Push: 100 Years of Evidence from the Tennessee Valley Authority.
Working Paper.
Kung, K.S., 1985. The origins and performance of China's Special Economic Zone. Asian
Journal of Public Administration 7 (1), 198–215.
Litwack, J.M., Qian, Y.Y., 1998. Balanced or unbalanced development: Special Economic
Zones as catalysts for transition. Journal of Comparative Economics 26 (1),
117–141.
Liu, Z.Q., 2008. Foreign direct investment and technology spillovers: theory and
evidence. Journal of Development Economics 85 (1–2), 176–193.
McGrattan, E., Prescott, E.C., 2009. Openness, technology capital and development.
Journal of Economic Theory 144 (6), 2454–2476.
Moretti, E., forthcoming. Real Wage Inequality. American Economic Journal: Applied
Economics.
National Statistical Bureau, 2002. The classification, methods and definitions of NSB's sta-
tistics. http://www.stats.gov.cn/english/classificationsmethods/.
National Statistical Bureau, 2007. Data of Gross Domestic Product of China (1952–
2004). China Statistics Press, Beijing.
National Statistical Bureau, 2009 and other various issues. China Statistical Yearbook.
China Statistics Press, Beijing.
Park, J.D., 1997. The Special Economic Zones of China and Their Impact on its Economic
Development. Praeger, London.
Persson, T., Tabellini, G., 2008. The growth effects of democracy: is it heterogenous and
how can it be estimated? In: Helpman, E. (Ed.), Institutions and Economic
Performance. Harvard University Press.
Peters, A.H., Fisher, P.S., 2002. State Enterprise Zone Programs: Have They Worked?
Upjohn Institute Press.
Prasad, E., Wei, S.J., 2007. The Chinese approach to capital inflows: patterns and
possible explanations. In: Prasad, Eswar, Wei, Shang-Jin (Eds.), Capital Controls
and Capital Flows in Emerging Economies: Policies, Practices and Consequences,
pp. 421–480.
Rolfe, R.J., Woodward, D.P., Kagira, B., 2004. Footloose and tax free: incentive
preferences in Kenyan export processing zones. South African Journal of
Economics 72 (4), 784–807.
Special Economic Zones in South Asia: Comparative Analysis of Bangladesh, Sri Lanka,
and India, mimeo, Harvard University.
Shangzifa, 2006. Ministry of Commerce Circular No. 257, 2006.
Shangzifa, 2008. Ministry of Commerce Circular No. 21, 50, 51, 64, 2008.
Shangzifa, 2009. Ministry of Commerce Circular No. 2, 6, 7, 8, 9, 2009.
Sianesi, B., 2004. An evaluation of the active labour market programmes in Sweden.
The Review of Economics and Statistics 86 (1), 133–155.
Smith, J.A., Todd, P.E., 2005. Does matching overcome LaLonde's critique of
nonexperimental estimators? Journal of Econometrics 125 (1–2), 305–353.
 J. Wang / Journal of Development Economics 101 (2013) 133–147 147

Wei, S.J., 1995. The open door policy and China's rapid growth: evidence from
city-level data. In: Ito, Takatoshi, Krueger, Anne O. (Eds.), Growth Theories in
Light of the East Asian Experience. University of Chicago Press, Chicago.
Whalley, J., Xin, X., 2006. China's FDI and Non-FDI economies and the sustainability of
future high Chinese growth. China Economic Review 21 (1), 123–135.
Willmore, L., 1996. Export processing in the Caribbean: lessons from four case studies.
Working Paper No. 42. United Nations Economic Commission for Latin America
and the Caribbean, Santiago.
Xu, C.G., 2011. The fundamental institutions of China's reforms and development.
Journal of Economic Literature 49 (4), 1076–1151.
Young, A., 2003. Gold into base metals: productivity growth in the People's Republic
of China during the Reform Period. Journal of Political Economy 111 (1),
1220–1261.
Zeng, Z., 2011. How Do Special Economic Zones and Industrial Clusters Drive China's
Rapid Development? World Bank policy research working paper 5583.
Zhongfa, 1979. The Central Government Circular No. 50, 1979.