Transportation Research Part A 113 (2018) 279–290

Contents lists available at ScienceDirect

Transportation Research Part A

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

Does growth follow the rail? The potential impact of high-speed

T
rail on the economic geography of China
Mi Diao
Department of Real Estate, National University of Singapore, Singapore

A R T IC LE I N F O ABS TRA CT

Keywords: The rapidly expanding high-speed rail (HSR) network in China has produced and will continue to
High-speed rail produce a progressive contraction of space by significantly shortening the rail travel time among
Accessibility major Chinese cities. Does economic growth follow the extension of HSR as a result of improved
Fixed asset investment accessibility? This study investigates the impact of HSR on the economic geography of China. We
Economic geography
find that HSR has improved the accessibility (as measured by weighted average travel time) of
China
Chinese cities by 12.11% at the national level from 2009 to 2013. However, the accessibility
benefit of HSR is not distributed evenly over space. Cities in the wealthy eastern region and with
HSR access enjoy higher accessibility benefit compared with cities in the hinterland and without
direct HSR access. Using a difference-in-differences analysis and an instrumental variable
strategy to address the non-random placement of HSR stations, we also find that on average, HSR
cities have experienced a significant increase in fixed asset investment after the inauguration of
HSR service, which could stimulate future economic growth. The treatment effect of HSR on
investment varies with city size. Second-tier cities with relatively large population bases benefit
more from HSR in attracting investment compared with small cities and mega cities, which could
experience marginal or negative investment growth.

1. Introduction

China’s economy has experienced dramatic growth in the last three decades, accompanied by rapid urbanization process. In an
attempt to meet the increasing demand for inter-city travel and promote economic growth, the Chinese State Council approved the
“Mid-to-Long Term Railway Network Plan” in 2004 to connect major cities across the country with bullet trains. The plan sets the
target at a total of 100,000 kilometer (km) of railroads by the year 2020, of which 12,000 km will be high-speed rail (HSR) lines with
an operating speed of 200 km per hour (km/h) and faster.1 HSR construction in China has accelerated since 2008, in part because of
the aggressive fiscal policies to stimulate economy in response to the global economic recession. The target set by the plan was
revised upward to 120,000 km of railroads by 2020, of which 16,000 km will be HSR lines.2 It is estimated that the total investment
on the HSR system during the 12th five-year plan of China (2011–2015) only was over 1.8 trillion RMB Yuan (approximately
300 billion US Dollars).3 With the ongoing HSR construction projects, China already has the longest HSR network in the world, which
is more than the rest of the world's HSR tracks combined.4 China is now making even more ambitious plans to extend the HSR

E-mail address: rstdm@nus.edu.sg.

1
State Council, 2004, http://www.gov.cn/ztzl/2005–09/16/content_64413.htm.
2
State Council, 2008, http://www.nra.gov.cn/zggstl/wggstlghqk/2008/201312/t20131227_4084.html.
3
http://finance.sina.com.cn/roll/20110121/23069293769.shtml.
4
Official website of the Central Government of China, http://www.gov.cn/xinwen/2016–09/10/content_5107333.htm.

https://doi.org/10.1016/j.tra.2018.04.024
Received 4 August 2017; Received in revised form 6 March 2018; Accepted 23 April 2018
0965-8564/ © 2018 Elsevier Ltd. All rights reserved.
M. Diao Transportation Research Part A 113 (2018) 279–290

network to reach Europe as well as Singapore.5 Fig. 1(a) plots the proposed national grid of passenger-dedicated lines (PDL), which
forms the skeleton of the HSR network in China.
The expansion of HSR network has produced and will continue to produce a progressive contraction of space by significantly
improving the accessibility by rail among major Chinese cities (Zhu et al., 2016). For example, the travel time between Beijing and
Shanghai has been shortened from approximately 10 h with the conventional rail service to 4.5 h with the Beijing-Shanghai HSR line.
Despite the measurable gains in rail travel time across cities, whether and to what extent the huge investment in HSR can stimulate
economic growth and reshape the economic geography of China is still unclear.
Transport infrastructure is often considered a key to promote economic growth and development by reducing the time cost of
travel and shipping. There is a rich body of literature that investigates the relationship between transport infrastructure investment
and various aspects of economic development, for example, long term GDP effects (Banerjee et al., 2012; Faber, 2013; Qin, 2017),
gains from trade (Donaldson, 2018), agricultural land value (Donaldson and Hornbeck, 2013), real estate price (Zheng and Kahn,
2013; Diao et al., 2017b), skill premium in local markets (Michaels, 2008), employment (Duranton and Turner, 2012), population
growth and urbanization (Atach et al., 2010), and suburbanization and urban form (Baum-Snow, 2007; Baum-Snow et al., 2012; Zhu
and Diao, 2016). These studies mainly investigate two fundamental issues. The first major concern is on the causality between
transport infrastructure investment and economic growth: does transport infrastructure investment induce or follow economic
growth? The second issue is concerned with the distributional effects of transport infrastructure investment: does the subsequent
transportation cost reduction lead to the diffusion of economic activities to peripheral regions, or does it reinforce the concentration
of production in existing urban centers? The empirical work has mainly focused on assessing the economic impacts of the railway
network (Atach et al., 2010; Donaldson and Hornbeck, 2013) and the interstate highway system (Baum-Snow, 2007; Michaels, 2008)
in the United States, the railway and highway networks in modern China (Banerjee et al., 2012; Baum-Snow et al., 2012; Faber, 2013;
Qin, 2017), the railway network in colonial India (Donaldson, 2018) or the intra-city rail transit systems in cities all over the world
(Diao, 2015; Diao et al., 2017b).
As an important transport innovation, HSR differs from conventional modes of transport such as highways and conventional
railroads in multiple ways, for example, the high construction cost, the high operating speed, and almost exclusively for passenger
transport. In the meantime, due to its considerable contribution to accessibility, HSR is likely to have significant implications for
nurturing development and driving population agglomeration and urban growth like other transport modes. However, empirical
knowledge on the relationship between HSR and economic growth is still limited. In Japan, regions served by the Shinkansen
achieved higher growth rates than those without direct Shinkansen service (Sands, 1993), but the impacts exclusively attributable to
the presence of the Shinkansen are hard to identify because of the many other confounding factors that could have stimulated
economic growth in the region (Givoni, 2006). In France, although the introduction of Train a Grande Vitesse (TGV) services between
Paris and Lyon led to significant transport benefits, its economic impacts vary across cities (Banister and Berechman, 2000).
The very recent upsurge of HSR in China offers us an empirical setting to examine the spatial-temporal change because of an
extensive HSR network developed over a relatively short time span, and to explore how transportation infrastructure investment will
influence the growth of Chinese cities and reshape the economic geography of China. An increasing number of studies have examined
the potential impact of HSR in China in multiple aspects, including accessibility (e.g., Shaw et al., 2014; Jiao et al., 2014), regional
development (Zhu et al., 2016; Chen and Haynes, 2017), and real estate market (Zheng and Kahn, 2013; Diao et al., 2017a). Shaw
et al. (2014), Jiao et al. (2014), and Zhu et al. (2016) find that the HSR could lead to substantial improvement in accessibility for
Chinese cities. However, the accessibility benefit of HSR is not distributed evenly over space, thus increasing the inequality between
different regions, between cities with different sizes, and between cities with different accessibility levels to HSR stations. HSR could
affect real estate markets through facilitating market integration and urban development. At the regional scale, Zheng and Kahn
(2013) find that HSR is associated with rising real estate prices in secondary cities connected by HSR, as HSR enables households and
firms to enjoy the agglomeration economy of mega city without having to living within its boundary. At the city scale, Diao et al.
(2017a) find that intra-city access to inter-city transport nodes such as HSR stations could translate into a price premium in the real
estate market, but this premium is affected by the connection between the station and the city. Salzberg et al. (2013) assess the
economic development benefit resulting from larger and better connected markets in four cities along a newly-constructed HSR line
and find that it is of the same order as the direct transport benefits.
These studies provide important insights into the potential role of HSR in the development of Chinese cities. However, research on
how HSR will reshape the economic geography of China while addressing the causality issue and the distributional effect issue in the
literature is still limited. This study aims to contribute to the literature on transport infrastructure investment and China’s economic
growth by: (1) estimating the accessibility gain in China in recent years due to the construction of HSR; (2) assessing the potential
impact of HSR on the economic growth of Chinese cities resulting from the changes in accessibility through the channel of fixed asset
investment (FAI); and (3) exploring the distributional effects of HSR investment among Chinese cities.
Utilizing a unique dataset, the national train schedules of China at a number of points of time between 2009 and 2013, we
construct accessibility measures to quantify the rail accessibility in all prefecture-level Chinese cities connected by railways. We find
that the ongoing construction of HSR network has significantly changed the spatial-temporal distance of the country by greatly
improving the inter-city accessibility for the connected cities. The weighted average travel time (WATT) by rail in prefecture-level
Chinese cities, on average, was reduced by 12.11% from 2009 to 2013. However, the accessibility gain is distributed unevenly in the
country. HSR cities experienced higher accessibility gains compared with non-HSR cities and the wealthy eastern region benefited

5
Source: A Closer Look at the Chinese High Speed Rail Juggernaut: The Chinese closer to Elon Musk's Hyperloop than the US. Forbes, August 4th, 2014.

280
M. Diao Transportation Research Part A 113 (2018) 279–290

1(a) Proposed National Grid of Passenger Dedicated Lines in China

1(b) Locations of cities in the DID analysis

Fig. 1. HSR in China.

more from HSR than the economically lagged central and western regions.
After presenting the accessibility changes, we proceed to investigate the potential economic growth resulting from the accessi-
bility change by assessing the impact of HSR on the FAI of Chinese cities. As defined by the National Bureau of Statistics of China, FAI
includes the investment in capital construction, investment in renovation and renewals of existing facilities, investment in real estate

281
M. Diao Transportation Research Part A 113 (2018) 279–290

development, investment in other fixed assets by state-owned units, investment in other fixed assets by collectively owned units, and
private investment in housing construction. As a lead economic indicator, FAI is one of the fundamental factors in facilitating
economic growth in China (Ahuja et al., 2012).
Employing a difference-in-differences (DID) analysis, we compare the monthly FAI of a “treatment” group and a “control” group
before and after the inauguration of HSR to estimate the causal effect of gaining HSR access on FAI. The treatment group consists of
cities that were connected to HSR before 2013, whereas the control group consists of cities that had no HSR access during this period.
To address the endogeneity concerns related to the non-random placement of HSR stations, we employ an instrumental variable (IV)
strategy based on China’s double-track railway network in 1978 as a robustness check on the DID estimates.
We find evidence that HSR cities have experienced a significant growth in investment after they were connected to the HSR
network. Our estimates indicate that HSR increases monthly FAI by approximately 3.28 billion RMB Yuan on average for HSR cities,
which is equivalent to 12.2% of the average monthly FAI of HSR cities in our sample in 2013. The treatment effect of HSR varies by
city size. Second-tier cities with relatively large population bases have been benefited more from HSR in attracting investment
compared with small cities and mega-cities, which could experience marginal or even negative growth in FAI after being connected
by HSR. Such distributional effects of HSR could lead to the restructuring of the urban system in China.
The remaining part of this paper proceeds as follows. Section 2 introduces data and methodology employed in this study; Section
3 presents results of the accessibility analysis; Section 4 presents results from the econometric analysis and discusses the relationships
between HSR and FAI in Chinese cities; and Section 5 draws conclusions and discusses policy implications.

2. Methodology and data

This section describes the data and methodology employed in this study to examine the evolution of the contour surface of rail
accessibility in China over time and investigate the impact of HSR on FAI in prefecture-level Chinese cities. A prefectural-level city is
an administrative division ranking below a province and above a county in Mainland China's administrative structure. There are four
province-level cities in China: Beijing, Shanghai, Tianjin and Chongqing. Despite these four cities having higher rank in the ad-
ministrative structure of China than prefecture-level cities, they are also included in the analysis.

2.1. Data

There is a growing body of research on the accessibility benefit of railways. However, due to data limitations, previous studies
often rely on simple geospatial database of railway networks to find the shortest distance between city pairs and apply hypothetical
speed or cost assumptions to estimate travel time and travel cost (e.g., Donaldson and Hornbeck, 2013; Zheng and Kahn, 2013).
Critical issues on accessibility such as the real travel speed, the availability and frequency of train services between city pairs were not
considered. The related measurement errors could bring biases to the analyses.
This study benefits from a unique dataset, which includes the national train schedules of Mainland China at four time points: April
10, 2009, December 16, 2010, January 1, 2012, and January 1, 2013.6 The train schedule data enable us to extract a complete picture
of the evolution of rail accessibility in China since 2009 with a degree of precision and coherence unavailable to earlier researchers.
These 4 points in time are of interest because they occur between the inaugurations of several major HSR lines, most notably, the
Wuhan-Guangzhou line on December 26, 2009; the Beijing-Shanghai line on June 30, 2011; the Harbin-Dalian line on December 1,
2012; and the Beijing-Wuhan line on December 26, 2012. The train schedules provide rich information about train services in China,
including the arrival and departure time of all trains at each and every station, covering approximately 4000 trains and 3000 stations
each. Based on the HSR definition given by the EU,7 there are three types of train services in China that can be classified as HSR,
whose IDs start with letters “D”, “G”, and “C”, respectively. The D-series trains (Dongche) have a peak speed range of 200–250 km/h.
The G-series trains (Gaosu) provide long-distance HSR services, and the C-series (Chengji) are regional intercity express railway
services. Both the G- and C-series of train services have peak speeds of over 300 km/h. As of January 1, 2013, there were 1031 D-
series train services that made stops at 279 stations in 105 prefecture-level cities and 963 G- or C-series train services serving 117
stations in 45 prefecture-level cities. With the highest operation speed among all trains and the strategic locations of cities that they
serve, the G- and C-series train services greatly shrink the spatial-temporal distance among major Chinese cities and could have
substantial impacts on the spatial-temporal and socio-demographic landscapes of China. In this study, we will examine the temporal
change in rail accessibility based on national train schedules and assess the impact of gaining access to G- and C-series of trains on the
FAI of prefecture-level cities.
The monthly FAI data are collected from the official websites of municipal statistical agencies.8 Among the 305 prefecture-level
cities in Mainland China, 80 cities provide monthly FAI data for the complete part of or a majority part of our study period from

6
The train schedule data were originally from Tielu.org and collected and compiled for non-commercial use by Qi Zhao. Tielu.org updates the databases on a real
time basis according to the latest official schedule for passenger trains released by the Ministry of Railways of China.
7
According to the European Commission Directive 96/48, 1996, the standard of speed for HSR is set at 250 km/h for dedicated new lines and 200 km/h for
upgraded lines in respect of the infrastructure capabilities.
8
The investment on fixed assets at municipalities in China is reported monthly. According to the National Bureau of Statistics of China, investments in cross-
province projects, such as railway, electricity, and water resource projects, were reported by the corresponding administration in-charge such as the ministry of
railways, not by the local governments. Therefore, the FAI data used in this research should not include investment in HSR (all cross-province projects). Source: http://
www.stats.gov.cn/tjzs/cjwtjd/201308/t20130829_74320.html.

282
M. Diao Transportation Research Part A 113 (2018) 279–290

February 2009 to July 2013. Among these cities, 25 cities were connected to the HSR network during this period, and 55 cities had no
HSR service, which can be employed as control group to isolate the impact of HSR on FAI. The locations of these cities can be found in
Fig. 1(b).

2.2. Travel time matrix and accessibility measure

Accessibility measures the level of easiness with which activities can be reached from a certain place and with a certain system of
transport (Morris et al., 1978). In this study, we employ the weighted average travel time (WATT) (Gutierrez and Urbano, 1996) to
characterize the accessibility of individual prefecture-level cities to the rest of the country.
To compute the accessibility measure, we first derive the shortest travel time between all station pairs directly or indirectly
connected by trains. A graph model is created based on the train schedule at a particular point in time p, in which each node
represents a train station and the edge between two nodes indicates a train service that directly connect the pair of stations re-
presented by two nodes. Both nodes and edges have attributes. The attributes of nodes include the GDP and population of corre-
sponding cities. The attributes of edges include the actual rail travel time and capacity of the train service at time p. A shortest-path
algorithm is applied to the graph model to create a station-level rail travel time origin-destination (OD) matrix Tstation,p for time p,
defined as:
Tstation,p = [tghp]n ∗ n (g = 1,2,…,n;h = 1,2,…,n)

Note that each element in the OD matrix tghp is computed by following the shortest path through multiple connections in the railway
network between an origin station g and a destination station h with a penalty of 2 h per transfer in time p. tghp = 0 when g = h. n is
the total number of stations.
Because large cities could have more than one train station, a prefecture-level OD matrix Tprefecture,p for time p is generated based on
the station-level OD matrix. The value of each cell in the prefecture-level OD matrix is the minimum travel time between a pair of
railway stations g and h that are located in the origin prefecture i and destination prefecture j, respectively. m is the total number of
prefecture-level cities.
Tprefecture,p = [tijp]m ∗ m (i = 1,2,…,m;j = 1,2,…,m)

tijp = Min (tghp) ( ∀ g ∊ i ,∀ h ∊ j )

Based on the prefecture-level rail travel time matrix, we compute the WATT of a city i in time p as the GDP weighted travel time to all
other cities with the following formula:
m
∑ (tijp Mj )
j
Aip = m
∑ (Mj )
j

where Aip is the WATT of city i in time p; tijp is the rail travel time between cities i and j in time p; and Mj is the GDP of city j in 2010.
This static GDP-weighting approach allows us to capture the accessibility change simply due to mobility improvement while ac-
counting for the relative importance of cities .9 With the WATT measure for all prefecture-level cities connected by trains in China, we
are able to interpolate the accessibility surface of China and visualize its evolution over time as the expansion of the HSR network.
It should be noted that we use inter-city travel time to construct the accessibility measure in this study. However, as Diao et al.
(2017a) indicate, the intra-city travel time to reach HSR stations could affect the accessibility benefit brought by HSR in China.

2.3. Econometric modeling

Investigating the relationship between HSR and investment is critical for understanding the potential HSR-induced growth in
China. FAI can be viewed as a lead indicator of economic growth, especially in China, a country where investment is one of the key
driving forces of the economy and constitutes approximately one-half of GDP growth (Ahuja et al., 2012). A detailed description of
FAI in China can be found in the endnote.10

9
Alternatively, researchers can also use GDP in time p as weights to compute WATT in time p. However, with this dynamic weighting approach, the accessibility
change is affected by both travel time change and GDP change. Therefore, the impact of pure mobility improvement cannot be identified.
10
To help interpret the results, we use the national FAI in 2010 as an example to illustrate FAI practices in China. In 2010, the total investment in fixed assets in the
whole country was 27812.19 billion yuan, among which 24143.09 billion yuan was invested in urban areas and 3669.10 billion yuan was invested in rural areas.
According to registration status, the FAIs from domestic funds, Hong Kong, Macao and Taiwan funds, and foreign funds were 26091.44 billion, 829.51 billion, and
891.24 billion yuan, respectively. Among domestic funds, 8331.65 billion was from state-owned, 1004.19 billion was from collective-owned, 144.56 billion was from
cooperative, 83.1 billion was from joint, 7032.15 billion was from limited liability, 1720.3 billion was from share-holding, 6057.23 billion was from private,
950.67 billion was from self-employed, and 767.6 billion was from others. In terms of usage, the investment in construction and installation reached 17135.18 billion
yuan in 2010, with another 6168.15 billion invested in the purchase of equipment and instruments, and 4508.85 billion in others. If FAI was grouped by jurisdiction of
management, 2279.06 billion yuan was classified into central investment, and 21864.02 billion belonged to local investment. Source: China Statistical Yearbook,
http://www.stats.gov.cn/tjsj/ndsj/2011/indexeh.htm.

283
M. Diao Transportation Research Part A 113 (2018) 279–290

The location determinants of investment in China have been explored by a number of studies. For example, Zhao and Zhu (2000)
found that at the city level, infrastructure, market size, and policy incentives are important location determinants of incoming
Foreign Direct Investment (FDI). In addition, labor qualities, industrial agglomerations and prior FDI stocks are other factors iden-
tified influencing the location decisions of FDI (Cheng and Stough, 2006). In this study, we apply a difference-in-differences (DID)
estimator to compare the prefecture-level cities connected by the HSR network to the ones without direct HSR access, before and after
the HSR stations commenced operation to estimate the treatment effect of HSR on FAI.
Cities that are connected to the HSR network might differ systematically from those that have no direct HSR access, which could
bias the estimated treatment effect of HSR. In this study, we utilize city fixed effects to eliminate the heterogeneity at the city level,
accounting for time-invariant city attributes. In the meantime, FAI could be associated with the temporal dynamics of the national
economy. To remove the underlying time trends, we include the month fixed effect in the model. Our approach constitutes a quasi-
experiment procedure that can isolate the causal impact of HSR on FAI. The DID model can be formulated as follows:
FAIit = β0 + β1 Cityi + β2 Montht + β3 HSRi ∗Afterit + εit (1)

where FAIit is the monthly FAI of city i in month t; Cityi and Montht control for city fixed effect and month fixed effect, respectively;
HSRi is a dummy variable indicating a HSR city; Afterit is a dummy variable indicating that the period is after HSR in service; the
interaction term of HSRi and Afterit captures the treatment effect of HSR, which takes the value of 1 for HSR cities after HSR was in
service, and 0 for non-HSR cities and HSR cities before HSR was in service; εit is an error term.
The impact of HSR could vary across cities with different economic sizes. To capture the city size effect, we further interact the
population and squared population of the city with the DID term. The model can be specified as follows:

FAIit = β0 + β1 Cityi + β2 Montht + β3 HSRi ∗Afterit + β4 HSRi ∗Afterit ∗Popi + β5 ∗HSRi ∗Afterit ∗Popi2 + εit (2)

Some HSR cities could have significant accessibility gains, while others may only enjoy moderate or minimal benefits from HSR
because of the differences in both the accessibility level in the pre-HSR era and the frequency and coverage of new HSR services
provided. To check for this heterogeneity, we compute the change in WATT before and after HSR operation for each HSR city and
integrate the actual accessibility gain into our DID estimator. Acce_1st and Acce_4th are two dummy variables indicating the 1st
quartile (with larger accessibility gain) and 4th quartile (with smaller accessibility gain) of HSR cities sorted by their savings in
WATT, respectively. These two variables are interacted with the DID term in Model (3).

FAIit = β0 + β1 Cityi + β2 Montht + β3 HSRi ∗Afterit + β4 HSRi ∗Afterit ∗Popi + β5 ∗HSRi ∗Afterit ∗Popi2 + β6 HSRi ∗Afterit ∗Acce1sti
+ β7 HSRi ∗Afterit ∗Acce 4thi + εit (3)

One strong ex ante assumption underlying the DID estimator is that the placement of HSR stations was randomly assigned. However,
in reality HSR could be used as an instrument to stimulate the economic growth of lagged cities or be located in cities that are
believed to be booming in the future so that they can provide sufficient ridership to support the successful operation of HSR. In either
case, the Ministry of Railway (MOR) of China is unlikely to allocate HSR stations randomly. To address the potential endogeneity, we
adopt an IV regression approach as a robustness check. The IV is a variable that predicts gaining HSR access controlling for other
factors but is otherwise uncorrelated with the FAI during the study period. It is employed to isolate the plausibly exogenous variation
in HSR access, similar to what would be the case if HSR access had been randomly assigned. Following the literature (e.g., Duranton
and Turner, 2011; Zheng and Kahn, 2013), we choose China’s double-track railway network in 1978 as the IV in this study. A double-
track railway usually involves running one track in each direction, compared with a single-track railway where trains in both
directions share the same track. Therefore, double-track railway lines have higher capacity than single-track ones. In 1978, the total
length of double track railway is approximated 7630 km, which accounts for 14.7% of the total track length in China11. The major
double-track railway lines in 1978 include Beijing-Shanghai line, Beijing-Guangzhou line (the segment between Beijing and Hen-
gyang had double tracks, while the remaining segment from Hengyang to Guangzhou was still single-track line), Harbin-Dalian line,
and Zhengzhou-Baoji line, connecting some of the major cities in China. The validity of the 1978 double-track railway network as an
instrument depends on the fact that the double-track railway network was constructed in the pre-reform era of China (1949–1978).
The year of 1978 marked a significant milestone in modern China when China started to switch from a centrally planned economy to
a market based economy. The double-track railway network of China before 1978 serve as a means for shipping raw materials and
manufactured goods according to the national and provincial plans rather than facilitating investment in cities. After 30 years of
reform, market is now playing a central role in the Chinese economy and economic activities would have ample time to relocate since
1978. This instrument is closely correlated with the likelihood that a city is connected by HSR but unlikely to be correlated with the
unobserved determinants of the city’s recent FAI.
Furthermore, to control for other factors that could influence investment and are correlated with the double-track railway net-
work, we include a set of city attributes such as GDP, population and region in the first stage of the IV regression. Thus conditional on
the control variables, we expect the 1978 double-track railway lines constructed in the pre-reform era of China to affect the FAI in
modern market-based Chinese cities only through their effects on the HSR network. We generate the instruments based on a digital
map of major railway lines of China in 1978. All prefecture-level cities connected to the 1978 double-track railway network are
identified directly from this map.

11
Source: National Bureau of Statistics of the People’s Republic of China, http://www.stats.gov.cn/ztjc/ztfx/jnggkf30n/200811/t20081111_65698.html11

284
M. Diao Transportation Research Part A 113 (2018) 279–290

Fig. 2. Surfaces of weighted average travel time by train.

3. Accessibility analysis

Accessibility is a widely used measure in geography and transport studies. Examining the accessibility surface of China enables us
to understand the overall spatial-temporal landscape of China and to evaluate a city's reachable opportunities and resources.
In this study, we compute the WATT for 280 prefecture-level cities connected by railways at four key points of time, respectively:
April 10, 2009, December 16, 2010, January 1, 2012, and January 1, 2013. Fig. 2 plots the surface of WATT by train over time. The
accessibility surface generally presents a concentric pattern around the geometric center of the economy and the railway network of
China. Areas with good accessibility spread to more areas as the development of HSR and the expansion of regular railroads in the
western part of China. On December 16, 2010, only a few cities in Central China such as Zhengzhou and Wuhan had a WATT of less
than 12 h. After the Beijing-Shanghai HSR line was put in service on June 30, 2011, a corridor with WATT of less than 12 h formed
along the middle section of the new HSR line. As increasingly more HSR lines were put in service in the following two years, this area
with good rail accessibility expanded to the triangle area with three vertices located at the regions surrounding Beijing, Shanghai and
Wuhan. The improved accessibility grants this area greater potential for economic growth.
Table 1 presents the means of WATTs in 2009 and 2013 at different levels of spatial aggregation. We find that the national
average of WATT decreased by 12.11% from 1315.76 min in 2009 to 1156.48 min in 2013, but the accessibility gain is not distributed
evenly across the country. HSR cities experienced higher accessibility gains during 2009–2013 compared with non-HSR cities, as we
would expect. The average WATT for the 45 HSR cities decreased by 24.67% from 2009 to 2013, compared with a 10.55% reduction
in WATT for the remaining 235 non-HSR cities. In the regional development literature, Mainland China is often divided into three

Table 1
Statistics on weighted average travel time (WATT).
Number of cities WATT 2009 (minute) WATT 2013 (minute) Change

Nation 280 1315.76 1156.48 −12.11%

HSR cities 45 901.24 678.90 −24.67%

Non-HSR cities 235 1395.14 1247.93 −10.55%

Eastern Region 101 1161.60 986.41 −15.08%

Central Region 111 1134.10 999.18 −11.90%
Western Region 68 1841.30 1665.85 −9.53%

285
M. Diao Transportation Research Part A 113 (2018) 279–290

Fig. 3. Area cartogram of prefectures resized in proportion to WATT 2009 and shaded by percentage change in WATT, 2009–2013.

regions geographically, the eastern, the central and the western regions (Fig. 1(a)), based on their different natural endowments and
socioeconomic characteristics. The eastern region is the wealthiest among the three regions, while the western region is lagging
behind in economic development. The regional gap has been widened since the opening up of China in 1978 due to the uneven
development policy (Fan, 1997). To boost economic efficiency, the eastern region was designed to be developed first. New economic
policies were first introduced in several open economic zones in the eastern region to attract foreign direct investment and gradually
spread to the hinterlands. Reducing the regional disparity has been a policy goal for the central government of China since the 1990s
and investment in transport infrastructure such as railways and highways to link the periphery to the center has been an important
component of the policy portfolio of the central government to reduce regional inequality. With the expansion of the HSR network, all
three regions experienced a steady reduction in WATT from 2009 to 2013. The eastern region benefitted the most, with the mean
WATT decreased 15.08% from 1161.6 min in 2009 to 986.41 min in 2013. While during the same period, the central and western
regions experienced 11.90% and 9.53% reductions in mean WATT, respectively. Therefore, instead of contributing to the spatial
integration of Chinese cities, the current HSR network has widened the accessibility gaps between cities connected to the HSR
network and cities without direct HSR access, as well as the gaps among different regions of China.
In Fig. 3, all prefecture-level cities in Mainland China are shaded based on the percentage change in WATT from April 10, 2009 to
January 1, 201312. Darker colors mean larger accessibility gains. To capture the differences in rail accessibility among the cities in the
base year 2009, we create a cartogram that transforms the boundaries of cities so that areas of cities are proportional to their WATT in
2009 while keeping the total area of China unchanged. Thus, cities with good accessibilities in 2009 (e.g., cities in the eastern region)
shrank while those with longer average travel times (e.g., cities in the western region) expanded in area. In Fig. 3, the thick gray line
depicts the boundary of China before the transformation and the narrow gray line shows the boundary of Chinese provinces after the
transformation. The major PDL lines as of January 1, 2013 are overlaid on top of the map of accessibility change. Fig. 3 suggests that
cities with larger accessibility gains during 2009–2013 correlate with cities that had better rail accessibility in 2009 and major PDL
corridors in 2013. Some peripheral cities in northwest China that had no railway service before 2009 also enjoyed significant
accessibility gains due to the construction of new conventional railways that connected them to the core area of the country. It
appears that the “preferential attachment” effect also exits in the new round of HSR construction.

12
For cities without rail access, the weighted average travel time is computed as the sum of the weighted average travel time of its nearest city with rail access and
the road travel time between the two cities. The road travel time is computed as the straight-distance between the two cities divided by a speed of 30 km/hour.

286
M. Diao Transportation Research Part A 113 (2018) 279–290

Table 2
Summary statistics for cities in the DID analysis.
Obs. Central Region Western Region GDP (10 Billion RMB Population (Million) WATT 2009 WATT 2013 (Minutes)
(0/1) (0/1) Yuan) (Minutes)

HSR cities 25 0.36 0.04 45.151 7.50 920.36 686.16

Non-HSR cities 55 0.35 0.22 16.079 4.87 1283.87 1123.99

Table 3
Descriptive statistics of variables in the models.
Obs. Mean Std. Dev. Min Max

1) Descriptive statistics for the first-stage logit model in the IV regression

HSR 80 0.313 0.466 0 1
Double-track rail in 1978 80 0.275 0.449 0 1
Population (million) 80 5.692 4.559 0.685 28.846
GDP (10 billion RMB yuan) 80 25.164 31.069 2.422 171.660
West Region 80 0.163 0.371 0 1
Central Region 80 0.513 0.503 0 1

2) Descriptive statistics for the DID model

FAI (billion RMB Yuan) 3589 14.913 16.211 0.000 126.985
HSR * After 3589 0.190 0.393 0.000 1.000
HSR * After * Pop 3589 1.474 3.742 0.000 23.019
HSR * After * Pop2 3589 16.171 64.013 0.000 529.881
HSR * After * Large Accessibility Gain (1st quartile) 3589 0.049 0.216 0.000 1.000
HSR * After * Small Accessibility Gain (4th quartile) 3589 0.052 0.223 0.000 1.000

4. Measuring the effect of HSR on fixed asset investment

China’s economic growth has been characterized by a heavy reliance on investment. In this section, we employ a quasi-experi-
ment approach to assess the causal effect of HSR on FAI and to understand the potential role of HSR in boosting the economic growth
of HSR cities. Following the model specifications described in the Methodology section, we employ a DID estimator to evaluate the
treatment effect of HSR and use IV regression to control for the non-random placement of HSR stations. All the 80 cities (out of 305
prefecture-level cities in China) with at least 25 monthly FAI reports during the period from February 2009 to June 2013 are included
in the model. Among them, 25 are HSR cities and the remaining 55 are non-HSR cities. The locations of the 80 cities are shown in
Fig. 1(b). Table 2 compares characteristics of HSR and non-HSR cities in our sample. The 25 HSR cities were connected by HSR lines
at 6 different points of time during the study period as a result of the continuous expansion of the HSR network. The DID terms in the
regression models are generated accordingly. Table 3 reports descriptive statistics for the IV regression and DID model, respectively.
Columns 1–3 of Table 5 show the results of the OLS estimation of the DID specification. All models control for city and month
fixed effects. Model 1 suggests that the average treatment effect of HSR on FAI is 1.512 billion RMB Yuan per month, compared with
the average monthly FAI of 26.9 billion in the 25 HSR cities in our sample in 2013. The HSR effect is significant at the 0.01 level.
Estimating specification (1) by OLS would imply the assumption that HSR cities were randomly assigned in the country. Due to the
huge investment in HSR, the MOR may target economically prosperous cities to locate HSR stations in order to generate sufficient
ridership to support the successful operation of HSR and achieve better investment returns. This concern is supported by descriptive
statistics reported in Table 2. HSR cities were on average larger, richer, and more likely to be located in the Eastern region compared
with non-HSR cities.
To address these concerns, we employ an IV strategies based on the 1978 double-track railway network in China. In the first stage
of the IV regression, we calibrate a logit model, utilizing a dummy variable of being connected to China’s double track rail network in
1978 and a set of city level characteristics (population, GDP, and region) to predict the probability that a city was connected to the
HSR network during 2009–2013. The estimation result of the logit model is presented in Table 4. The IV has a positive and highly
significant coefficient, which suggests that the 1978 double-track rail network is a good predictor of being a HSR city between 2009
and 2013, conditional on control variables. The predicted probability is then integrated in the second stage of the DID specification.
Columns 4–6 of Table 5 report the results of the IV regression. Controlling for the non-random placement of HSR stations (Model 4),
the impact of HSR on the monthly FAI of average cities increases to 3.280 billion Yuan, which is equivalent to 12.2% of the average
monthly FAI of the 25 HSR cities in our sample in 2013. The HSR effect is significant at the 0.01 level.
While the coefficient of the DID term documents the average change in FAI resulting from HSR, we hypothesize that the HSR
effect varies across cities of different sizes. To check for this city size effect, we add to the model interactions of the DID term with the
population and population squared of cities, respectively. The results are shown in columns 2 and 5 of Table 5. The coefficients of the

287
M. Diao Transportation Research Part A 113 (2018) 279–290

Table 4
Logit regression on the probability of being HSR cities in 2009–2013.
Coef. Std. err.

Double-track rail in 1978 (dummy) 2.524 0.709***

Population (million) −0.029 0.137
GDP (10 billion RMB Yuan) 0.044 0.024*
Western Region (dummy) −2.422 1.344*
Central Region (dummy) 0.577 0.801
Constant −2.587 0.788***
LR Chi2(5) 37.670 ***
Pseudo R2 0.379

***Statistically significant at the 1% level; ** 5% level; * 10% level.

Table 5
Monthly fixed asset investment regression.
OLS IV Regression

Model (1) Model (2) Model (3) Model (4) Model (5) Model (6)

HSR*After 1.512*** −10.804*** −10.201*** 3.280*** −14.993*** −15.078***

(0.544) (1.844) (1.997) (0.752) (3.051) (3.100)
HSR*After*Large Accessibility Gain (1st quartile) −0.928 −0.784
(1.264) (2.197)
HSR*After*Small Accessibility Gain (4th quartile) −3.279*** −2.073
(1.171) (2.060)
HSR*After*Pop 2.471*** 2.526*** 3.427*** 3.494***
(0.380) (0.392) −0.572 (0.578)
HSR*After*Pop2 −0.083*** −0.083*** −0.118*** −0.117***
(0.015) (0.015) (0.021) (0.021)
Month effects Yes Yes Yes Yes Yes Yes
City effects Yes Yes Yes Yes Yes Yes

n 3589 3589 3589 3589 3589 3589

Adjusted R2 0.808 0.810 0.811 0.808 0.810 0.810

Dependent variable is the monthly fixed asset investment in billion RMB Yuan.
All specifications include month and city fixed effects. Standard error in parentheses.
*** Statistically significant at the 1% level; ** 5% level; * 1% level.

Fig. 4. Impact of HSR on monthly FAI (billion CNY) by population (Million). (Solid Line: IV; Dashed Line: OLS).

288
M. Diao Transportation Research Part A 113 (2018) 279–290

DID term and the two interaction terms are all significant at the 0.01 level. We then plot the HSR effect on FAI by population size
(from 0.5 to 25 million)13 in Fig. 4. Initially, the HSR effect increases with the population size and reaches a maximum value of
8 billion Yuan (Model 2: OLS) and 10 billion Yuan (Model 5: IV) for a city with a population of approximately 15 million. Then, the
HSR effect begins to decline with population size. This relationship suggests that second-tier cities with considerable population bases
tend to benefit more from HSR. The impact of HSR on very small cities and megacities could be trivial or even negative. The results
provide some evidences to support the existence of distributional effect of HSR investment along HSR corridors. Two factors may
contribute to these findings. First, HSR connects small and large cities more tightly. The closer connection may enable firms to
relocate from megacities to nearby second-tier cities on the HSR corridor to reap the agglomeration economy of megacities while
avoiding their negative externalities such as congestion, pollution and high land cost. In the meantime, some economic activities in
very small cities may also be more conveniently diverted to large cities to enjoy the agglomeration spillovers. Second, megacities such
as Beijing and Shanghai generally have well-developed modes of transportation networks, including frequent airlines connecting to
other domestic cities. Thus, the marginal productivity increase from new investments on HSR may not significantly change the
overall multi-modal accessibility level.
The accessibility gain from HSR varies across cities due to the differences in HSR services. To capture the effect of heterogeneous
accessibility gains, we rank HSR cities according to the improvement of WATT before and after HSR was in service and generate two
dummy variables indicating cities with relatively larger (in the 1st quartile of cities) and smaller accessibility gain (in the 4th quartile
of cities) from HSR, respectively. We then include the interactions of the DID term and these two dummy variables in Models 3 and 6.
The results of Model 3 (OLS) suggest that being in the group of HSR cities with relatively smaller accessibility gains could lower
monthly FAI, other factors being constant. This effect is significant at the 0.01 level. However, it becomes insignificant in the IV
regression, as shown in Model 6.

5. Discussion and conclusion

Investment in transportation infrastructure is believed to be beneficial to local economic growth by improving accessibility. The
HSR construction in China provides an empirical setting to study the impact of transport enhancement on the economic growth of
cities. In this study, we employ national train schedules during the period of 2009–2013 to investigate the changing rail accessibility
surface of China in the HSR era, and apply DID and instrumental variable techniques to evaluate the treatment effect of HSR on FAI of
cities, using the 1978 double-track railway network in China as a source of exogenous variation.
We find that the reduced travel time among cities resulted from HSR could help to increase the economic growth potential by
facilitating travel and interaction among cities and regions. The construction of the HSR lines have increased the average accessibility
of Chinese cities by approximately 12.11% (as measured by WATT) from 2009 to 2013 and reshaped the spatial-temporal landscape
of China. Cities with relatively better accessibilities before the HSR era and cities connected by HSR lines enjoy larger accessibility
gains compared with other cities. In the meantime, the wealthier eastern region benefits the most from HSR as of the year of 2013,
while the western region benefits the least. Our research findings suggest the spatial unevenness of China may increase with bullet
trains due to the widen gaps in accessibility across cities and regions. However, with the continuous expansion of HSR network to
central and western regions, it could be expected that there would be rises in accessibility in many central and western cities in the
following years.
With a difference-in-differences estimator, we also find that the accessibility advantage brought by HSR could translate into
higher FAI. Investment crowded into the HSR cities in pursuit of higher returns because of the expected growth. Therefore, economic
growth may follow the extending of the HSR tracks in general. The HSR effect varies with city size. Second-tier cities with con-
siderable population bases benefit the most from HSR, while the impact on small cities and megacities could be marginal or even
negative. It should be noted that although many cities may not have direct rail connection, they can still benefit from HSR services
because travelers in non-HSR cities can access HSR through other transport modes. Due to the spillover effect of HSR services, our
DID analysis may underestimate the impact of HSR on FAI.
This study provides useful insights in understanding the changing economic geography of China, especially the role that the
rapidly expanding HSR network could play in this process. HSR greatly reduced market accessibility gaps among cities connected by
bullet trains. For example, the size of accessible markets from a second-tier city on the Beijing-Shanghai HSR line is not significantly
different from that from Beijing and Shanghai. As a result, more firms and skilled labor may relocate from mega cities to second-tier
cities connected by HSR to enjoy the agglomerated economy while avoid the negative externalities of megacities. In the meantime,
small HSR cities may experience an outflow of investment because economic activities may be more conveniently shifted to larger
cities with HSR. Cities bypassed by HSR lines could suffer because of the increased accessibility inequality, which can make non-HSR
cities even less attractive in drawing investment and firms. China’s economy is in a dynamic transition into the future. Our research
findings could be helpful for municipalities and firms in understanding the changing economic geography from the transportation
network’s perspective and maximize their benefits in the HSR-induced economic growth.

13
Chongqing has the largest population size (29 million) among all Chinese cities. Beijing has a population of 20 million, and Shanghai has a population of
23 million.

289
M. Diao Transportation Research Part A 113 (2018) 279–290

Acknowledgement

I would like to thank Yi Zhu and the anonymous referees for their valuable comments and inputs as well as Yaojie Tang and Qi
Zhao for their excellent research assistance. All errors remain mine.

References

Ahuja A., Chalk, F., Nabar, M., N’Diaye, P., Porte, N., 2012. An End to China’s Imbalances? IMF Working Paper.
Atach, J., Bateman, F., Haines, M., Margo, R.A., 2010. Did railroads induce or follow economic growth? Urbanization and population growth in the American Midwest,
1850–1860. Soc. Sci. History 34, 171–197.
Banerjee A., Duflo, E., Qian, N., 2012. On the Road: Access to Transportation Infrastructure and Economic Growth in China. NBER Working Paper No. 17897.
Banister, D., Berechman, Y., 2000. Transport Investment and Economic Development. UCL Press, London.
Baum-Snow, N., 2007. Did highways cause suburbanization? Quart. J. Econ. 122 (2), 775–805.
Baum-Snow, N., Brandt, L., Henderson, J.V., Turner, M.A., Zhang, Q., 2012. Roads, Railroads and Decentralization of Chinese Cities. Working Paper.
Chen, Z., Haynes, K., 2017. Impact of high-speed rail on regional economic disparity in China. J. Transp. Geogr. 65, 80–91.
Cheng, S., Stough, R.R., 2006. Location decisions of Japanese new manufacturing plants in China: a discrete-choice analysis. Ann. Region. Sci. 40 (2), 369–387.
Diao, M., 2015. Selectivity, spatial autocorrelation, and the valuation of transit accessibility. Urban Stud. 52 (1), 159–177.
Diao, M., Zhu, Y., Zhu, J., 2017a. Intra-city access to inter-city transport nodes: the implications of High-Speed-Rail station locations for the urban development of
Chinese cities. Urban Stud. 54 (10), 2249–2267.
Diao, M., Leonard, D., Sing, T.F., 2017b. Spatial-difference-in-differences models for impact of new mass rapid transit line on private housing values. Reg. Sci. Urban
Econ. 67, 64–77.
Donaldson, D., 2018. Railroads of the Raj: estimating the impact of transportation infrastructure. Am. Econ. Rev. 108 (4–5), 899–934.
Donaldson, D., Hornbeck, R., 2013. Railroads and American Economic Growth: A “market access” Approach. NBER Working Paper No. 19213.
Duranton, G., Turner, M.A., 2011. The fundamental law of road congestion: Evidence from US cities. Am. Econ. Rev. 101 (6), 1616–2652.
Duranton, G., Turner, M.A., 2012. Urban growth and transportation. Rev. Econ. Stud. 1, 1–36.
Faber, B., 2013. Trade Integration, Market Size, and Industrialization: Evidence from China’s National Trunk Highway System. Working Paper.
Fan, C., 1997. Uneven development and beyond: regional development theory in post-Mao China. Int. J. Urban Reg. Res. 21 (4), 620–639.
Givoni, M., 2006. Development and impact of the modern high-speed train: a review. Transp. Rev. 26 (5), 593–611.
Gutierrez, J., Urbano, P., 1996. Accessibility in European Union: the impact of the trans-european road network. J. Transp. Geogr. 4, 15–25.
Jiao, J., Wang, J., Jin, F., Dunford, M., 2014. Impacts on accessibility of China’s present and future HSR network. J. Transp. Geogr. 40, 123–132.
Michael, G., 2008. The effect of trade on the demand for skill: Evidence from the interstate highway system. Rev. Econ. Stat. 90 (4), 683–701.
Morris, J.M., Dumble, P.L., Wigan, M.R., 1978. Accessibility indicators for transport planning. Transp. Res. Part A 13 (2), 91–109.
Qin, Y., 2017. No county left behind? The distributional impact of high-speed rail upgrade in China. J. Econ. Geogr. 17 (3), 489–520.
Salzberg, A., Bullock, R.G., Fang, W., Jin, Y., 2013. High-Speed Rail, Regional Economics, and Urban Development in China. China Transport Topics. no. 8. World
Bank, Washington DC, http://documents.worldbank.org/curated/en/663511468220788389/High-speed-rail-regional-economics-and-urban-development-in-
China.
Sands, B., 1993. The development effects of high-speed rail stations and implications for Japan. Built Environ. 19 (3/4), 257–284.
Shaw, S., Fang, Z., Lu, S., Tao, R., 2014. Impacts of high speed rail on railroad network accessibility in China. J. Transp. Geogr. 40, 112–122.
Zheng, S., Kahn, M.E., 2013. China’s bullet trains facilitate market integration and mitigate the cost of mega city growth. Proc. Natl. Acad. Sci. U. S. A. 110 (14),
E1248–E1253.
Zhao, H., Zhu, G., 2000. Location factors and country-of-origin differences: an empirical analysis of FDI in China. Multinatl. Bus. Rev. 8 (1), 60–73.
Zhu, Y., Diao, M., 2016. The impacts of urban mass rapid transit lines on the density and mobility of high-income households: a case study of Singapore. Transp. Policy
51, 70–80.
Zhu, Y., Diao, M., Fu, G., 2016. Featured graphic: the evolution of accessibility surface of China in the High-Speed-Rail era. Environ. Plann. A 48 (11), 2108–2111.

290