To Build Outward or Upward?

The Spatial Pattern of Urban Land Development in China

Zhi Wang, Qinghua Zhang, Li-An Zhou

December 30, 2016

Abstract

This paper studies the determinants of the spatial pattern of urban land development.
Our study is motivated by an interesting fact observed in the past decade, the fact that
the fastest-growing Chinese cities have undergone outward expansion of urban land
development with relatively low use intensity. We link this pattern with the career
concerns of local leaders who act as city developers. We first develop a theoretical
model to demonstrate how a city leader whose career advancement depends on city
economic growth chooses the spatial pattern of urban land development. Two major
testable predications are derived. First, city leaders with high career incentives are more
likely to expand the city outward in order to boost land sale revenues for financing
public infrastructure and enhancing economic growth. Second, the average use intensity
of newly developed land is lower for city leaders with stronger career incentives due to
the trade-off between a city’s upward and outward expansion. We test the theory using
a large dataset of residential land transactions matched with city leaders from 2000 to
2012. We find robust evidence consistent with the predictions of our model.

JEL Classification: R1; R52; H7

Keywords: Urban land development; Spatial patterns; Land use regulations; City
developer; Political economy

 We would like to thank Vernon Henderson for substantive comments and suggestions on this research
project. We thank all seminar participants at the 2016 Meetings of the Urban Economics Association, the
China Meeting of Econometric Society 2016, Singapore National University, Jinan University, Xiamen
University and Fudan University for helpful comments and suggestions. Zhang gratefully acknowledges
support from the Chinese National Science Foundation (Grant#71673010). Zhang gratefully
acknowledges the support of a Global Research Program on Spatial Development of Cities, funded by
the Multi Donor Trust Fund on Sustainable Urbanization of the World Bank and supported by the UK
Department for International Development. Any remaining errors are our own.

School of Economics, Fudan University, No. 600 Guoquan Road, Shanghai, 200433, China. Email
address: [email protected], Phone: +86 21 6564 3021.

Guanghua School of Management, Peking University, Beijing, China, 100871, Email address:
[email protected], Phone: +86 10 6275 7960.

Guanghua School of Management and IEPR, Peking University, Beijing, China, 100871. Email address:
[email protected].
1

Electronic copy available at: https://ssrn.com/abstract=2891975

1. Introduction
China has experienced a massive process of urbanization in the past two decades
on an unprecedented scale in human history. According to the National Bureau of
Statistics of China, the nation’s urbanization rate increased from 26% to 49.9% from
1990 through 2010. In this process 372 million people, the vast majority of whom were
originally from rural villages, moved into the urban population. Built-up urban land
area also expanded substantially, increasing at an annual rate of 5.73%, or 134,550
hectares per year from 1990 through 2010. This astonishing scale of urbanization has
been accompanied by tremendous growth of real GDP in cities.
In these two decades of urbanization, China has undergone a very interesting land-
development pattern. Figure 1 depicts the correlation between outward expansion of
land development and population growth for China’s top 50 cities from 1990-2010. It
is clear that faster population growth is associated with increased outward expansion of
land use. For the same set of cities, Figure 2 shows that the cities with higher population
growth have lower floor-to-area ratios (FARs) imposed on newly developed residential
land. Contrary to the common impression that heavily populated cities in developed
regions such as New York, London, Singapore, Tokyo, and Hong Kong tend to build
more intensively, China’s intensely populated cities, such as Beijing, Shanghai, and
Guangzhou, have much lower FARs than some second-tier or even third-tier cities like
Xi’an and Guiyang (provincial capitals of Shaanxi and Guizhou, respectively, in the
western region of China).
This paper attempts to explain this puzzling pattern by investigating the decision-
making of urban land development policies. Chinese city leaders play a central role in
urban land development. They act as city developers and set the key parameters shaping
decisions about how much new land is to be developed and where, and the regulation
of the intensity of land use such as the building density (i.e., FAR) for each land parcel
(Lichtenberg and Ding, 2008, 2009; Cai, Wang and Zhang, 2016; Brueckner, Fu, Gu
and Zhang, 2016). This is in stark contrast to many developed countries where
homeowners have considerable influence on local land development. In the meantime,
Chinese city leaders, who hold critical control over local economic activities, are placed
in promotion tournaments for economic growth (Chen, Li, Zhou, 2005; Li and Zhou,
2005; Xu, 2011; Wu, Deng, Huang, Morck, Yeung, 2013; Yao and Zhang, 2015; Gong,
Knight, Xiao, Zhang, 2016) in which their promotions are closely linked with regional
economic outcomes (such as growth of GDP and fiscal revenues).
2

Electronic copy available at: https://ssrn.com/abstract=2891975

 In this paper, we link the peculiar pattern of urban land development we discussed
earlier with the career concerns of city leaders. There are very few studies in the
literature that examine the determinants of land development patterns in developing
countries (Gyourko and Molloy, 2015). Most of the relevant studies are set in the
context of the developed world (e.g., Burchfield, Overman, Puga, Turner, 2006). Our
paper fills this gap. More importantly, the current literature emphasizes the idea that
incumbent home-owners have incentive to influence the local planners to restrict land
development in their neighborhoods in order to keep their property values from falling
(Fischel, 2001; Glaeser, Gyourko and Saks, 2005; Hilber and Robert-Nicoud, 2013;
Ortalo-Magne and Prat, 2015). The influence of developers neither shows up in their
framework nor constitutes the focus of analysis in this line of literature, although land
developers may have also played a salient role in determining land development (e.g.,
Molotch, 1976; Logan and Molotch, 1987; Sole-Olle and Viladecans-Marsal, 2012).1
We first develop a theoretical model to demonstrate how a city leader whose career
advancement depends on city economic growth chooses the spatial pattern of urban
land development. Specifically, she faces a choice regarding urban boundary (i.e., how
far the outward expansion of urban land development goes) and land use intensity
within the city reflected by a certain regulatory FAR schedule. Her payoffs depend on
the likelihood of her promotion, which hinges on both the ex ante career-concern
intensity determined by her initial age and political hierarchy level when she first took
office, and the ex post economic performance during her term. If she has a higher ex
ante career-concern intensity, she will have a stronger incentive to promote the city’s
economic performance since doing so can more effectively enhance the likelihood of
her promotion and her expected payoffs.
We assume that the city government collects revenues from the sales of the long-
term leaseholds of land use rights to private developers, which is consistent with
China’s current land property institutions. The land sale revenues are supposed to
finance urban infrastructure construction, as is commonly done in China, and in turn

1
There are some studies that argue for the importance of local governments and business developers in
urban development. Molotch (1976) proposes the idea of a "growth engine" and highlights how the close
connections between local government officials and local business leaders help to create land policies
that are in favor of local economic growth. Logan and Molotch (1987) argue that “urban growth
machines”, which combine real estate, banking, and commercial interests to support the expansion of
cities, are a typical feature of many U.S. cities. However, these arguments do not provide a formal theory
or empirical analysis of the role of developers in the formation of land development policies. Sole-Olle
and Viladecans-Marsal (2012) find some indirect empirical evidence pointing to the influence of
developers on local land development regulations in Spain.
3
are expected to boost local economic growth. Developing more land at the city edge
(i.e., outwardly expanding the city) can create more land sale revenues for the city
treasury because the development costs near the city edge are much lower than those
near the city center. However, land expansion is subject to the quota restrictions
imposed by the upper-level government. As a result, the city leader has to exert greater
effort to bargain with the upper-level government in order to obtain more quotas for
rural-to-urban land conversion if she wants to outwardly expand the city (Xie, 2015).
Such efforts may be rather costly to the city leader. Only those city leaders with high
career-concern intensity are willing to exert more effort to seek more quotas from the
upper-level government in exchange for the benefits of increasing land sale revenues
and boosting local GDP. Therefore, we would expect that city leaders with stronger
career concerns will be more likely to expand the city outward. This is the first
prediction of our model.
The city leader also sets the density constraints of land use (i.e., regulatory FAR
schedule) in the city. In doing so, she takes into account of the value of land considering
various benefits and costs including the possible externality of high density
development. Our theory shows that, if we hold housing demand factors such as income
and population constant, the FAR schedule is lower as the city's boundary expands
outward. This reflects a trade-off between the city's outward and upward expansion.
Combining the first prediction of the model that city leaders with higher career-concern
intensity tend to expand the city more outwardly, we derive the second prediction of the
model: city leaders with higher career-concern intensity tend to set lower regulatory
FARs.
Our empirical analysis utilizes the database on residential land transactions
provided by the China Index Academy. The sample for our analysis covers over 30,000
completed residential land transactions (including residential mixed with commercial
land) in 202 Chinese cities during 2000-2012 and contains detailed information on
prices, sale date, total area, FAR upper limits and location on each transacted parcel of
residential land.
The outward dimension of urban land development is usually measured in the
current literature by the change in total urban land area (e.g., Brueckner and Fansler,
1983; McGrath, 2005; Deng, Huang, Rozelle, Uchida, 2008; Lichtenberg and Ding,
2009). A serious limitation of the above measure lies in its inability to distinguish
between urban land developments that fill in open space between existing developments
4
within the city boundary and those at the city edge that push the urban boundary further
away from the city core. The latter case is precisely the outward expansion of the urban
boundary that we examine in this paper. Taking advantage of our micro-level land data,
which contains information on the geographic coordinates for each land parcel, we
calculate the distance from each land parcel to the city center and develop a measure of
outward development based on the top percentiles in the distribution of the distances to
the city center of all the land parcels.
We match the land transaction data with the personal data on city leaders taking
office in about two hundred cities matched with our land sample during the same time
period. Based on each leader's initial age and political hierarchy level at the start of her
or his office term, we estimate the leader’s ex ante likelihood of getting promoted,
independent of her or his ex post economic performance during the term, and define it
as the leader's career-concern intensity.
We then empirically examine the relationship between a city leader's career-
concern intensity and the spatial pattern of urban land development. We find that a one-
standard-deviation increase in the leader's career-concern intensity leads to an 11 km
extra outward urban expansion, a 41% increase relative to the sample average.
According to our theory, the key mechanism that drives this positive relationship is a
land-fiscal mechanism. Specifically, by selling more land at the city edge, the city
government can gather more land sale revenues, which help finance the construction of
urban infrastructure. This in turn stimulates economic growth. We find evidence that
indicates that an extra 12 km of average outward expansion of urban land generates
extra land sale revenues of 10 billion RMB, and increases total local GDP by 45 billion
RMB, amounting to 84% and 20% of their sample averages respectively. In addition,
we find that that a one-standard-deviation increase in the leader's career-concern
intensity lowers the FAR level by 0.04, which explains 12% of the variations in FAR
across city leaders.
The above analyses focus on residential land because its sales accounted for nearly
three quarters of total land revenues for local governments and land sale revenues are
at the heart of our political economy story about urban land development. As a
supplementary analysis, we also examine the patterns of both industrial and commercial
land development. We find that industrial land exhibits an outward expansion pattern
similar to that of residential land; that is, during the terms of higher-incentive city
leaders, more industrial land is developed at the city edge. This evidence implies that
5
the outward expansion of residential land complements that of industrial land in
promoting the city’s GDP growth, which strengthens the logic of our political economy
story. By contrast, commercial land development has no such pattern in relation to the
local leaders’ career-concern intensity.
Our paper contributes to the literature by explicitly studying how city leaders as
city developers play an important role in shaping the pattern of urban land development.
China’s unique institutional background provides an interesting context for us to
explore this issue. Chinese city governments are responsible for stimulating local
economic growth, providing public goods, and promoting urbanization. Because the
promotion of city leaders is closely related to their cities’ economic performance, city
leaders with strong career concerns will naturally adopt land development policies that
are favorable to local economic growth. Unlike U.S. homeowners, China’s homeowners
have relatively weak ability to affect land use regulations. Our paper adds to the existing
literature 2 by offering an important complementary channel through which land
development regulations are determined. Moreover, our dataset demonstrates rich
variations across cities and over time, enabling us to identify this channel.
Our paper also relates to the literature concerning the role of local public finance
in land policies. One frequently discussed motive for land use regulations in the
literature is fiscal externality on the provision of local public goods (e.g., Epple, Romer
and Filimon, 1988; Calabrese, Epple and Romano, 2007). More recently, Henderson
and Venable (2009) demonstrate that revenues from land development play an
important role in the dynamic formation of cities. Using a panel dataset from coastal
provinces in China, Lichtenberg and Ding (2009) find evidence for the role of fiscal
incentives in expanding the area of urban land. Han and Kung (2015) show that fiscal
pressures induced by a tax-sharing reform tend to push the urban land area to grow
more in China. Our paper enriches this literature by introducing the role of local leaders’
career incentives in driving land sales and shaping land development patterns, and also
by explicitly demonstrating the trade-off between upward and outward expansion of
urban space.
There is a small yet growing body of literature on land development and
regulations in urban China (Deng et al.,2008; Lichtenberg and Ding, 2009; Brueckner
et al., 2016; Cai, Wang,and Zhang, 2016). However, none of these studies addresses

2
See Gyourko and Molloy (2015) for a comprehensive review.
6
how city leaders make decisions regarding urban land development policies.
The rest of the paper is organized as follows. Section 2 is a brief description of the
institutional background of land development as well as of local leaders' career
incentives in China. In Section 3, we present a simple theory that demonstrates the
decision-making process of city leaders with regard to urban land development. Section
4 discusses the data and the main variables. We present our primary empirical results
regarding the outward expansion of residential land in Section 5. In Section 6, we show
the empirical results regarding land use intensity. Section 7 briefly discusses the
development patterns of industrial and commercial land. Section 8 concludes.

2. Institutional Background
2.1 Urban Land Development Regulations in China
Land quota and urban growth control
In the 1980s and early 1990s, there was an unprecedented loss of arable land in
China due to large-scale industrialization and urbanization. The sizable decline in arable
land strengthened concerns about China’s food security.3 In the meantime, the conflicts
between local governments and farmers due to insufficient compensation for rural-to-
urban land conversion have intensified sharply. In order to keep urban land expansion
under control, the central government amended the Land Administrative Law in 1998
and enacted a new set of arable land protection provisions aimed at prohibiting any
additional loss of cultivated land. In the same year, a new ministry, the Ministry of Land
and Resources of the People’s Republic of China, was established to strengthen the
central government’s control over land development.
Since 1998, an urban land quota system has been implemented via a hierarchical,
top-down planning process. The central government makes the nation’s long-term land
development plan. The first plan covered the period from 1997 through 2010. The plan
dictates the maximum amount of newly developed urban land for each province in the
long term as well as the minimum amount of arable rural land in reserve. Given these
two important constraints, provincial governments make their own long-term plans for
land development and allocate land use quotas to cities under their administrative

3
At the beginning of the economic reforms, in 1978, the cultivated rural land per capita in China was
only 1.55 mu, far lower than that in the U.S.
7
control. In addition, governments at various levels also make short-term (five-year or
annual) land development plans that provide more detailed guidelines.
The city governments make decisions regarding the size and location of land to be
developed in a city. When land development involves conversion of rural land to urban
land, the conversion must be approved by the upper-level government. If the conversion
is within the quota constraints imposed by the upper-level government, then it is
relatively easy to obtain approval for the application. Otherwise, the city government
has to make extra efforts to lobby the upper-level government (sometimes even the
central government) for extension of the land quota.4
When land conversion is involved, the city government has to compensate the
farmers who currently use the land. By China’s Land Administration Law, the total
compensation fees should not exceed 30 times the average annual value of product
generated from the land within the three years before the conversion.5 This is typically
far below the market value of the land, which can be 500 times higher than the
compensation fees (Zhang, 2007; Liu, Chen and Zhao, 2012). Naturally, insufficient
compensation led to discontent among farmers and protests. However, in most cases,
the farmers are politically weak and have to obey the rules set by the local governments.
This is why developing rural land at the city edge is lucrative to city governments: they
can auction off the land at a much higher price after conversion. By contrast, brownfield
development near city centers usually involves high compensation fees (close to market
value) paid to urban residents, which makes the net revenues from such development
relatively low.
City governments can bypass the quota constraints on land conversion in various
ways (Xie, 2015). They typically figure out legitimate reasons to persuade the higher
governments to grant extra land use quotas after their short-term quotas are used up.
These reasons may include creating a college town at the city edge, building a cultural
and historic town (which typically requires relatively low FAR) or developing a high-
tech industrial park. As a result of these exceptions, the long-term land quota of land
conversion is often exhausted years before the end of the plan. For example, by 2004,

4
China’s Land Administration Law (1998) requires that any conversion of farm land to non-agricultural
use be approved by higher level authorities and be offset by conversion or reclamation of other land to
agricultural use so that the amount of land used for agriculture, adjusted for quality, remains constant
(Ding, 2003; Lichtenberg and Ding, 2008).
5
The law stipulates that the compensation fee can be increased to more than 30 times the upper bound
under some special circumstances. However, local governments are generally unwilling to do so.
8
Shandong and Zhejiang provinces had used up 80% and 99% (respectively) of the total
land quota of the long-term plan of 1997-2010. The zeal of local governments for urban
land development placed tremendous pressure on the central government when making
the next long-term development plan. Meanwhile, the city governments actively engage
in lobbying and bargaining with the upper-level governments for favorable treatment
in the next long-term plan, which may effectively legalize their current urban expansion
beyond the quota constraints set by the old plan. In addition, sometimes the city
governments even tolerate some small-scale land development at the city edge without
any approval from the upper governments. In general, going beyond the quota
constraint is possible, but at considerable cost. Costs include lengthy bargaining,
proposing “creative” rationales for extra quotas, and networking with upper-level
government officials.
After rural land is converted to urban land, it is at the disposal of the city
government. By law, all urban land is owned by the state. Since 1988, the use rights of
vacant urban land parcels have been allocated through leaseholds by city land bureaus.
In the 1990s, most use rights allocations were done by “negotiation” between
developers and government officials. To control the widespread corruption in such
negotiated land deals, the Ministry of National Land and Resources banned negotiated
sales on August 31, 2004. Since then, all urban leasehold sales for private development
have been conducted through public auctions. Land auctions are held by local land
bureaus, with details of all transactions posted online and publicly available. All land
sale revenues are turned in to the city treasury.

Land use intensity restrictions

Now let us turn to regulations on the intensity of land use. The authority that is in
charge of land use planning and general guideline setting is the city land reserve and
allocation committee, whose members include the city’s key leaders and bureau
directors from relevant government agencies. Apparently the city leaders have decisive
voices in urban planning and land development strategies. After setting up the urban
development strategies and guidelines, the committee typically delegates to the city’s
urban planning bureau the routine decisions, such as use type and details of
development restrictions (e.g., regulatory FAR limit, building height, green area rate,
etc.) for each parcel of land to be auctioned. When designing land use restrictions, urban

9
planning bureaus aim to maximize the value of land by weighing various benefits and
costs including the possible congestion costs caused by high-density development.
The FAR regulation is one of the most important land use regulations in urban
China. By law, any land parcel must have a designated regulatory FAR level before it
can be auctioned off. Also, after the land is developed, the city’s planning bureau is
required to complete an official inspection of the residential project before it is put up
for sale, in order to ensure compliance with the FAR regulation. In most cases, the FAR
regulation takes the form of an upper-bound constraint on the ratio of a building’s total
floor area to the lot size on which the building is to be constructed. As Cai, Wang, and
Zhang (2016) show, even though the FAR constraints can be adjusted under certain
circumstances, they are largely binding in urban China. Therefore, FAR regulations
play an important role in shaping the urban spatial patterns of Chinese cities. We use
the regulatory FAR upper limits to measure the land use intensity, which, intuitively,
can be thought of as the upward dimension of urban land development.

2.2 Local Leaders’ Career Concerns and Land-Finance Policies

Prefecture cities are under the supervision and control of provinces. In China's
centralized personnel system, it is the upper-level committee of the Communist Party
that determines the promotion of local leaders. The top leaders at the prefecture level
include the party secretary and the city mayor. The party secretary is more powerful
than city mayor due to the ruling position of the Chinese Communist Party. The regional
economic performance is one of the key performance indicators used by the superior to
evaluate local leaders (Li and Zhou, 2005; Xu, 2011; Yao and Zhang, 2015). Since the
establishment of mandatory retirement age in the early 1980s, age has become critical
to career advancement. When a local leader approaches the retirement age, he or she is
less likely to be promoted. This retirement system has been enforced rather strictly in
the past decades. The retirement age varies with the hierarchical level of local officials.
The retirement age of a prefecture-level official is 60 while that of a province-level
official is 65. 6 This means that both age and hierarchical levels are important
determinants of promotion, other things being equal. Given a local official’s current age
and hierarchical level, her distance from the glass ceiling of retirement can easily be

6
The political hierarchy levels at which a city leader may stand include prefecture level, deputy-province
level, province level or above.
10
estimated. Local leaders who are far away from the glass ceiling should have high
career-concern intensity. If a local leader is nearing the retirement age, she will have
little incentive to work hard to improve local economic performance.
Since the late 1970s, China has gone through several waves of fiscal reforms in an
effort to decentralize its fiscal system and fiscal management (Zhang and Zou, 1998).
Fiscal decentralization led to the perpetually declining share of central fiscal income,
which led the central government to act to turn this tide. In 1994, the tax-sharing reform
was implemented and consequently 75 percent of value-added tax, the largest source of
tax revenue, goes to the central government. Corporate income tax, which was originally
designated as a local tax in 1994, was reclassified as a shared tax between the central and
local governments after 2000. As a result of repeated rearrangements of tax revenues in
favor of the central government, local governments felt increasing fiscal pressures.
Against this backdrop, land sale revenues rose to prominence and became the major
source of extra-budgetary income for local governments. In the past two decades, city
governments have increasingly relied on land-leasing revenues to finance urban public
goods provision such as infrastructure investment (World Bank, 2012), which in turn can
help boost local GDP growth. Some studies show that the ratio of land sale revenues to
local fiscal revenues of prefecture city governments increased from 10% in 1999 to nearly
60% in 2005 (Han and Kung, 2015). It is worth mentioning that about three quarters of
land sale revenues come from the sale of residential land. City governments tend to
deliberately lower the sale prices of industrial land in order to attract the entry of new
firms, which again may enhance GDP growth.

3. A Simple Model
In this section, we present a theoretical framework to demonstrate how city leaders
make decisions regarding urban land development. We derive testable predictions from
this framework. Our model assumes a mono-centric city with linear form.7 All workers
commute to the Central Business District (CBD) to work. We assume free migration
across cities.8

A. Individual

7
See Duranton and Puga (2015) for a comprehensive review on urban land use in a mono-centric city
framework.
8
Adding migration friction to the model does not change the main predictions of our theory.
11
 An individual worker’s utility comes from housing services as well as from
consumption of a composite good of all other commodities. An individual worker’s
objective function is as follows:

max u ( x ( r ), h ( r ))  x ( r ) h ( r )1
x ( r ), h ( r )

s.t. , (1)
w  x ( r )  h ( r ) p ( r )  2tr
where r indicates the location where the worker lives, h(r ) is housing consumption,
x(r ) is consumption of the composite good, w is wage income, t is commuting
cost per unit distance, and p (r ) is unit floor area house price at location r.
Solving the individual’s utility maximization and considering the location
equilibrium condition, we have
1
   (1   )1 ( w  2tr ) 1
p(r )    , (2)
 V 
where V is the individual worker’s indirect utility and
(1   )( w  2tr )
h(r )  . (3)
p (r )

B. Firm
Assume the city's production is conducted by a representative firm. The firm's
production function is
Y  AGL , (4)
where A is technology, G is public infrastructure provided by the city government,
and L is labor.9 G improves the productivity of labor. We assume that the firm sells
products in the national market and there is free trade among cities. The product price
is normalized to be 1.
Prefect competition among firms gives zero profit, which implies that
w  AG . (5)
Eq. (5) pins down the wage rate. Suppose each worker provides 1 unit of labor.
When the labor market clears, L=N, where N is the city’s population size.

9
For simplicity, we do not include capital in the production function because it is not the focus of this
paper. One can easily incorporate capital into the model by assuming an infinite elastic supply of capital
to the city from the national capital market. This would complicate the model solution without providing
further insights.
12
C. Build upward: floor-to-area ratio
Informed by the institutional background in urban China as discussed in Section
2, we assume that the city government delegates an agent to set the floor-to-area ratio
F ( r ) for land development at each location r . In making the decision, the agent aims
to maximize the value per unit of land; that is,
F (r )  arg max F p(r ) F  q( F ) F . (6)

The construction cost per unit floor area is q( F ) . The construction cost q( F )
increases monotonically with F . For example, foundation costs may increase with
building height. These costs may also include expenditures incurred in correcting
negative externalities such as noise and traffic jams that arise from high-density
development. Note that the assignment of FAR by the agent is similar to what a private
developer would do (DiPasquale and Wheaton, 1996), except that the agent considers
the possible externality costs. Let the construction cost per unit floor area be
q ( F )   F ,   0 . Then from (6), the optimal FAR level is
p(r )
F (r )  . (7)
2

D. Build outward: the benchmark case where the market determines the urban
boundary
Let us first look at the benchmark case where it is not the city government, but the
market force that drives urban outward expansion. In this case, the city government
does not assume the developer role. Private developers choose which land to develop.
They obtain land from current users at the market price, denoted as pl () . Land at any
location r is developed only when it can generate non-negative profits. Competition
among developers bids up the land price at each location such that in equilibrium, the
marginal revenue from developing land at each location equals the unit land price. The
urban boundary S is thus determined by the following condition:
p(S )F (S )  q(F (S ))F (S )  pl (S )  p . (8)

The left-hand side of the above equation is the marginal revenue of urban land
development at the city edge. The right-hand side is the market price of land at the city
edge, which is equal to p ; that is, the opportunity cost of converting agricultural land

to urban use. In equilibrium, this equality must hold.

13
 The city government collects land sales tax at a rate t . The tax revenues are all
used to finance urban public infrastructure G . Thus the government’s budget
constraint is
S
G  t  pl ( r ) dr  t
0  0
S S

p (r ) F (r )dr   q ( F ( r )) F (r )dr .
0
(9)

Also note that in equilibrium, free migration across cities implies that
V v , (10)
and the housing market clearance implies that
S F (r )
0 h(r )
dr  N . (11)

Combining the above three equations (9)-(11) with equations (2), (3), (5) and (7), we
can solve for the equilibrium for this benchmark case. Let us denote the equilibrium
urban boundary in this benchmark case as S0 .

E. Build outward: the planning case where the city government assumes the role of
developer
Now let us turn to the case where the city government behaves as a developer and
sets the urban boundary. In making a decision regarding urban boundary, a city leader
considers her own pay-offs which derive from the success of her political career. If
promoted, the city leader’s payoff is K  0 ; otherwise, her payoff is zero. All the
revenues from land development are collected by the city government and are all used
to finance urban public infrastructure, G , which will improve the city’s total factor
productivity and boost GDP growth. Because greater fiscal revenues can help her
political career, the city leader has incentive to grab more revenues from land
development.
The net revenues from land development are equal to the total house value
minus the land compensation cost and construction cost. 10 Compensation costs are
unavoidable in land development. Such costs typically include the costs of demolishing
existing structures and compensating current residents. In our model, we use a
continuous function, c ( r ) , to measure the compensation costs. As such, the
government’s budget constraint is

10
In urban China, the city government is the monopoly supplier of city land. It holds auctions to sell
long-term leaseholds of land to private developers. Through competitive bidding, these land sales can
grab the developers’ profits from land development.
14
 S S S
G   p(r ) F (r )dr  q( F (r )) F (r )dr   c(r )dr . (12)
0 0 0

Developing land near the CBD usually involves much higher compensation costs
(close to the land market value) than developing land far away from the CBD (Zhu,
2004). Thus, we assume that c ( r ) decreases with r. Further away from the CBD, the
compensation costs deviate more and more from the market value of the land. In
particular, at the city edge, to convert farm land to urban land, the city government is
only required to pay the farmers a compensation cost that is far below the market value
of the land (see Section 2.1). Thus if the city edge is still at the previous market
equilibrium S0 , the LHS of equation (8), which is the marginal revenue from land
development, would exceed the compensation costs.11 Intuitively, this will induce city
governments that crave additional revenue to expand their city boundaries beyond S0 .
Specifically, the decision process is as follows. Let us use the city’s total output
Y to capture economic performance in the model. We use a parameter  to capture
a city leader’s career-concern intensity, which depends on her personal characteristics
such as age and hierarchical level when taking office. The higher  is, the further
away the local leader is from the career glass ceiling, so the leader is more intensely
concerned with her career. As such, we define the city leader’s promotion likelihood
function as f (  Y ) , with f '()  0 . So the leader’s promotion likelihood is an

increasing function of the product of her career-concern intensity and her ex post
economic performance. 12 When the career-concern intensity is higher, improving
performance can increase the leader’s chance of promotion more effectively.

11
For illustrative purposes, suppose that the tax rate in the benchmark case and the compensation cost
schedule in the planning case are such that the total government fiscal revenue of the planning case is the
same as that of the benchmark market equilibrium if the city edge is set at . Because the fiscal revenues
are all used to finance public infrastructure, this means the productivity and wage income (see equation
(5)) is the same in the two cases as well, which in turn leads to the same house prices and FAR levels
(see equations (2) and (7)). Simple calculations then show that the LHS of equation (8) is the same for
the two cases. According to the definition of the benchmark market equilibrium, the LHS of equation (8)
equals the land market value at the city edge , therefore, we can conclude that the LHS of equation (8)
is higher than the compensation cost in the planning case.
12
In China’s political system, the likelihood of promotion of a politician hinges on both the politician’s
personal conditions prior to her taking office and her ex-post performance during the term. The two
factors are complementary to each other. People may be concerned that a leader’s effort may decrease
with her ex-ante promotion chance when her ex-ante chance gets sufficiently high as documented in the
literature on western political regimes (e.g., Besley, Persson and Sturm, 2010; Sole-Olle and Viladecans-
Marsal, 2012). To explore this nonlinear effect of λ on the choice of the spatial pattern of urban land
development by the city leader, we run regressions that include both λ and its quadratic term. We also
draw non-parametric graphs of the relationships between the spatial patterns of urban land development
and our measure of λ. We find no evidence of such nonlinear effect due to the prior chance of promotion.
15
 The city leader faces a trade-off. On the one hand, expanding the city outward can
bring in more fiscal revenues so as to finance public infrastructure and accommodate
more people in the city, which in turn boosts GDP growth. On the other hand, pushing
the urban boundary outward usually requires the city leader to seek more urban land
quotas from the upper-level government, which is costly. As discussed in section 2.1,
when the central government realized that the city governments’ costs of obtaining land
at the city edge were much lower than the market value of land, it started making long-
term plans for urban land development that set quotas of rural-to-urban land conversion
for each province. In these plans, the provincial governments allocate these quotas
across cities and over time. Such quota constraints are represented by constraints on the
city’s boundary S in our model.
Suppose that the provincial government sets the city boundary constraint at the
benchmark market equilibrium boundary S0 .13 If a city leader wants to expand the

spatial boundary of her city beyond S0 , she needs to lobby the upper-level
governments. In this sense, such boundary constraints are not strictly binding but are
subject to costly bargaining. As described in Section 2.1, Chinese city leaders use
various ways in practice to seek additional quota and push the urban boundary outward.
The lobbying efforts for more urban land quota are reflected in the effort cost function,
denoted as E (S ; S0 ) . We assume that these lobbying costs increase with S when the

city expansion goes beyond the boundary constraint S0 . Specifically, we define the
effort costs of the city leader as
0, if S  S0
E ( S ; S0 )   , (13)
e   ( S  S0 ) , if S  S0


where   0 ,   0 , and e  0 .
The city leader's objective function is described as follows:

13
In our simulation, we suppose that the city boundary constraint is set by the upper-level government
at the benchmark level of the market equilibrium. Our model’s main predictions are robust to other
choices of .
16
 max f ( AGN ) K  E ( S ; S0 )
S

s.t.
S S S
(i ) G   p(r ) F (r )dr   q( F (r )) F (r )dr   c(r )dr (14)
0 0 0

(ii ) V  v
S F (r )
(iii )  dr  N
0 h( r )

In (14), (i) is the city government’s budget constraint, (ii) is the intercity spatial
equilibrium condition under free migration, and (iii) means that the total housing supply
should accommodate the total housing demand.

F. Solution and simulation

Plugging equations (2), (3), (5), and (7) into (14), we can derive the first-order
conditions for S . With the first-order conditions, we can solve for four unknown
variables, i.e., S , G , N and Y . These fully characterize the equilibrium.
Since an analytical solution is impossible to obtain, we use reasonable parameters
to conduct simulations. From the simulations, we can do comparative statics and
demonstrate how urban land development patterns are determined by relevant factors.
Figure 3 shows that stronger career concerns drive a city leader to expand the city
boundary more outwardly. Figure 4 helps us further understand the mechanism
underlying such outward expansion. It shows a positive relationship between the city's
spatial boundary and several economic outcomes in our model. The more a city expands
outwardly, the greater its fiscal revenues, population and total output. These factors will
enhance the city leader’s chances of getting promoted.
In addition, outward urban expansion has an implication on upward building
intensity; i.e., the floor-to-area ratio. As a city’s urban boundary expands outward, if we
keep the wage and population of the city unchanged, individuals can have greater
housing consumption. This implies that the individual’s utility level increases.
According to eq. (2), the housing price at each location declines. As a result, the FAR
level at each location decreases too, according to eq. (7). This reflects a trade-off
between outward and upward urban expansion.
In sum, we have the following testable predictions. First, city leaders with high
career incentives are more likely to expand the city outward. The underlying
mechanism that drives this is that by doing so the city leader can generate more land

17
sale revenues for financing public infrastructure, which in turn enhances her city’s
economic performance. Second, the land use intensity is lower for city leaders with
higher career incentives due to the existence of a trade-off between a city’s upward and
outward expansion.

4. Data and Variable Construction

4.1 Measures for the Spatial Pattern of Urban Land Development
To characterize the spatial pattern of urban land development, we exploit the
database on all residential land transactions through public auction14 provided by the
China Index Academy, China’s largest independent think tank focusing on the real
estate market. The sample for our analysis contains information on over 30,000
completed residential land transactions in 202 Chinese cities during the period 2000-
2012.15, 16
For each land parcel, we know the use type, land area, regulatory FAR,
reserve price, selling price, transaction date, auction type, etc.
As for outward urban development, we are essentially interested in measuring the
extent to which a city’s edge of urban land development is pushed further away from
the city center. Taking advantage of the micro-level land data, we are able to do so with
detailed geographic locations of the city’s residential land developments. Specifically,
we obtain the geographic coordinates for each land parcel from www.map.baidu.com.
Using the coordinates, we calculate for each land parcel the distance to the city center.17
Panel A of Table 1 shows the summary statistics of land distance to the city center by
land transaction time and by region (i.e., coastal, central, northeast, southwest, or

14
The Chinese government passed a law in 2002 requiring all the land intended for business purposes
(including residential, commercial and industrial land) to be sold through public auction. This law was
strictly enforced beginning in August 2014. We understand that there still exist some land transactions
through negotiated sales instead of public auction. However, we note that most negotiated sales involve
industrial land. Even for residential land sold through negotiations, the sale price is far below the market
value (less than 100 RMB per sq. meter). This suggests that such negotiated sales are not for the purpose
of creating fiscal revenues for the local government which is at the heart of this paper’s story. Thus our
land transaction data from public auctions is representative and suitable for the research purposes of this
paper.
15
Among the total of 202 cities in the sample, 198 are prefecture-level cities, and 4 are provincial-level
cities (i.e., Beijing, Shanghai, Tianjin, and Chongqing).
16
The original sample includes 36,058 residential land parcels with completed sales. 5,229 land parcels
are dropped because their regulatory information (i.e., FAR) is missing. An additional 11 land parcels are
dropped because the information on their geographic location is not available. Finally, 801 land parcels
are dropped because they lack information on city leaders. The final study sample contains 30,017
residential land parcels.
17
We use the coordinates of the 1992 light center (i.e., the brightest cell at night in each city’s central
area) from Baum-Snow et al. (2016) to identify the actual city center. They suggest that despite the
enormous increase of light over the past two decades, the light centers have remained unchanged.
18
northwest). More than half of the residential land transactions happened in the coastal
regions. Around the country, the average and 90th percentile in the distribution of the
land distance to the city center were 15.5 and 35.7 km (respectively) during 2000-2008
and they increased to 25.3 and 63.2 km (respectively) in the period from 2011-2012,
which indicates a marked outward expansion of city boundaries over time. Across
regions, one can see that coastal cities in general have more outward urban development
than the nation’s average.
To investigate the relationship between outward development and local leaders’
career incentives, we measure the extent of outward development during the term of
each city leader with the top percentiles in the distribution of the land distance to the
city center among all the land parcels sold during the city leader’s term. The outward
expansion of urban land development is a typical form of urban spatial expansion.
Previous studies usually measure the expansion by the change in total urban land area
(e.g., Brueckner and Fansler, 1983; McGrath, 2005; Deng et al., 2008; Lichtenberg and
Ding, 2009). A clear limitation of such aggregate measures is that the growth of urban
land may be caused by either new developments filling in open space among existing
developments or by developments pushing the existing urban boundary further away
from the city center. Rigorously speaking, the latter type of development is more
consistent with the outward expansion of urban boundaries and our outward measure is
better able to capture it; also, it is more relevant to our land-fiscal revenues story. In
Table A1 of the appendix, we also check the correlation between our outward measure
and other urban land area measures. The table shows that the increase in our outward
measure is positively correlated with the growth of urban land area based on several
widely used data sources, such as Landsat images, the Chinese Yearbooks of Land
Resource, and the Chinese Urban Statistical Yearbooks.18
We use the regulatory upper limit of FAR level for each land parcel to measure the
land use intensity, i.e., the extent of upward development. The regression analysis
examining the relationship between upward development and local leaders’ career
incentives is at the land parcel level, so we can control for the effects of location

18
The Geography Institute of China’s Academy of Science provides detailed land-type information for
all the cities nationwide based on the US Landsat TM/ETM images (a satellite remote sensing data) for
the years 2000, 2005, 2008, 2010, and 2013, from which we can calculate the area of all built-up areas
that are contiguous to urban settlements for the corresponding year. The Chinese Yearbooks of Land
Resource provide data on the area of newly added urban land on a yearly basis from 2003 to 2013. The
Urban Statistical Yearbooks report the built-up land area between 2000 and 2013.
19
attributes on residential building density. Panel B of Table 1 reports the average
regulatory FAR levels by land transaction time and by region. In general, the average
FAR has increased over time. Compared with interior cities, the coastal cities on
average have lower regulatory FARs.

4.2 Measures for the Career-Concern Intensity of City Leaders

In China, the party secretary of a city is the highest ranked city official and holds
authority over all key decisions concerning economic and social issues, while the city
mayor, who has a lower rank than the party secretary, focuses more on the daily
operation of the city administration and on policy implementation. As such, this paper
treats a city’s secretary as the city’s leader. Chinese government officials are ranked by
formal hierarchical levels which closely correlate with their political power and
resources. By and large, promotion means moving from a lower level to some upper
level.
The hierarchical ranks the leaders of Chinese cities can enjoy consist of four
possible levels, in descending order: politburo, province, deputy-province, and
prefecture. The hierarchical ranks of city leaders are correlated with the administrative
status of the cities of which they take charge.19
We collected information on 974 city secretaries who were in office from 2000-
2012 in the 202 Chinese cities matched with our land sample. The information includes
their date of birth, education attainment, their start time and end time of office, their
hierarchical level when they first took office, and their power status when they left
office (e.g., whether promoted, laterally moved or retired, etc.). Among the 974 city
secretaries, at the time when they took office, 798 were prefecture-level leaders, 157
were deputy-province level leaders, 10 were province level leaders, and 9 were
politburo level leaders.
The city leader’s age at the beginning of the term (“start age”) matters a lot in
terms of chances for promotion. For example, given that the mandatory retirement age
for prefecture-level politicians is 60, younger city leaders should have a longer career

19
A majority of prefecture city leaders hold prefectural-level positions, although a small portion of them
hold deputy-province level positions. The party secretaries of four centrally administered cities (Beijing,
Shanghai, Tianjin, and Chongqing) are mostly province-level (though some of them are politburo-level
leaders). The party secretaries of 15 significant cities (e.g., certain provincial capital cities and all
separate-planning cities) usually hold deputy-province-level positions. The 15 deputy provincial-level
cities are Qingdao, Shenyang, Dalian, Changchun, Ha’erbin, Nanjing, Hangzhou, Ningbo, Xiamen, Jinan,
Wuhan, Guangzhou, Shenzhen, Chengdu, and Xi’an.
20
horizon and therefore they should have greater incentives to seek good economic
performance.20 Moreover, the hierarchical level at the starting year of the term (“start
level”) also matters for promotion prospects. For example, the retirement age is 65 for
provincial-level city leaders and 68 for politburo city leaders. Also, although deputy-
province-level city leaders have the same retirement age as prefecture-level city leaders,
there are fewer significant positions available at higher hierarchical levels than deputy-
province. Therefore, after the current terms end, they are more likely to retire or to move
to some honorary positions in the Chinese People’s Political Consultative Conference
(CPPCC) or the People’s Congress (PC) at either the provincial or national level. As
such, they have relatively low career incentives compared to prefecture-level leaders.
We define a promotion dummy variable, which equals one if the city leader gets
promoted to a higher-level position by the end of the term. Note that the new position
cannot be an honorary position in CPPCC or PC. Otherwise, even if the city leader’s
hierarchical level moves up, it is not counted as a promotion. Therefore, the promotion
dummy is set to be zero if by the end of the term, the city secretary either stays on the
same level or retires or moves to some honorary positions in the CPPCC or PC at either
the province or national level. 21 We observe the promotion outcomes of 735 city
secretaries. Among them, 258 city leaders were promoted.
Panel C of Table 1 shows the summary statistics for the 974 city leaders in our
sample. As shown in Panel C of Table 1, about 70% of the province- or politburo-level
city leaders were promoted. While 40% of prefecture-level city leaders successfully
climbed to a higher hierarchical level, only 6% of the deputy-province-level city leaders
were promoted. Panel D of Table 1 reports the percentage of leaders promoted by start
age for each start level. For prefecture-level city leaders, it is clear that the likelihood
of promotion strictly decreases with start age, from about 60% for the age interval of
38-47 to around 14% for the age interval of 53-65.
Thus a city leader’s start age and level determine her career incentives at the
beginning of the term, which in turn will affect her effort level during the term and

20
The retirement scheme and regulations on terms of cadre in China have served as powerful tools to
guarantee that relatively young politicians reach leadership positions (Li and Zhou, 2005; Kahn, Li, and
Zhao, 2014).
21
For city leaders who were either arrested or died during their terms, or who had not finished their
terms by the end of 2015, we consider their promotion status as being missing. We also experiment with
alternative definitions of promotion. For instance, we set the promotion dummy to be zero if the city
leader was either arrested or died during her term, or if the city leader had not finished her term by the
end of 2015. The main results are similar.
21
hence her actual chance of being promoted later on. To measure the career-concern
intensity of city leaders, we first run regressions of the promotion dummy on start age,
the dummies of start levels, and the interactions of start age and dummies of start levels.
Column (1) of Table 2 shows the LPM results, and column (4) reports the Probit results.
The relationships between promotion, start age and start level are consistent with the
patterns shown in Table 1. Variations in start age and hierarchical level explain about
20% of the variation in promotion. For 697 observations, we also know whether the
city where the politician was currently working was located in her birth province, and
whether the politician worked previously in the central government, both of which are
considered relevant to promotion by the Chinese political literature (Li and Zhou, 2005;
Zhang and Gao, 2007). As shown in columns (2)-(3) and (5)-(6), these two variables
alone explain only a tiny part of the variation in promotion and the estimate results are
robust to including these two variables. Note that we intentionally avoid including
variables that are endogenous to the actual effort level, such as the total tenure length.
Using the parameter estimates reported in column (4), we predict the prior
promotion likelihood for each of the 974 city leaders based on start age and start level
only. Each politician’s career incentives should be an increasing function of the
predicted prior promotion likelihood. As such, we measure the career-concern intensity
of each city leader by the predicted promotion likelihood.

4.3 Event Study: City Leader Turnovers and Spatial Pattern of Urban Land
Development
Before doing regression analysis, we first use an “event-study” approach to
illustrate what happens to the spatial pattern of urban land development when a city is
undergoing a turnover from a city leader with low career-concern intensity to one with
high career-concern intensity. Specifically, according to the career-concern intensities
estimated in Section 4.2, we group all city leaders in our sample into two categories —
the high-incentive group if their career-concern intensities are above the median (0.5),
and the low-incentive group otherwise. Then we select those cases from the entire
sample where a city experiences a turnover from a high-incentive leader to a low-
incentive one, or vice versa. Further restricting the sample to those city leaders who
sold at least one residential land parcel during their terms, we identify 19 high-to-low-
type turnovers of city leaders and 13 low-to-high-type turnovers of city leaders. As
shown in Figure 5, within the same city, switching from a high-incentive city leader to
22
a low-type one is associated with a sharp decrease in the outward expansion of the city’s
urban land development, as measured by the 90th percentile of distance to the city center
of newly-developed land. It is also associated with a sharp increase of land-use intensity,
as measured by the average regulatory FAR limits. Although it is still preliminary, this
exercise clearly shows that the spatial pattern of urban land development, measured in
both outward and upward dimensions, has a close relationship to city leaders’ career
concerns.

5. Empirical Analysis: The Outward Dimension of Land Development

5.1 The Impact of Career Concerns on Outward Development
We use the following specification to examine the relationship between a city
leader’s career-concern intensity and the extent of the city’s outward development
during her term:
zsp,c   out , p s ,c  city fixed effects +X s ,c  us ,c , (15)
p
where z s , c represents the p-th percentile of the distance to the city center of land

parcels sold during city leader s’s term in city c, s,c is city leader s’s career-concern

intensity, and u s , c is an error term. Including the city fixed effects allows us to

compare the spatial patterns of urban development between different city leaders within

the same city. We expect 

out , p
to be positive as our model predicts that city leaders
with high career incentives are more likely to expand the city outward.

In (15), Xs,c represents a set of control variables. We use the dummies of the year

when each city leader takes office (“on year”) and the dummies of the year when each
city leader leaves office (“off year”) to capture the national trends of urban outward
expansion at the turnover time of city leaders. Province-specific linear trends of both
on year and off year are included to address the concern that the arrival of a high-
incentive city leader may be associated with the regional policy changes (which usually
cover one or a few adjacent provinces) that may influence the urban development of a
city. 22 In addition, to address the concern that high-incentive city leaders may be
systematically appointed to growing cities, we control for the economic conditions

22
The results remain unchanged if we instead control for the province-specific on-year dummies, except
that the estimates’ significance drops as the regression’s degree of freedom sharply decreases.
23
before each city leader took office, including the levels of various economic variables
in the year before the city leader took office and the percentage changes of these
variables over the two years before the city leader took office. In the baseline
regressions, these variables include total registered population, GDP per capita,
investment in fixed assets, and investment in real estate development. In all regressions,
we also control for the square root of built-up land area at the end of the year before the
arrival of each city leader. We have a total of 386 city-secretary-level observations
matched with the land sample of 30,017 transactions, resulting in a sample of 377
observations containing complete information on all relevant city level variables.23
The heterogeneity in the physical landscape may affect the cost of outward
expansion (Burchfield et al., 2006). In addition, the compensation fee for land
conversion has risen over years due to the increasing bargaining power of farmers (Zhao
and Webster, 2011). We believe that controlling for the city fixed effects and time
controls (including both time dummies and province-specific time trends) in the
subsequent regressions should address these concerns.
Columns (1)-(3) of Table 3, Panel A present the regression results for the 90th
percentile, 85th percentile, and mean of distance to the city center. The growth in the top
percentiles represents outward expansion beyond the fringe of existing urban residential

development. Consistent with the model’s prediction, the coefficient 

out , p
is positive
and significant for both the 90th and 85th percentiles, suggesting that a city leader with
greater career concerns tends to promote an outward-oriented urban development. The
local leader’s effect is economically large: a one-standard-deviation increase in career-
concern intensity (0.21) leads to an increase of the 90th percentile distance to the city
center by about 11 km, representing about a 41% increase over the sample mean (which
is 27 km). Column (3) shows that the effect of career concerns on the mean distance to
the city center is positive but small in magnitude and insignificant, suggesting that the
positive effects of career concerns are mainly reflected in the expansion of urban
fringe.24
One concern is that the urban development in the four provincial-level cities, i.e.,

23
Among the 974 city leaders in our sample, 588 of them do not have any residential land transaction
records in our data. A simple comparison shows that within a given city, the career-concern intensities of
leaders who have land transaction records are not statistically different from those of leaders who do not.
24
Table A2 in the appendix presents the regression results for different percentiles of distance to the city
center. The effects of career concerns on the lower percentiles turn gradually from being positive and
significant to negative and insignificant as we move from the 90th percentile to the 10th percentile.
24
Beijing, Shanghai, Tianjin, and Chongqing, may be heavily influenced by the will of
the leaders in the central government. Also, the party secretaries of these four cities may
have quite different career horizons as most of them eventually ended up as central
leaders of the country. Columns (4)-(6) of Panel A in Table 3 show that the results are
fairly robust to exclusion of the observations from these four large cities.
Note that each regression in Table 3 is weighted by the number of land parcels sold
during the term to reduce the efficiency loss due to the heteroskedasticity of the outward

measures. Because s is a generated regressor, the standard errors are calculated on

the basis of 1,000 bootstrap replications.

5.2 Robustness Checks

In this subsection, we conduct several robustness checks to address various
concerns regarding our identification strategy. These results are reported in Panel B of
Table 3 as well as in columns (1)-(3) of Table 4.

Additional controls
A city leader who has greater career concerns may stay in office for a longer time.
Meanwhile, the total tenure length may affect the number of total land parcels sold
during the term, which in turn influences the outward urban expansion. Columns (1)-
(3) of Panel B in Table 3 show that the main results remain robust to inclusion of the
city leader’s total tenure length.
According to the political favoritism literature (e.g., Jia, Kudamatsu and Seim,
2015; Chen, Henderson and Cai, 2016), high-incentive city leaders may have better
political connections which can help them bargain more easily for additional rural-to-
urban land quota. To address this concern, we use a dummy indicating if the city leader
previously worked in the central government as a proxy for her political connections.
We find no evidence of a systematic relationship between our career-concern intensity
measure and this political connection dummy. As shown in columns (4)-(6) of Table 3,
Panel B, the main results change little after we further control for this political
connection dummy.
Jia, Kudamatsu and Seim (2015) suggest that older party officials tend to have
more power than junior ones while officials of similar age compete with each other for
high office. Thus, it is possible that city leaders show more loyalty towards provincial

25
leaders who are older than they are and in turn are more highly favored by those older
upper-level superiors. On the other hand, a younger provincial leader tends to carry out
a more sensible match between incentives (in terms of appointing city leaders) and
resources (in terms of granting land quotas) in order to maximize the province’s overall
economic performance. To address these concerns, we further include the age of the
provincial leader as of the year when the city leader took office. The results are fairly
robust as shown in columns (7)-(9) of Panel B in Table 3.
Finally, urban growth due to the industry composition effect and changes in
national industry structure may be correlated with the outward expansion of urban land.
Meanwhile, the appointment of a city leader may be affected by the city’s initial
industrial structure. To address this concern, we construct a Bartik-type variable that
measures the industry composition effect of structural changes during the term of each
city leader. Specifically, we multiply a city’s employment share in each industry sector
at some early year by the national growth rate in the corresponding sector during the
term of the city leader (Bartik, 1991)25 and then take the average across all sectors. As
shown in columns (10)-(12) of Table 3, Panel B, the main results remain robust after
we include this control variable.

Pre-trends of outward development

The time span of a city’s land development may go beyond the term of a single
city leader. Thus, the results in Table 3 may be driven by the possibility that the spatial
pattern of urban land development under the current city leader simply continues that
of her immediate predecessor. In this case, the correlation between a city leader’s career
incentives and urban land expansion that we observe in Table 3 is spurious. In order to
address this concern, we investigate the relationship between the career-concern
intensity of the current city leader and the spatial pattern of the land parcels sold in the
3 years just before she took office. For this exercise, we end up matching 14,097 land
parcel sales with 183 city leaders from 131 cities in our sample period.
Columns (1)-(3) of Table 4 report the regression results that respectively relate

25
The yearly data on national employment by industry sector is obtained from the website of the
National Bureau of China. This data is available for the time period 2004-2014 and includes a total of 19
industry sectors. The data on each city’s employment by industry sector is from the Urban Statistical
Yearbooks. The earliest year when the city-level employment data has the same definitions of industry
categories as those of the national-level data is 2004. Thus, we use the city-level data in 2004 to calculate
each city’s initial industrial structure.
26
the 90th percentile, the 85th percentile, and the mean of the distance to the city center of
the land parcels sold in the 3 years just before the current leader took office with her
career-concern intensity. The estimated coefficient on career-concern intensity turns out
to be negative and insignificant for all three percentile outcomes, contrary to what we
found in Table 3. Therefore, we can conclude that the effect of career incentives on
urban outward development is not driven by the continuance of the predecessor’s spatial
pattern.

5.3 Underlying mechanism: Land sale revenues, GDP, and outward expansion
The OLS results
According to our theoretical model, the underlying mechanism that drives a
strong-incentive city leader to outwardly develop the city more than a weak-incentive
leader is that by doing so she can generate more land sale revenues to finance public
infrastructure. Increasing public infrastructure in turn boosts GDP growth and thus
improves her promotion prospects by the end of her term. To check whether this
mechanism works, we investigate the relationships between the degree of a city’s
outward expansion and the city’s land sale revenues and GDP. Note that we run city-
year level regressions for this mechanism check.26 We use the mean distance to the city
center among all the land parcels sold in each city in each year to measure the yearly
outward expansion. 27 We have a total of 871 city-year-level observations for our
regression analyses.
The regression specification is given by
yc,t   zc,t  city fixed effects  year fixed effects  Vc,t 1  c,t , (16)

where z c , t represents the mean distance to the city center of all the land sold in city c

in year t, Vc ,t 1 is a set of city characteristics in the previous year, such as the city’s

levels and growth trends of registered population size and GDP per capita as well as the

square root of its built-up land area, and c ,t is an error term. We use city fixed effects
to remove the effects of time-invariant city characteristics and we use year fixed effects

26
Instead of collapsing data at the city-secretary level, we run city-year regressions here in order to
obtain a larger sample size and stronger instruments, as discussed later.
27
The effect of noise increases as the study time period is shortened, and the top percentiles are more
likely to suffer from measurement errors due to noise than the mean. Therefore, we choose to focus on
the mean distance when the study time period is one year. Note that, on average, a single term of a city
leader lasts for 4 years.
27
to control for national shocks. In the left-hand side of (16), yc ,t represents various

social and economic outcomes in city c in year t (land sale revenue, GDP, etc.).
We collect data on each city’s annual land sale revenues from the Yearbooks of
Land Resource 2001-2013. Other city-level data including citywide GDP (measured as
value added) and its sectorial decomposition in primary, secondary, and tertiary
industries, registered urban population size, number of employed persons in urban
private enterprises and self-employed individuals are obtained from the Urban
Statistical Yearbooks 2001-2013. All price variables are converted to constant 2012
price levels using the provincial-level CPI reported by the National Bureau of Statistics
of China.
Columns (1) and (3) of Table 5 report the OLS results for land sale revenues and
GDP, respectively. The effect of outward expansion on land sale revenues is positive
and significant, while the effect of outward expansion on GDP is positive but less
significant.

The 2SLS results

The OLS regressions may yield biased estimates due to the endogeneity of urban
land expansion. The degree of outward expansion of a city may correlate with some
unobserved city-level shocks, which may also affect the local land sale revenues and
GDP. For example, the entrance of new industrial plants may increase the need for
expansion of urban land at the city edge and boost the local GDP at the same time. In
such a case we will end up with an upward bias in the estimate of  . We may also
have a downward bias in the estimate of  if a negative shock hits the local rural
sector. This may result in cheaper farmland and at the same time a stagnant local
economy.
We address this endogeneity issue by seeking a valid instrumental variable for
urban land expansion. Motivated by the institutional features of China’s personnel
system and land regulations described in Section 2, we use the turnovers of provincial
leaders as instrumental variables for the city’s degree of outward expansion.
As discussed in Section 2, a city’s land development plan must be approved by its
upper-level governments.28 The turnovers of provincial leaders may affect the extent

28
For cities with populations above one million, land use plans must be further approved by the State
Council. The land use plan specifies the maximum amount of agricultural land to be used for construction,
28
of the city’s outward expansion through the implementation of the provincial quota
constraints. If the new provincial leader strictly enforces the land quota, then there
exists little room for a city leader to bargain for extra land quota. It is considered that
provincial leaders with greater promotion opportunities tend to enforce orders from the
central government more strictly. The literature has shown that age is a critical
determinant for the career concerns of provincial leaders (Li and Zhou, 2005; Kahn, Li,
Zhao, 2014). As such, using turnover information for provincial secretaries, we
construct the instrumental variable, an interaction term between a dummy indicating
whether city c is led by a provincial secretary who has a younger start age than his
immediate predecessor in year t, and the absolute value of the start age gap between the
current provincial secretary and his immediate predecessor. We essentially compare the
current provincial leader with his immediate predecessor because the land quota budget
of the former is more likely to be affected by the land-use quota decisions of the latter,
given that they are both subject to a fixed land quota constraint designated in the long-
term master plan of the central government.29
Column (5) of Table 5 reports the first-stage result. Given that the provincial leader
is younger than his immediate predecessor, a one-year increase in the gap of their start
ages will lead to a decrease in the mean distance to the city center of about 1 km. This
suggests that, relative to his immediate predecessor, a younger provincial leader is more
likely to stick to the land quota designated by the central government.
Columns (2) and (4) of Table 5 report the 2SLS results for land sale revenues and
GDP, respectively. The effects of outward expansion on land sale revenues and GDP
are positive and significant. A one-standard-deviation increase in the mean distance to
the city center (12 km) leads to an increase in land sale revenues of 10 billion yuan and
an increase in GDP of 45 billion yuan in 2012 RMB, representing about 84% and 20%
of the sample averages of the land sale revenues (11.9 billion) and GDP (229.2 billion),
respectively.30
To have the exogenous condition of a valid instrument hold, we need to assume

the maximum amount of cultivated land to be developed, the maximum per capita land use in cities and
towns, the minimum amounts of protected farmland and “cultivated land,” and so on. For more details,
see Feng, Lichtenberg, and Ding (2015).
29
The first national land use master plan covers the period from 1997-2010, and the second one covers
the period from 2006-2020. Immediately consecutive provincial leaders are likely to be subject to the
same land quota constraint, which is specified in the master plan.
30
For the main regressions in Tables 5 and 6, we include four provincial-level cities. The results remain
unchanged if we drop these four cities.
29
that, for the same city, after we control for the initial economic conditions, the turnovers
of provincial leaders only affect the degree of urban outward development, and do not
directly influence the city’s economic outcomes. One may be concerned that a
provincial leader who is much younger than his predecessor is also more likely to carry
out local economic policies to promote GDP growth within his jurisdiction. To address
this concern, we control for provincial-level GDP in each year (in log) in all the
regressions, assuming that the possible direct effect of our instrument variable on local
economic performance is province-wide and can be largely captured by a province’s
GDP measure. The results in Table 5 remain unchanged.

More mechanism tests

Columns (2)-(4) of Table 6 show the 2SLS results of the effects of outward
expansion on GDP by sector. Pushing the edge of urban residential development
outward has significant effects on the GDP of both the primary and secondary sectors,
which may benefit more from investment in local public infrastructure such as roads
and rails. Note that the effect on the secondary sector has a much larger magnitude than
that on the primary sector.31
Our theoretical model also implies that a city’s population size increases with the
degree of outward expansion. Data on the actual city population size calculated based
on current residents is not available on a yearly basis. Alternatively, in column (5), we
investigate the non-agricultural registered population from the Urban Statistical
Yearbooks instead. Consistent with our model, the effect of outward expansion on
population size is positive and significant. To address the concern that the registered
population size does not include non-registered migrants, in column (6), we further
report the results for the number of employed persons in urban private enterprises and
self-employed individuals, which may partially capture the migrant population. The
effect is also positive and significant.
We do not have a direct measure for local spending on public infrastructure.
Instead, we obtain data on local budgetary expenditures, a city’s public library
collections, number of beds in hospitals, and number of doctors, all from the Urban
Statistical Yearbooks. Local budgetary expenditures include public infrastructure and

31
A one-standard-deviation increase in the mean distance to the city center leads to an increase in the
primary-sector GDP of 3.2 billion yuan and an increase in the secondary-sector GDP of 52.3 billion yuan
in 2012 RMB, representing about 22% and 46% of their sample averages.
30
other public services, especially those related to local education and health care. While
the total local budgetary expenditure increases with the outward expansion as shown in
column (7) of Table 6, the increased public expenditure does not seem to go toward the
provision of public goods such as library collections or health care resources, which do
not directly contribute to GDP growth in the short term (see columns (8)-(10) of Table
6). This result provides indirect evidence that spending on local public infrastructure
increases with the city’s outward development.

6. Empirical Analysis: Career Concerns and Upward Land Development

In this section, we investigate the determinants of a city’s upward development,
i.e., the regulatory FAR limits, with special attention paid to the career-concern intensity
of city leaders. Taking advantage of our land-level data, we can control for the effects
of intra-city location on the FAR level for each land parcel. By doing so, we essentially
compare the FARs of two land parcels that share the same location attributes but were
sold during the terms of two city leaders with different career-concern intensities. The
baseline regression specification is given by
Fi , s ,c ,t   FAR s  city fixed effects  xi c  Wc ,t 1  t  ui , s ,c ,t , (17)

where Fi,s,c,t is the FAR level of land i sold during the term of city leader s in city c

in year t, and ui ,s,c,t is an error term. In (17), x represents the land parcel’s intra-city

location attributes, and the coefficient of x ,  c , is allowed to vary by city. As we did

in (15), we include a set of control variables. First, Wc,t 1 is a set of the city’s time-

variant economic variables that control for housing demand factors, including the city's
total registered population, GDP per capita, investment in fixed assets and investment
in real estate development in the year prior to the land sale, and the percentage change
of these city characteristics over the two years just before the land sale. In Wc,t 1 , we

also include the square root of the built-up land area at the end of the year immediately
preceding the land sale. In addition, t represents the fixed effects of land transaction
year, the fixed effects of land transaction quarter, and the province-specific linear trends
of land transaction year, which control for national and regional shocks. Again,
including the city fixed effects allows us to compare the FAR schemes of different city
leaders within the same city. We expect  FAR to be negative, for the equilibrium FAR

31
schedule becomes lower as the city is built more outwardly by a high-incentive city
leader.

Baseline regression results

In the baseline regression, x includes the distance to the city center, which is
regarded as the most important determinant of intra-city building density in the
literature (e.g., Fujita, 1989; DiPasquale and Wheaton, 1996; Duranton and Puga, 2015).
In order to show the FAR-distance gradient that has been well documented for
developed countries in the literature, we first restrain  c to be uniform across cities
and report the estimate in column (1) of Table 7, Panel A. The estimated coefficient on
the log of land distance to the city center suggests that a 100% increase in the distance
to the city center is associated with a decrease in FAR of 0.26. Note that the average
FAR is about 2. In all the columns of Table 7, we also report the estimated coefficients
of the city’s registered population size and GDP per capita, which are two typical
housing demand factors that can increase building density within a city. Consistent with
our expectation, the results show that these two city characteristics are positively
correlated with FAR.
Column (2) of Panel A in Table 7 reports the results when we let  c vary by city.

Consistent with the theory model’s prediction, the coefficient  FAR is negative and

significant. On average, a one-standard-deviation increase in s leads to a decrease in

the regulatory FAR of 0.04, which represents 12% of the standard deviation of FAR
across the terms of city leaders (0.34).
Our model also implies that when a city is governed by a high-incentive leader,
the FAR levels should be lowered to a larger extent for the land parcels located near the
city center than for those located at the city edge. To confirm this heterogeneous effect
of s by distance to the city center, we further include the interaction between s
and the distance to the city center and report the corresponding results in column (3).
As expected, for land parcels located near city centers, a one-standard-deviation
increase in s leads to a decrease in FAR of 0.08. This effect is attenuated with
increasing distance to the city center. 32 The four largest cities, Beijing, Tianjin,

32
To address the concern that low-career-concern city leaders may sell little land near the city fringe,
we trim the sample by dropping the land observations with a distance to the city center at or above the
32
Shanghai, and Chongqing, due to their special economic and political positions in China,
may have urban development patterns different from typical prefecture-level cities. To
address this concern, we re-run the regressions of columns (2) and (3) on a subsample
excluding these four big cities. Columns (4) and (5) show that the results remain
similar. 33 Again, because s is a generated regressor, the standard errors are
calculated on the basis of 1,000 bootstrap replications.

Buffer fixed effect results

Apart from distance to the city center, the neighborhoods of the transacted land
parcels may differ across city leaders in terms of other intra-city location attributes that
we do not observe (e.g., school quality and accessibility to public services such as
subways, parks, hospitals, etc.). These characteristics can affect FAR as well. To control
for the unobserved neighborhood characteristics, we adopt a buffer strategy to compare
the regulatory FARs of land parcels located in the same neighborhood but sold by
different city leaders. Specifically, we draw a buffer with a 1-km radius around each
land parcel in the sample and call each buffer the “neighborhood” of all land parcels
falling into the buffer. For regression analysis, we keep those buffers that contain at
least two land parcels (including the centroid land parcel). We have a total of 25,479
buffers that contain a total of 269,405 land parcels (“buffer sample”).34 We re-run the
regressions corresponding to specification (17) on the buffer sample and replace the
distance to the city center with the buffer dummies. Note that the city fixed effects are
dropped when the buffer dummies are included. Panel B of Table 7 reports the results,
which are very similar to the baseline results shown in Panel A.

Pre-trends of upward development

Similarly to the robustness check on the pre-trends of outward development
included in Section 5.2, we investigate the relationship between the career-concern
intensity of the current city leader and the FARs of the city’s land parcels sold within
the 3 years just before she took office to address the concern regarding her continuance

95th percentile of the city and re-run the regression. The result remains very similar.
33
Taking advantage of the large sample size of the land-level data, we also re-run the regression by
further dropping 15 deputy-provincial-level cities, and the result changes little.
34
On average, the 1-km buffer around a single land parcel contains about 10 land parcels. Accordingly,
a land parcel can fall into multiple buffers. On average, each land parcel belongs to 10 buffers. The
average distance to the centroid land parcel is about 200 meters.
33
of her predecessor’s spatial development pattern. Columns (4)-(6) of Table 4 report the
results of the land-level regressions and show that the FAR levels of the previously-sold
land parcels are positively correlated with the career-concern intensity of the current
city leader, and that the correlation is statistically insignificant, both of which are
contrary to what we found in Table 7.

Start age and start level

According to the construction of the ex ante promotion likelihood of city leaders,
the variations in their career concern intensities are mainly driven by the differences in
their initial ages and political hierarchy levels when they took office. Readers may
worry that our key results are driven by the way we predict the ex ante likelihood of
promotion. To address this concern, we re-run the regressions of outward development
and upward development respectively in specification equations (15) and (17) by
replacing the predicted career-concern intensity with the city leader’s start age and start
level. The results are reported in Table A4 (Panels A and B) in the appendix. Note that
the effect of the dummy indicating province-level or above is subsumed by the city
fixed effects. The estimates show a consistent story: younger city leaders tend to build
their cities more outwardly with a lower land-use intensity; other things being equal,
city leaders at the deputy-province level are more likely to adopt a less-outward pattern
of urban land development with a higher land-use intensity since they have lower ex
ante promotion chances and thus lower career incentives.

7. Spatial Pattern of Development of Other Land-Use Types

In the previous sections, we focus on residential land development. By doing so,
we implicitly assume that the production takes no land. This section presents an overall
picture of the spatial pattern of development of other land use types in addition to
residential land. For the 202 cities in our sample, we collect the transaction information
for 12,718 purely commercial land parcels and 42,096 industrial land parcels. All of
these land transactions are conducted through public auction. As shown in Table A3 in
the appendix, while the total area of industrial land is comparable to that of residential
land, the former contributes only 11% of the total land sale revenues of all three land
use types.
The summary statistics in Table A3 show that industrial land development is
generally located farthest away from the city center with an average distance of 27 km.
34
The average distance to the city center of purely commercial land is 19.6 km, while that
of residential land is 21.4 km. This is consistent with classic urban models. As
commercial activities benefit the most from agglomeration economy, business
developers are willing to bid the highest price for the land located near the city center.
Industrial production usually has the greatest need for space and therefore tends to be
located in the suburbs. Residential land, on average, lies in between.35
Table 8 Panel A shows that the positive relationship between city leaders’ career
concern intensities and outward development also holds for industrial land. This is
because the outward expansion of residential developments brings more land sale
revenues to the city treasury and leads to more public infrastructure being built at the
city edge (e.g., paved roads, highways, sewage systems, the supply of utilities), which
may enhance the industrial development nearby (Baum-Snow, Brandt, Henderson,
Turner, Zhang, 2016). Moreover, as Zheng, Sun, Wu, Kahn (2015) shows, industrial
development will generate demand for housing and commercial services at the city edge,
which may in turn raise housing prices and land values. Such complementarity works
to boost a city’s GDP growth.36
On the other hand, as Table 8 Panel B shows, there is no significant relationship
between the outward development of purely commercial land and city leaders’ career
incentives. Note that the sample of purely commercial land consists mainly of land
parcels for office buildings that are occupied by headquarters, banks and various
financial service firms. Compared with other production activities, these businesses
enjoy more agglomeration economy and tend to benefit more from being located near
the city center. Therefore their bids for land decline sharply as the land gets farther away
from the center.

8. Conclusion
Land development is essential for urbanization because it will profoundly shape
the urban internal structure and influence the spatial distribution of economic activities
within a city. This impact may even last for decades to come in the future (Glaeser,

35
We also run a linear regression of the distance to the city center on a dummy indicating residential
land and a dummy indicating industrial land, controlling for the city fixed effects. The results show that,
on average, the residential and industrial land parcels are located farther away from the city centers than
the purely commercial land parcels by 2.60 km and 6.23 km, respectively.
36
The residential land type in this paper includes both purely residential and residential mixed with
commercial.
35
2011). This is especially relevant for the fast-growing cities in developing countries.
Land development is subject to government regulations in most cities all over the world
(Duranon and Puga, 2015; Gyourko and Molloy, 2015). It is thus very important to
understand the key factors influencing the patterns of urban land development and
regulations.
This paper examines the determinants of urban land development policies by
focusing on the decisions of city leaders who behave like city developers. China’s
institutional background offers a unique setting for studying the role of developers in
influencing urban land development regulations as homeowners have relatively weak
power to affect local land policies in urban China.
We highlight that the career incentives of city leaders play an important role in
driving the unique pattern of land development in Chinese cities. Unlike the
homeowners in the developed economies that are prone to under-expansion of cities,
promotion-motivated city leaders in China tend to over-expand cities spatially
(compared to benchmark market equilibrium), which in turn results in low-intensity
land use. Exploiting a large micro-level land dataset that features rich variations across
cities and over time, we empirically test the predictions of our theory and find strong
supportive evidence.
An intriguing question arising out of our study is the welfare implication of the
spatial pattern of urbanization driven by city leaders as city developers. We find that
career-concern-driven urban expansion creates opposing welfare impacts. On the one
hand, the total factor productivity is increased when outward expansion creates more
land sale revenues and thus channels more money into public infrastructure investment.
Further, higher productivity accompanied by outward expansion can attract more
people and foster urban agglomeration. On the other hand, commuting costs increase
when the spatial size of a city expands (Bertaud and Brueckner, 2005). Meanwhile, a
more widely spread spatial distribution of population and economic activities may hurt
agglomeration. A recent study shows that as large numbers of firms cluster in the
industrial parks at the city edge, they can enjoy localization economy benefits and the
workers can also live in nearby residential developments (Zheng et al., 2015). This
implies that the downsides of outward expansion we just described may be to some
extent mitigated by the creation of industrial parks at the edges of many cities. Since
we do not know the exact magnitudes of these offsetting effects of outward expansion,
it is hard for us to draw any conclusions on the welfare implications of career-concerns-
36
driven urban expansion. Even with this limitation, our paper makes its contribution by
offering a positive analysis of China’s urban land development policies, thus laying a
foundation for a comprehensive welfare analysis in the future.
This paper also opens venues for future research in several other directions. For
instance, it is worth further studying how the pattern of urban land development
influences the spatial distribution of economic activities within cities. Baum-Snow et
al. (2016) find that road infrastructure facilitates the decentralization of population and
manufacturing production in Chinese cities. It would be interesting to examine how
urban land development directed by city governments affects the provision of public
infrastructure, and how this in turn shapes the internal urban structure in terms of the
locations of firms, employment and population. It would also be worth investigating
how residential land developments complement industrial land developments in
boosting local GDP and enhancing the formation of edge cities in suburban areas.

References

Bartik, T., 1991. “Who Benefits from State and Local Economic Development Policies?”
W.E. Upjohn Institute for Employment Research, Kalamazoo, MI.
Baum-Snow, N., Brandt, L., Henderson, J.V., Turner, M.A., Zhang, Q., 2016. “Roads,
Railroads and Decentralization of Chinese Cities,” Review of Economics and Statistics,
forthcoming.
Bertaud, A., Brueckner, J.K., 2005. “Analyzing Building-Height Restrictions: Predicted
Impacts and Welfare Costs.” Regional Science and Urban Economics, 35, 109-125.
Besley, T., Persson, T, Sturm, D., 2010. “Political Competition, Policy and Growth:
Theory and Evidence from the United States,” Review of Economic Studies, 77(4),
1329-1352.
Brueckner, J.K., Fansler, D.A., 1983. “The Economies of Urban Sprawl: Theory and
Evidence on the Spatial Sizes of Cities,” Review of Economics and Statistics, 65, 479-
482.
Brueckner, J.K., Fu, S., Gu, Y., Zhang, J., 2016. “Measuring the Stringency of Land-
Use Regulation and Its Determinants: The Case of China’s Building-Height Limits.”
Review of Economics and Statistics, forthcoming.
Burchfield, M., Overman, H.G., Puga, D., Turner, M.A., 2006. “Causes of Sprawl: A
Portrait from Space.” The Quarterly Journal of Economics, 121(2), 587-633.
Cai, H., Wang, Z., Zhang, Q., 2016. “To Build Above the Limit? Implementation of
Land Use Regulations in Urban China,” Journal of Urban Economics, forthcoming.
Calabrese, S., Epple, D., Romano, R., 2007. “On the Political Economy of Zoning,”
Journal of Public Economics, 91, 25-49.

37
Chen, Y., Henderson, J. V., Cai, W., 2016. “Political Favoritism in China’s Capital
Markets and Its Effect on City Sizes,” Journal of Urban Economics, forthcoming.
Chen, Y., Li, H.B., Zhou, L.A., 2005. "Relative Performance Evaluation and the
Turnover of Provincial Leaders in China," Economics Letters, 88(3), 421-425.
Deng, X., Huang, J., Rozelle, S., and Uchida, E., 2008. “Growth, Population and
Industrialization, and Urban Land Expansion of China,” Journal of Urban Economics,
63, 96-115.
Deng, X., Huang, J., Rozelle, S., and Uchida, E., 2010. “Economic Growth and the
Expansion of Urban Land in China,” Urban Studies, 47(4), 813-843.
Ding, C., 2003. “Land Policy Reform in China: Assessment and Prospects,” Land Use
Policy, 20, 109-120
DisPasquale, D., Wheaton, W.C., 1996. Urban Economics and Real Estate Markets.
New Jersey: Prentice Hall.
Duranton, G., Puga, D., 2015. “Urban Land Use,” in Gilles Duranton, J. Vernon
Henderson, and William S. Strange (ed.): Handbook of Regional Science and Urban
Economics, 5, Amsterdam: North-Holland, 467-560.
Epple, D., Romer, T., Filimon, R., 1988. “Community Development with Endogenous
Land Use Controls,” Journal of Public Economics, 35, 133-162.
Feng, J., Lichtenberg, E., Ding, C., 2015. “Balancing Act: Economic Incentives,
Administrative Restrictions, and Urban Land Expansion in China,” China Economic
Review, 36, 184-197.
Fischel, W.A., 2001. The Homevoter Hypothesis, Cambridge, MA: Harvard University
Press.
Fischel, W.A., 2008. “Political Structure and Exclusionary Zoning: Are Small Suburbs
the Big Problem?” Fiscal Decentralization and Land Policies, Gregory K. Ingram and
Yu-Hung Hong, eds. Cambridge, MA: Lincoln Institute of Land Policy.
Fujita, M., 1989. Urban Economic Theory: Land Use and City Size. Cambridge:
Cambridge University Press.
Glaeser, E.L., Gyourko, J., Saks, R.E., 2005. “Why Have Housing Prices Gone Up?”
American Economic Review, 95(2), 329-333.
Glaeser, E.L., 2011. Triumph of the City. New York: Penguin Press.
Gong, L., Knight, B., Xiao, J., Zhang, Q., 2016. “Political Budget Cycles and Incentive
from Chinese Prefectures.” Peking University Working Paper.
Guo, X., 2001. “Land Expropriation and Rural Conflicts in China,” The China
Quarterly, 166, 422-439.
Gyourko, J., Molloy, R., 2015. “Regulation and Housing Supply,” in E.S., Mills (ed.):
Handbook of Regional Science and Urban Economics, 5, Elsevier.
Han, L., Kung, J. K., 2015. “Fiscal incentives and policy choices of local governments:
Evidence from China,” Journal of Development Economics, 116, 89-104.
Henderson, J.V., Venables, T., 2009. “The Dynamics of City Formation,” Review of
38
Economics Dynamics, 12, 233-254.
Hilber, C.A.L., Robert-Nicoud, F., 2013. “On the Origins of Land Use Regulations:
Theory and Evidence from US Metro Areas,” Journal of Urban Economics, 75, 29-43.
Jia, R., Kudamatsu, M., Seim, D., 2015. “Political Selection in China: The
Complementary Roles of Connections and Performance,” Journal of the European
Economic Association, 13(4), 631-668.
Kahn, M., Li, P., Zhao, D., 2015. “Pollution Control Effort at China’s River Borders:
When Does Free Riding Cease?” American Economic Journal: Economic Policy,
forthcoming.
Li, H.B., Zhou, L.A., 2005. “Political Turnover and Economic Performance: The
Incentive Role of Personnel Control in China,” Journal of Public Economics, 89, 1743-
1762.
Lichtenberg, E., Ding, C., 2009. “Local Officials as Land Developers: Urban Spatial
Expansion in China,” Journal of Urban Economics, 66, 57-64.
Lichtenberg, E., Ding, C., 2008. “Assessing Farm land Protection Policy in China,”
Land Use Policy, 25, 59-68.
Liu, X., Chen, Z., Zhao, Y., 2012. “The Procedural Fairness Prior to the Monetary
Compensation: The Determinants of the Degree of Farmers’ Satisfaction at Levying
their Land,” Management World (in Chinese), 2, 44-51.
Logan, J., Molotch, H., 1987. Urban Fortunes: The Political Economy of Place. Los
Angeles: University of California Press.
McGrath, D.T., 2005. “More Evidence on the Spatial Scale of Cities,” Journal of Urban
Economics, 58, 1-10.
Molotch, H., 1976. “The City as a Growth Machine: Towards a Political Economy of
Place,” American Journal of Sociology, 82(2), 309-332.
Ortalo-Magne, F., Prat, A., 2014. “On the Political Economy of Urban Growth:
Homeownership versus Affordability,” American Economic Journal: Microeconomics,
6 (1), 154-181.
Sole-Olle, A., Viladecans-Marsal, E., 2012. “Lobbying, Political Competition, and
Local Land Supply: Recent Evidence from Spain,” Journal of Public Economics, 96,
10-19.
Wu, J., Deng, Y., Huang, J., Morck, R., Yeung, B., 2013. “Incentives and Outcomes:
China's Environmental Policy,” NBER Working Paper # 18754.
Xie, Z., 2015. Administration upon Spring: Institutional Analysis of the Land
Administration in China (in Chinese). The Commercial Press.
Xu, C., 2011. “The Fundamental Institutions of China’s Reforms and Development,”
Journal of Economic Literature, 49, 1076–1151.
Yao, Y., Zhang, M. 2015, “Subnational Leaders and Economic Growth: Evidence from
Chinese Cities,” Journal of Economic Growth, 20, 405–436.
Zhao, Y., Webster, C., 2011. “Land Dispossession and Enrichment in China’s Suburban

39
Villages.” Urban Studies, 48(3), 529-551.
Zhang, 2007. “The Implementation and Protection of Land Property Right Under the
Background of Urbanization),” Management World (in Chinese), 12, 31-47.
Zhang, J., Gao, Y., 2007. “Term Limits and Rotation of Chinese Governors: Do They
Matter to Economic Growth?” Economic Research Journal (in Chinese), 11, 91-103.
Zheng, S., Sun, W., Wu, J., Kahn, M., 2015. “The Birth of Edge Cities in China:
Measuring the Spillover Effects of Industrial Parks,” NBER Working Paper # 21378.
Zheng, S, Cao, J., Kahn, M.E., Sun, C., 2014. “Real Estate Valuation and Cross-
Boundary Air Pollution Externalities: Evidence from Chinese Cities,” Journal of Real
Estate Finance and Economics, 48, 398-414.
Zheng, S, Kahn, M., Sun, W., Luo, D., 2014. “Incentives for China’s Urban Mayors to
Mitigate Pollution Externalities: The Role of the Central Government and Public
Environmentalism,” Regional Science and Urban Economics, 47, 61-71.
Zhu, J., 2004. “From Land Use Right to Land Development Right: Institutional Change
in China’s Urban Development,” Urban Studies, 41(7), 1249-1267.

40
 Table 1: Summary statistics
Coastal Central Northeast Southwest Northwest
GDP per capita in 2012 (10,000 yuan per person) 6.48 4.44 5.24 3.06 4.35
Percentage of land sales (%) 56.66 19.29 10.40 10.55 3.09
Panel A: Land parcel's distance to CBD, km
Nation Coastal Central Northeast Southwest Northwest
2000-2008 15.54 18.42 7.90 10.87 14.64 7.91
<35.71> <39.77> <15.58> <25.49> <31.70> <11.33>
(15.96) (17.45) (7.83) (11.06) (14.77) (9.05)
[8,801] [5,444] [1,316] [578] [1,215] [248]

2009-2010 22.84 26.14 15.33 18.38 29.47 9.59

<54.76> <61.44> <35.54> <48.66> <74.83> <14.63>
(26.38) (27.76) (17.03) (21.02) (36.21) (13.87)
[12,417] [7,124] [2,493] [1,252] [1,075] [473]

2011-2012 25.33 28.25 16.47 21.53 39.59 10.94

<63.21> <65.31> <37.17> <54.18> <79.45> <18.15>
(28.54) (26.43) (19.47) (26.69) (46.69) (15.75)
[8,799] [4,440] [1,982] [1,293] [877] [207]
Panel B: Regulatory FAR level
Nation Coastal Central Northeast Southwest Northwest
2000-2008 2.22 1.93 2.65 2.11 2.98 3.01
(1.20) (0.99) (1.24) (1.15) (1.41) (1.65)
[8,801] [5,444] [1,316] [578] [1,215] [248]

2009-2010 2.41 2.19 2.63 2.48 2.92 3.17

(1.13) (0.99) (1.19) (1.18) (1.28) (1.40)
[12,417] [7,124] [2,493] [1,252] [1,075] [473]

2011-2012 2.51 2.27 2.73 2.55 2.94 3.37

(1.16) (0.99) (1.17) (1.33) (1.34) (1.20)
[8,799] [4,440] [1,982] [1,293] [877] [207]
Notes: Standard deviations are in parentheses. The 90th percentiles are in angle brackets. Numbers of
observations are in brackets. We divide China into 5 regions according to the geographical divisions of
the country, i.e., coastal, central, northeast, southwest, and northwest. The provinces in the coastal region
are Shandong, Jiangsu, Anhui, Zhejiang, Fujian, Shanghai, Beijing, Tianjin, Guangdong, Guangxi, and
Hainan. The provinces in the central region are Hubei, Hunan, Henan, Jiangxi, Hebei, Shanxi, and
Neimenggu. The provinces in the northeast region are Liaoning, Jilin, and Heilongjiang. The provinces
in the southwest region are Sichuan, Yunnan, Guizhou, and Chongqing. The provinces in the northwest
region are Ningxia, Xinjiang, Qinghai, Shaanxi, and Gansu.

41
 Table 1: Summary Statistics (cont’d)
Panel C: Start age and promotion status by start level
Start level
Deputy-
Prefecture Province Politburo
province
Start age 49.54 50.82 57.00 57.44
(3.76) (3.80) (4.64) (4.67)
[798] [157] [10] [9]
Percentage of being promoted by 0.40 0.06 0.71 0.67
the end of term [610] [112] [7] [6]
Panel D: Percentage of being promoted by the end of term, by start level and by start age
Start level
Deputy-
Prefecture Province Politburo
province
Start age intervals
38-47 0.59 0
[192] [22] [0] [0]
48-50 0.45 0.14 1
[174] [29] [1] [0]
51-52 0.26 0.07 0
[126] [15] [1] [0]
53-65 0.14 0.04 0.80 0.67
[118] [46] [5] [6]
Notes: Standard deviations are in parentheses. Numbers of observations are in brackets.

42
 Table 2: Start age, start level, and promotion
Dummy: Promotion of City Leader
LPM LPM LPM Probit Probit Probit
(1) (2) (3) (4) (5) (6)
Dummy: Deputy-province -2.691*** -2.650*** -2.752*** -7.937*** -7.906*** -8.414***
(0.263) (0.266) (0.274) (1.398) (1.396) (1.492)
Dummy: Province or above -0.570 -0.506 -0.667 -1.549 -1.299 -1.060
(0.758) (0.759) (0.802) (2.386) (2.386) (2.949)
Start age -0.048*** -0.046*** -0.046*** -0.136*** -0.130*** -0.130***
(0.005) (0.005) (0.005) (0.015) (0.015) (0.015)
Start age* Dummy: Deputy-province 0.048*** 0.047*** 0.049*** 0.136*** 0.135*** 0.144***
(0.005) (0.005) (0.005) (0.027) (0.028) (0.029)
Start age*Dummy: Province or above 0.022 0.020 0.022 0.060 0.055 0.047
(0.013) (0.013) (0.014) (0.042) (0.042) (0.050)
Dummy: Working in birth province -0.004 -0.012
(0.035) (0.113)
Dummy: Having center experience 0.152** 0.587**
(0.075) (0.267)
Constant 2.750*** 2.686*** 2.674*** 6.370*** 6.121*** 6.092***
(0.229) (0.232) (0.233) (0.751) (0.753) (0.756)
Observations 735 697 697 735 697 697
R-squared 0.194 0.189 0.192
Adjusted R-squared 0.190 0.180 0.180
Pseudo R-squared 0.162 0.158 0.160
Notes: *** p<0.01, ** p<0.05, * p<0.1 Heteroskedasticity-robust standard errors clustered by city are in
parentheses.

43
 Table 3 Panel A: Outward development and career concerns of city leaders
Full sample Drop four province-level cities
90th pct. 85th pct. Mean 90th pct. 85th pct. Mean
[26.9] [23.4] [13.5] [26.3] [22.8] [13.1]
(1) (2) (3) (4) (5) (6)
City leader's career-concern intensity 51.057** 32.796* 5.229 50.724** 32.963* 5.381
(24.794) (19.018) (9.567) (24.834) (19.089) (9.357)

City fixed effects Yes Yes Yes Yes Yes Yes

On-year & off-year dummies Yes Yes Yes Yes Yes Yes
Provincial linear year trends (on & off years) Yes Yes Yes Yes Yes Yes
Sqrt. of initial built-up land area Yes Yes Yes Yes Yes Yes
Initial city time-variant variables Yes Yes Yes Yes Yes Yes
Observations 377 377 377 366 366 366
R-squared 0.972 0.974 0.981 0.9685 0.9707 0.9753
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are calculated on the basis of 1,000 bootstrap replications. Each regression is weighted by the number
of land parcels sold during the city leader's term. Each regression controls for the city fixed effects, dummies of the year when the city leader took office (on years), dummies
of the year when the city leader left office (off years), province-specific linear trends (for both on & off years), square root of built-up land area at the end of the previous year
before the city leader took office, city's total registered population, GDP per capita, investment in fixed asset, investment in real estate development in the previous year before
the city leader took office, and the % change of total registered population, GDP per capita, investment in fixed asset, investment in real estate development over the two years
before the city leader took office. The means of the distance-based outward measures are in brackets, and the unit is km.

44
 Table 3 Panel B: Outward development and career concerns of city leaders, additional controls
90th pct. 85th pct. Mean 90th pct. 85th pct. Mean 90th pct. 85th pct. Mean
(1) (2) (3) (4) (5) (6) (7) (8) (9)
City leader's career-concern intensity 51.057** 32.668* 5.220 52.913** 35.323* 6.924 52.028** 32.871* 5.605
(25.105) (19.367) (9.917) (25.284) (20.063) (9.413) (24.198) (18.712) (9.638)
City leader's total tenure length -0.001 -0.158 -0.011
(0.501) (0.449) (0.206)
Dummy: City leader had center experience 15.294 16.464 11.066
(15.129) (13.756) (6.744)
Provincial leader's age 1.173 0.902 0.300
(1.105) (0.941) (0.468)
City fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
On-year & off-year dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes
Provincial linear year trends (on & off years) Yes Yes Yes Yes Yes Yes Yes Yes Yes
Sqrt. of initial built-up land area Yes Yes Yes Yes Yes Yes Yes Yes Yes
Initial city time-variant variables Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 377 377 377 375 375 375 368 368 368
R-squared 0.972 0.975 0.981 0.973 0.976 0.983 0.973 0.975 0.981
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are calculated on the basis of 1,000 bootstrap replications. Each regression is weighted by the number
of land parcels sold during the city leader's term. Col (1)-(3) further control for the total tenure length of the city leader (measured in months). Col (4)-(6) additionally include
a dummy indicating if the city leader used to work in the central government. Col (7)-(9) further include the age of the provincial leader as of the year when the city leader took
office. Col (10)-(12) additionally controls for an industry shift-share type variable for city growth during a 3-year period after each city leader took office.

45
 Table 4: Pre-trends of the spatial pattern of urban development and the career concerns of current city leaders, residential land sold
during the most recent 3 years before each city leader took office
Pre-trends of outward development Pre-trends of upward development

Exclude 4
Sampling distribution of distance to CBD,
Full sample province- Buffer sample
km
level cities

90th pct. 85th pct. Mean FAR FAR FAR

(1) (2) (3) (4) (5) (6)
City leader's career-concern intensity -7.798 -2.393 -13.84 0.248 0.248 -0.005
(97.9229) (128.6126) (51.3596) (0.1532) (0.1558) (0.0653)

Observations 183 183 183 14,087 11,583 86,847

R-squared 0.9992 0.9987 0.9996 0.3718 0.3035 0.5818
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are calculated on the basis of 1,000 bootstrap replications. The regressions in col (1)-col (3) are
weighted by the number of land parcels sold during the three years before the city leader took office. In the regressions of (1)-(3), we control for the city fixed effects, dummies
of the year when the city leader took office, dummies of the year when the city leader left office, province-specific linear trends (for both on & off years), square root of built-
up land area at the end of the previous year before the city leader took office. Because of the limited sample size, we are unable to further control for the city's total registered
population, GDP per capita, investment in fixed asset, investment in real estate development in the year before the city leader took office, and the % change of total registered
population, GDP per capita, investment in fixed asset, investment in real estate development over the two years before the city leader took office as in Table 3. In the regressions
in (4) and (5), we control for the city fixed effects, log of (land's distance to CBD+1), and the interactions of city fixed effects and log of (land's distance to CBD+1), the fixed
effects of the land transaction year, the fixed effects of the land transaction quarter, and province-specific linear transaction year trends. Each regression also controls for city's
total registered population, GDP per capita, investment in fixed asset, investment in real estate development in the year before the land sale, the % change of total registered
population, GDP per capita, investment in fixed asset, and investment in real estate development over the two years before the land sale, and the square root of built-up land
area at the end of the year before the land sale. The regression in (6) uses the buffer sample created from the 14,097 land parcels sold during the 3 years before the 183 city
leaders took office following the algorithm stated in Section 6. Col (6) controls for the buffer fixed effects, the fixed effects of the land transaction year, the fixed effects of the
land transaction quarter, province-specific linear transaction year trends, and the same set of city time-variant controls as included in (4) and (5).

46
 Table 5: Land sale revenue, GDP, and outward development
Mean
distance to
Land conveyance
Gross Domestic Product, CBD in
revenue, 100 million
100 million (2012 RMB) each
(2012 RMB)
city/year,
km
OLS 2SLS OLS 2SLS First stage
(1) (2) (3) (4) (5)
Mean distance to CBD in each city/year, km 4.777*** 8.652** 10.2891 37.148*
(1.734) (4.320) (11.0321) (20.310)
Excluded instrument:
A younger provincial secretary being successor * Absolute value of start age gap -1.166***
(0.308)
City fixed effects Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes
City initial characteristics Yes Yes Yes Yes Yes
Observations 821 821 868 868 869
R-squared 0.846 0.836 0.9731 0.970 0.815
First stage F 13.9 14.3 14.3
Notes: *** p<0.01, ** p<0.05, * p<0.1 Each regression is weighted by the number of land sales in each city/year. Heteroskedasticity-robust standard errors in parentheses are
clustered by city level. Each regression controls for the city fixed effects, year fixed effects, city's registered population and GDP per capita in year (t-1), % change of registered
population and GDP per capita over the past 2 years before (t-1), the square root of built-up land area at the end of year (t-1), and log (GDP of the province that contains the
city). For regressions in col (2) and (4), the excluded instrumental variable is the interaction term between the dummy indicating if in year t city c is led by a provincial secretary
who has a smaller start age than his immediate predecessor, and the absolute value of the start age gap between current provincial secretary and his immediate predecessor. The
corresponding first-stage results are reported in col (5).

47
 Table 6: Outward development and urban growth
GDP (100 GDP in GDP in GDP in Non- Number of Local Percentage Percentage Percentage
million primary secondary tertiary agricultural employed budgetary growth in growth in growth in
yuan, sector sector sector registered persons in expenditure, public number of number of
2012 (100 (100 (100 population urban 100 million library beds in doctors
RMB) million million million (10,000 private yuan collections hospital
yuan, yuan, yuan, persons) enterprises
2012 2012 2012 and self-
RMB) RMB) RMB) employed
individuals
(10,000
persons)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Mean distance to CBD in each city/year, km 37.148* 2.650*** 43.572*** 4.541 31.394*** 4.525*** 8.697** -0.016 -0.013 -0.017
(20.310) (0.677) (13.779) (9.628) (7.902) (1.556) (3.899) (0.027) (0.026) (0.023)
Observations 868 868 868 868 868 854 869 670 670 670
R-squared 0.970 0.984 0.952 0.955 0.837 0.751 0.923 0.092 0.071 0.094
First stage F 14.3 14.3 14.3 14.3 14.3 14.4 14.3 12.6 12.6 12.6
Notes: *** p<0.01, ** p<0.05, * p<0.1 Each regression is weighted by the number of land sale in each city/year. Heteroskedasticity-robust standard errors in parentheses are
clustered by city level. Each regression controls for the city fixed effects, year fixed effects, city's registered population and GDP per capita in year (t-1), % change of registered
population and GDP per capita over the past 2 years before (t-1), the square root of built-up land area at the end of year (t-1), and and log (GDP of the province that contains
the city). The excluded instrumental variable is the interaction of the dummy indicating if in year t city c is led by a provincial secretary who has a smaller start age than his
immediate predecessor and the absolute value of the start age gap between current provincial secretary and his immediate predecessor.

48
 Table 7 Panel A: FAR and career concerns of city leaders, baseline regression

Regulatory FAR upper limit

Exclude 4 provincial-
Full sample
level cities
(1) (2) (3) (4) (5)
City leader's career-concern intensity -0.235*** -0.189*** -0.366*** -0.251*** -0.547***
(0.072) (0.071) (0.072) (0.070) (0.081)
City leader's career-concern intensity * land distance to the city center (km) 0.011*** 0.018***
(0.001) (0.002)
Log (land distance to the city center+1) -0.257***
(0.008)
City registered population size (10,000 persons) 0.003*** 0.002*** 0.002*** 0.002*** 0.002***
(0.001) (0.001) (0.001) (0.001) (0.001)
GDP per capita (10,000 yuan per person) 0.021** 0.007 0.010 0.002 0.005
(0.010) (0.010) (0.009) (0.009) (0.010)
City fixed effects Yes Yes Yes Yes Yes
Log(land distance to the city center+1)*city dummies Yes Yes Yes Yes
Transaction year fixed effects Yes Yes Yes Yes Yes
Transaction quarter fixed effects Yes Yes Yes Yes Yes
Province-specific linear transaction year trends Yes Yes Yes Yes Yes
Other initial city controls Yes Yes Yes Yes Yes
Observations 29,957 29,957 29,957 26,567 26,567
R-squared 0.280 0.321 0.323 0.306 0.308
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are calculated on the basis of 1,000 bootstrap replications. Each regressions in columns 2-5 control for the city fixed effects,
log of (land's distance to CBD+1), and the interactions of city fixed effects and log of (land's distance to CBD+1), transaction year fixed effects, transaction quarter fixed effects, and province-
specific linear transaction year trends. Each regression controls for the square root of built-up land area at the end of the year before the land sale. Each regression also controls for city's total
registered population, GDP per capita, investment in fixed asset, investment in real estate development in the year before the land sale, and the % change of total registered population, GDP per
capita, investment in fixed asset, and investment in real estate development over the two years before the land sale.

49
 Table 7 Panel B: FAR and career concerns of city leaders, buffer regression
Regulatory FAR upper limit
Exclude 4 provincial-level
Buffer sample
cities

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

City leader's career-concern intensity -0.148*** -0.412*** -0.217*** -0.474***
(0.029) (0.042) (0.029) (0.037)
City leader's career-concern intensity * buffer centroid's distance to the city center (km) 0.019*** 0.022***
(0.002) (0.002)
City registered population size (10,000 persons) 0.001*** 0.002*** 0.002*** 0.002***
(0.000) (0.000) (0.000) (0.000)
GDP per capita (10,000 yuan per person) 0.028*** 0.030*** 0.033*** 0.041***
(0.004) (0.004) (0.005) (0.004)
Buffer fixed effects Yes Yes Yes Yes
Transaction year fixed effects Yes Yes Yes Yes
Transaction quarter fixed effects Yes Yes Yes Yes
Province-specific linear transaction year trends Yes Yes Yes Yes
Other initial city controls Yes Yes Yes Yes
Observations 269,221 269,221 240,529 240,529
R-squared 0.513 0.514 0.495 0.495
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are calculated on the basis of 1,000 bootstrap replications. Each regression controls for the buffer fixed
effects, transaction year fixed effects, transaction quarter fixed effects, and province-specific linear transaction year trends. Each regression controls for the square root of built-
up land area at the end of the year before the land sale. Each regression also controls for city's total registered population, GDP per capita, investment in fixed asset, investment
in real estate development in the year before the land sale, and the % change of total registered population, GDP per capita, investment in fixed asset, and investment in real
estate development over the two years before the land sale.

50
 Table 8 Panel A: Outward development and career concerns of city leaders, industrial land

Distribution of distance to CBD, industrial land sold within each city leader's term, km
Mean 90th pct. 80th pct. 70th pct. 60th pct. 50th pct. 40th pct. 30th pct. 20th pct. 10th pct.
[16.5] [29.0] [24.1] [20.9] [17.4] [15.1] [12.8] [10.5] [8.5] [6.4]
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
City leader's career-concern intensity 16.586* 28.090 32.472 30.802* 20.690* 18.511* 9.781 7.509 4.869 2.408
(8.469) (22.331) (23.883) (17.501) (11.316) (10.454) (8.425) (7.135) (6.787) (6.342)
City fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
On-year & off-year dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Provincial linear year trends (on & off years) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Sqrt. of initial built-up land area Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Initial city time-variant variables Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 365 365 365 365 365 365 365 365 365 365
R-squared 0.990 0.975 0.966 0.974 0.985 0.985 0.985 0.984 0.981 0.973
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are calculated on the basis of 1,000 bootstrap replications. Each regression is weighted by the number
of land parcels sold during the city leader's term. Each regression controls for the city fixed effects, the dummies of the year when the city leader took office, the dummies of
the year when the city leader left office, the province-specific linear trends (for both on & off years), the square root of built-up land area at the end of the year before the city
leader took office, city's total registered population, GDP per capita, investment in fixed asset, investment in real estate development in the year before the city leader took
office, and the % change of total registered population, GDP per capita, investment in fixed asset, investment in real estate development over the two years before the city leader
took office. The means of each distance-based outward measures are in brackets, and the unit is km.

51
 Table 8 Panel B: Outward development and career concerns of city leaders, purely commercial land

Distribution of distance to CBD, purely commercial land sold within each city leader's term, km
Mean 90th pct. 80th pct. 70th pct. 60th pct. 50th pct. 40th pct. 30th pct. 20th pct. 10th pct.
[12.9] [25.7] [19.5] [15.8] [13.4] [11.0] [8.9] [7.1] [5.6] [4.3]
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
City leader's career-concern intensity 3.943 10.956 12.501 7.629 0.266 -0.816 1.381 0.544 0.848 -0.351
(12.843) (30.023) (28.921) (24.706) (16.887) (12.493) (9.718) (6.037) (5.086) (4.083)
City fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
On-year & off-year dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Provincial linear year trends (on & off years) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Sqrt. of initial built-up land area Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Initial city time-variant variables Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 358 358 358 358 358 358 358 358 358 358
R-squared 0.964 0.960 0.939 0.922 0.945 0.952 0.958 0.973 0.958 0.957
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are calculated on the basis of 1,000 bootstrap replications. Each regression is weighted by the number
of land parcels sold during the city leader's term. Each regression controls for the city fixed effects, the dummies of the year when the city leader took office, the dummies of
the year when the city leader left office, the province-specific linear trends (for both on & off years), the square root of built-up land area at the end of the year before the city
leader took office, city's total registered population, GDP per capita, investment in fixed asset, investment in real estate development in the year before the city leader took
office, and the % change of total registered population, GDP per capita, investment in fixed asset, investment in real estate development over the two years before the city leader
took office. The means of each distance-based outward measures are in brackets, and the unit is km.

52
 Figure 1: Outward development and population growth

Pct.90 distance to CBD/sqrt prefectural land area

Hangzhou Dongguan

1
Xuzhou
Nanjing

Dalian
.8
Qingdao
Lanzhou Shenzhen
Ningbo
Xiamen
Tianjin Guangzhou
Zibo
.6

WuxiShanghai
Jieyang
Luoyang Zhongshan
Suzhou
Chengdu Beijing
Shantou
Zhengzhou
.4

Wuhan
Luzhou
Chongqing Xi'an
Taizhou
Jinan
Wenzhou
Changsha
Yueyang Fuzhou
Shenyang
.2

Kunming
Hefei
Guiyang
Nanchang Taiyuan
Weifang
Nanning
JilinHa'erbin
Fuyang Changchun
Shijiazhuang
Linyi
NanyangChaozhou
Zunyi
0

0 .5 1 1.5 2
Pop growth, 1990-2010
Note: The size of each buffer indicates the number of land parcels
The panel includes the top 50 cities in terms of urban pop size in 2010

Notes: The variable on y-axis is the ratio of the 90th percentile of land’s distance to the city center of all
residential land sold in each city during 2000-2012 to the square root of the city’s land area.

Figure 2: Regulatory FAR upper limits and population growth

4

Xi'an
Nanyang Guiyang
Lanzhou
3.5

Zunyi Jieyang
Luoyang
Nanning Hefei
Average regulatory FAR

Yueyang
Changsha Shantou
Shijiazhuang Taiyuan
Chengdu
3

Jilin
Chongqing
Fuyang Wuhan
Kunming
Chaozhou
WeifangFuzhou Guangzhou
Ha'erbin Zhengzhou Xiamen
2.5

Shenzhen
XuzhouTaizhou
Jinan Zhongshan Dongguan
Nanchang
LinyiShenyang
Wenzhou
Dalian
LuzhouChangchun Beijing
Hangzhou
2

Wuxi
QingdaoNingbo
Tianjin
Zibo Nanjing Suzhou
1.5

Shanghai

0 .5 1 1.5 2
Pop growth, 1990-2010
Note: The size of each buffer indicates the number of land parcels
The panel includes the top 50 cities in terms of urban pop size in 2010

53
Figure 3:Urban outward expansion versus city leader’s career-concern intensity
(Starred curve is the benchmark equilibrium urban boundary)
Extra Payoffs from Expanding Beyond the Boundary Constraint
200

100

-100
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Urban Outward Expansion

170

160

150
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Career Concern Intensity

Figure 4:Economic outcomes and urban outward expansion

10
x 10
Fiscal Revenues

8.5

8.45

8.4
154 156 158 160 162 164 166 168
9
x 10
3.2
Population

3.1

3
154 156 158 160 162 164 166 168
12
x 10
2.7
Output

2.6

2.5
154 156 158 160 162 164 166 168
Urban Outward Expansion

54
Figure 5: The trends of the spatial pattern transition of urban land development
Panel A: The transition of outward development

Average 90th percentile distance to CBD, km

30

28

26

24

22

20
high to low low to high

Panel B: The transition of upward development

Average FAR
2.5

2.4

2.3

2.2
-1 1 -1 1
high to low low to high

Notes: The city leaders whose career-concern intensities are above 0.5 are grouped into the high-type
category, and those whose career-concern intensities are below 0.5 are in the low-type category. In our
sample, there are 19 high-to-low-type pairs of adjacent city leaders and 13 low-to-high-type pairs of
adjacent city leaders. Panel A plots the average of the 90th percentile distance to the city center of land
sold during the term of the first city leader in each type of pairs (indicated by “-1”), and the average of
the 90th percentile distance to the city center of land sold during the term of the second city leader in
each type of pairs (indicated by “1”). Panel B graphs the average regulatory FAR limits of land sold
during the term of the first city leader in each type of pairs (indicated by “-1”), and the average regulatory
FAR limits of land sold during the term of the second city leader in each type of pairs (indicated by “1”).

55
 Appendix
Table A1: Outward expansion and urban land area
Sqrt. Newly added
Sqrt. All built-up Sqrt. Built up land
urban land area
area contiguous to area (sq. km).
(sq. km). Source:
urban settlements Source: City
Chinese
(sq.km). Source: Statistical
Yearbooks of
Landsat, 2005, Yearbooks, 2000-
Land Resource,
2008, 2010, 2013 2013
2003-2013
(1) (2) (3)
City's 90th pct. distance to CBD 0.0101*
(land sold during the corresponding time period) (0.0053)
City's mean distance to CBD 0.0120*** 0.0084*
(land sold during the corresponding year) (0.0044) (0.0043)
City fixed effects Yes Yes Yes
Sqrt of initial built-up land area Yes Yes Yes
Observations 541 816 871
R-squared 0.9780 0.7450 0.9469
Notes: For the regression corresponding to col (1), we divide our land sample into four time periods
based on the year of land transaction: 2000-2005 (46), 2006-2008 (95), 2009-2010 (199), and 2011-2012
(201). Numbers of cities sampled in each period are in the parentheses following the period. The
regression in col (1) relates the square root of a city’s built-up land area in the end of each period to the
city’s 90th percentile distance to the city center of land sold during the period, controlling for the city
fixed effects and square root of the city’s built-up land area in the beginning of the period (source:
Landsat). The regression in col (2) relates the square root of a city’s newly added urban land area in each
year (source: Chinese Yearbooks of Land Resource) to the city’s mean distance to the city center of land
sold in the year, controlling for the city fixed effects and square root of the city’s built-up land area in the
end of the previous year (source: Urban Statistical Yearbooks).The regression in col (3) relates the square
root of a city’s built-up land area in each year (source: Urban Statistical Yearbooks) to the city’s mean
distance to the city center of land sold in the year, controlling for the city fixed effects and square root of
the city’s built-up land area in the end of the previous year (source: Urban Statistical Yearbooks).
*** p<0.01, ** p<0.05, * p<0.1 Heteroskedasticity-robust standard errors are in parenthese.

56
 Table A2: Career concerns and various percentiles of land distance to the city center
Distribution of distance to CBD, residential land sold within each city leader's term, km
90th pct. 85th pct. 80th pct. 75th pct. 70th pct. 65th pct. 60th pct. 55th pct. 50th pct.
[26.9] [23.4] [20.8] [18.6] [16.8] [15.1] [13.7] [12.6] [11.4]
(1) (2) (3) (4) (5) (6) (7) (8) (9)
City leader's career-concern intensity 51.057** 32.796* 17.510 2.281 2.615 -1.066 1.564 -3.515 -5.680
(24.794) (19.018) (19.621) (16.040) (13.586) (13.225) (12.900) (12.020) (11.483)
Observations 377 377 377 377 377 377 377 377 377
R-squared 0.972 0.974 0.970 0.974 0.975 0.975 0.965 0.967 0.965
Distribution of distance to CBD, residential land sold within each city leader's term, km
45th pct. 40th pct. 35th pct. 30th pct. 25th pct. 20th pct. 15th pct. 10th pct. Mean
[10.2] [9.2] [8.4] [7.3] [6.3] [5.6] [4.8] [3.9] [13.5]
(10) (11) (12) (13) (14) (15) (16) (17) (18)
City leader's career-concern intensity -5.820 -7.637 -6.067 -5.355 -3.310 -2.218 -3.042 -1.805 5.229
(8.788) (8.483) (7.227) (5.769) (4.887) (4.125) (4.077) (3.438) (9.567)
Observations 377 377 377 377 377 377 377 377 377
R-squared 0.972 0.966 0.965 0.968 0.973 0.971 0.965 0.952 0.981
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are calculated on the basis of 1,000 bootstrap replications. Each regression is weighted by the number
of land parcels sold during the city leader's term. Each regression controls for the city fixed effects, the dummies of the year when the city leader took office, the dummies of
the year when the city leader left office, the province-specific linear trends (for both on & off years), the square root of built-up land area at the end of the year before the city
leader took office, city's total registered population, GDP per capita, investment in fixed asset, investment in real estate development in the year before the city leader took
office, and the % change of total registered population, GDP per capita, investment in fixed asset, investment in real estate development over the two years before the city leader
took office. The means of each distance-based outward measures are in brackets, and the unit is km.

57
 Table A3: Characteristics of land parcels by land use type

Commercial Residential Industrial

Total land sale revenue (100 million
RMB) 11,289 46,813 7,497
Share land sale revenue 17% 72% 11%
Total land area (sq.km) 386 1,693 1,756
Summary statistics
Distance to the city center (km) 19.61 21.43 26.91
(22.07) (24.83) (22.85)
[12,718] [30,017] [42,096]
Price per land unit (yuan per sq.m) 4,858.47 3,120.17 1,061.97
(13,475.40) (5,002.70) (42,813.40)
[12,707] [29,972] [42,068]
Regulatory FAR 2.43 2.38 1.54
(1.68) (1.17) (0.58)
[12,034] [30,017] [20,722]
Notes: Standard deviations are in parentheses. Numbers of observations are in brackets.

58
 Table A4 Panel A: Outward expansion, start age and start level
Full sample Drop four province-level cities
90th pct. 85th pct. Mean 90th pct. 85th pct. Mean
(1) (2) (3) (4) (5) (6)
Start age -1.977 -1.372 -0.128 -1.977* -1.372 -0.128
(1.209) (0.949) (0.481) (1.191) (0.935) (0.474)
Dummy: Deputy-province -88.330 -31.509 13.042 -88.330 -31.509 13.042
(112.788) (95.926) (42.554) (111.132) (94.517) (41.929)
City fixed effects Yes Yes Yes Yes Yes Yes
On-year & off-year dummies Yes Yes Yes Yes Yes Yes
Provincial linear year trends (on & off years) Yes Yes Yes Yes Yes Yes
Sqrt. of initial built-up land area Yes Yes Yes Yes Yes Yes
Initial city time-variant variables Yes Yes Yes Yes Yes Yes
Observations 377 377 377 366 366 366
R-squared 0.974 0.976 0.982 0.970 0.972 0.977
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are clustered by city. Each regression is weighted by the number of land parcels sold during the city
leader's term.

59
 Table A4 Panel B: FAR, start age and start level

Regulatory FAR upper limit

Buffer
Full sample
sample
excluding 4 Buffer
Full sample excluding 4
provincial- sample
provincial-
level cities
level cities
(1) (2) (3) (4)
Start age 0.013** 0.013** 0.014*** 0.014***
(0.006) (0.006) (0.002) (0.002)
Dummy: Deputy-province 0.774 0.789 0.859*** 0.866***
(0.562) (0.558) (0.164) (0.164)
City fixed effects Yes Yes
Buffer fixed effects Yes Yes
Log(land distance to the city center+1)*city dummies Yes Yes
Transaction year fixed effects Yes Yes Yes Yes
Transaction quarter fixed effects Yes Yes Yes Yes
Province-specific linear transaction year trends Yes Yes Yes Yes
Other initial city controls Yes Yes Yes Yes
Observations 29,957 26,567 269,221 240,529
R-squared 0.321 0.306 0.514 0.495
Notes: *** p<0.01, ** p<0.05, * p<0.1 Standard errors (in parentheses) are clustered by city.

60