Spatial spillover effects of transport infrastructure: evidence from Chinese regions

Journal of Transport Geography 28 (2013) 56–66

Contents lists available at SciVerse ScienceDirect

Journal of Transport Geography journal homepage: www.elsevier.com/locate/jtrangeo

Spatial spillover effects of transport infrastructure: evidence from Chinese regions Nannan Yu a,b,⇑, Martin de Jong a,b, Servaas Storm b, Jianing Mi a a b

Harbin Institute of Technology, West Dazhi Street 92, 150001 Harbin, China Delft University of Technology, Jaffalaan 5, 2628BX Delft, The Netherlands

a r t i c l e

i n f o

Keywords: Transport infrastructure Spillover effects Economic growth Spatial Durbin Model China

a b s t r a c t This paper examines the possibility of spatial spillover effects of transport infrastructure in Chinese regions. We estimate the regional spillovers of the transport infrastructure stock by applying a spatial Durbin Model for the time-period 1978–2009, and also three sub-periods, 1978–1990, 1991–2000 and 2001–2009. The results indicate that positive spillovers exist in each period due to the connectivity characteristic of transport infrastructure at the national level. At the regional level, transport infrastructure spillover effects vary considerably over time among China’s four macro-regions: the eastern region enjoyed positive spillovers all the time; the northeastern region had no significant spillover effects in 1978–1990, negative spillovers in 1991–2000, and positive spillovers in 2001–2009; the central region had negative spillovers for the three sub-periods; for the western region, negative spillovers can be observed after the 1990s. The analysis indicates that changes in spillovers among regions are closely associated with the migration of production factors in China during the last decades. Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction Plenty of studies have been conducted on the impact of transport infrastructure development on regional economic growth over the last decades, mostly aiming to examine the economic returns of transport investments in order to find a reasonable investment pattern (Aschauer, 1989; Munnell, 1990; Ozbay et al., 2003; Canning and Bennathan, 2007). Even though the range of the measured economic growth effects varies widely among studies, the positive relationship between transport investment and economic development is now commonly accepted (Banister and Berechman, 2001; Berechman et al., 2006). However the finding that the impact of transport infrastructure at the regional level is generally lower than the results observed at the national level leads some researchers to conclude that there exist significant spillover effects across regions. Subsequent research has tried to confirm this (Munnell, 1992; Holtz-Eakin, 1994; Cohen and Paul, 2004; Cantos et al., 2005; Berechman et al., 2006; Ozbay et al., 2007). Attempts have been made to corroborate the claim that the positive benefits accruing from these investments derive not only from investments made by individual states, but that there are also positive externalities from network expenditures made by neighboring states (Lall, ⇑ Corresponding author at: Harbin Institute of Technology, West Dazhi Street 92, 150001 Harbin, China. Tel.: +44 7453098837. E-mail addresses: [email protected] (N. Yu), [email protected] (M. de Jong), [email protected] (S. Storm), [email protected] (J. Mi). 0966-6923/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jtrangeo.2012.10.009

2007). That is because some effects induced by transport infrastructure will extend outside the limits of this area, generating spillover effects (Munnell, 1992; Boarnet, 1995, 1996, 1998). Only a few of studies have focused on ascertaining the possible existence of regional spillovers from transport capital, probably because it is difficult to find cases of countries in which its territory is divided in regions with substantial political power as the USA and Spain (Cantos et al., 2005). For the case of the USA, on the one hand, Munnell (1992) found that the impacts of highway capital became smaller as the geographic focus narrowed Thus she hypothesized that highway public capital can create positive cross-state spillovers because of productivity leakages (spillovers), since the transport infrastructure has network characteristics. But Holtz-Eakin and Schwartz (1995) rejected this argument after measuring the spillover effects separately. On the other hand, Boarnet (1995, 1996) hypothesized that public capital influences economic activity Iargely by shifting that activity from one location to another, and sees this claim confirmed in the case of the US street-andhighway capital. Considering these two arguments, Berechman et al. (2006) investigated the spillovers of transportation at the state, county and municipality levels of the USA respectively, and they concluded that the spillovers exist at the small geographic areas (at the municipality level) but that they cannot be found at the state and county levels. Ozbay et al. (2007) calculated the contribution of transport investments to county output using the data from the New York/New Jersey metropolitan area, and their results showed that the spillover effects decreased with distance from the

57

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66

investment location. For the case of Spain,1 the empirical findings show the variance: Cantos et al. (2005) confirmed the existence of very substantial spillovers; Álvarez et al. (2006) did not find evidence of the existence of spillover effects of public infrastructure using a panel data set of the 47 mainland Spanish provinces; Moreno and Lopez-Bazo (2007) investigated the economic returns to local and transport infrastructure and found negative spillovers across Spanish regions in transport capital investments. This variance of the results is probably because of the difference between studies in the definition of public capital and the econometric models. What’s worthy to note is that most empirical studies measure the spillover effects of transport infrastructure restricting attention to the first round neighbors for the purpose of a specification that is linear in the parameters (Boarnet, 1996; Berechman et al., 2006; Ozbay et al., 2007; Moreno and Lopez-Bazo, 2007). With the development of spatial econometric techniques (Anselin, 2001; LeSage and Pace, 2009), some advanced spatial models recently developed have been employed to capture the spatial externalities of infrastructure (including transport infrastructure) (Gomez-Antonio and Fingleton, 2009; Del Bo and Florio, 2012). Due to its huge size, China is disaggregated into many small administrative units—provincial (or municipality) governments,2 which have acquired substantial financial power after the Chinese fiscal decentralization, carried out in the early 1990s (Zhang et al., 2007). Each local government (provincial level) has the fiscal and political power to make decisions on the planning and investment of its transport construction, considering its individual interests. Hence, following fiscal decentralization, it is now possible to investigate for China whether regional transport infrastructure investment does have spillover effects – in terms of economic growth – on neighboring regions. In the case of China, only a small number of previous studies provide a separate analysis of spatial spillover endowments of transport infrastructures (Liu et al., 2007; Zhang, 2009; Liu, 2010; Hu and Liu, 2009). Liu et al. (2007) investigated the spillovers using the panel data for 11 cities of Zhejiang province and summarized that the highway infrastructure of other contiguous regions had positive spillover effects on local economic growth. Zhang (2009) estimated the spillover effects in the period 1993– 2004 and confirmed the existence of spillovers at the national level. Liu (2010) examined the contributions of highway and waterway infrastructure for different geographic levels, including both direct effects and externalities and suggested that the Chinese government should take the spatial correlation of investment impacts into consideration in its policy making regarding transport investment. Hu and Liu (2009) investigated the external spillover effects of transportation on China’s economic growth based on the spatial models and found the positive spillover with an elasticity of 0.06. These studies attempted to verify the existence of spillovers of transport facilities (or several types of transport infrastructure) in China and some of them indeed found empirical evidence of spillovers. However, most of these studies do not estimate spillover effects at the sub-national level, which would be more useful for the public decision making on the planning for large transport projects. This is why we propose this study, in order to obtain more detailed information of regional output productivity with respect to transport investment in a spatial framework, find changes in (the magnitude of) spillover effects over time, and try to interpret our findings in light of the actual situation in China.

1 For the case of Spain, several studies on the topic of cross-border spillover effects recently emerged. However, these papers adopted a methodology based on the accessibility calculation in a Geographic Information System support, which was not very related to our paper. Thus, we did not review this literature here. 2 The administrative hierarchy in China is: county–city–province (or municipality)–state. Since the financial reform in 1994, the provincial (and municipalities) governments have obtained the discretion over priority-setting in public investment.

Table 1 Transport system mileage in China. Source: The data is obtained from China Transportation Yearbook (1984–2010). Year

Roads (1000 km)

Railway (1000 km)

Waterways (1000 km)

Civil aviation (1000 km)

1950 1970 1980 1990 2000 2005 2009

99.65 636.74 883.31 1028.30 1402.79 3345.71 3860.21

22.2 43.79 52.98 57.83 68.70 75.48 85.56

73.64 148.42 108.53 109.27 119.37 123.31 123.75

8.22 42.50 231.38 506.82 1529.14 1998.52 2345.19

Our study aims to test for the presence of regional spillovers of transport capital and to measure their magnitude both in the country as a whole and in specific parts of China. Of particular emphasis in this paper is the regional difference in the spatial effects of transport infrastructure on economic growth. The structure of the paper is as follows. The next section gives a brief descriptive analysis of the expansion of the transport infrastructure network and the regional distribution of transport facilities in China. Section 3 introduces the methodology and database to quantify spatial spillovers of transport investment in the Chinese regions, and it also presents the results. To improve our understanding of the regional differences in spillover, a deeper analysis of the changes in spillover effects of transport infrastructure among Chinese regions will be presented in Section 4. The paper ends with conclusions and policy implications.

2. Transport infrastructures in China: an overview In the past decades, investment in transport infrastructure in China has seen remarkable growth in parallel with its booming economy. After 30 years of construction, all types of transport infrastructure have seen significant expansion as shown in Table 1. In the past six decades the transport network in China has begun to take shape. The patterns of the current railway and highway networks in China in 2009 are presented in Fig. 1. In 2009, the total length of the Chinese railway network reached 103.16 thousand kilometers. A government official from the Ministry of Railways, Mr. Liu Zhijun,3 has stated that, in the long-term plan for Chinese railways, the total railway mileages will increase to 120 thousand kilometers, including 16 thousand kilometers of high-speed railways in 2015. As to the highways, the investment in the highway construction was as high as RMB 623.11 billion yuan (about $93 billion dollars) in 2009 and kept a high growth rate from 1978, above 10% per year. The total mileage of expressways was 45 thousand kilometers in 2009, which was an 80% increase compared with the length in 2002. The central government allocates its investment budget mostly to those transport facilities, the construction of which is likely to generate high economic returns, such as toll roads, ports and inter-city high-speed rail between high-density metropolitan areas. However, because regional Chinese administrative units have their own discretion with respect to the distribution of public investment, local governments make the investment decision in view of their individual economic growth and (often) neglect the (spillover) impact of their investments on the neighboring areas. As a result, there is considerable underinvestment in the connecting highways (State Roads and Provincial Roads) and rural roads, which have low economic returns but high social returns. 3 Mr. Liu Zhijun was the head of Ministry of Railway Transportation in China during the period of 2003–2011, which is independent from the Ministry of Transportation.

58

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66

Fig. 1. China’s railway network (left) and highway network (right) of China in 2009. Source: The information is collected from the official websites of Ministry of Railways and Ministry of Transport of China, 2010. Available on line: http://www.china-mor.gov.cn/, http://www.moc.gov.cn/.

Meanwhile, as we have discussed in our previous paper (Yu et al., 2012), there exists a wide variety in transport infrastructure facilities among Chinese regions4 (shown in Fig. 2), which clearly appears as stair steps decreasing gradually from eastern China to western China. Most coastal provinces are well endowed with a high quantity and quality of transport facilities, whereas the transport network density remains very low in remote western provinces. What is therefore clear is that the transport network has expanded considerably since the establishment of modern China in 1949, and because of network characteristics and spatial clustering of transport infrastructures we argue that it is important and necessary to consider spatial factors when estimating the potential economic benefits of transport facilities. Could the transport facilities yield more economic benefits than just their direct regional effects on the region alone? How do the spillovers change over time at the sub-national level if the existence of spillovers can be verified? To answer these questions, we will examine the spillover effects at the national and regional level in the next section. 3. Measuring spatial spillover effects of transport infrastructure in China 3.1. Model specification and data collection 3.1.1. Model specification On the subject of impacts with respect to transport infrastructure, most of the previous studies were conducted within the framework of a Cobb–Douglas (C–D) production function (Boarnet, 1996, 1998; Holtz-Eakin and Schwartz, 1995; Hu and Liu, 2009; Del Bo and Florio, 2012). Therefore, the baseline empirical model is constructed in framework of production function as:

Y ¼ f ðL; Kc; Kg; TIÞ

ð1Þ

where Y denotes output, L denotes labor input, Kc denotes private sector capital stock, Kg represents pubic capital stock (except for 4 In our paper, China is divided into four macro regions, based on their level of economic development and geographic position: the eastern region, northeastern region, central region and western region.

the transport capital stock) and TI stands for the transport infrastructure capital stock. As usual, in the log-linearized reduced version, the estimated parameters can be thought of as GDP elasticities to each regressor:

ln Y ¼ b0 þ b1 ln L þ b2 ln Kc þ b3 ln Kg þ b4 ln TI þ e

ð2Þ

5

Given that Moran’s I statistic suggested that the data were affected by spatial autocorrelation, it is necessary for our study to consider the role of a change in own and neighboring explanatory and dependent variables by an appropriate spatial econometric model (Anselin, 2001). Based on LeSage and Pace (2009), a general spatial model, Spatial Durbin Model (SDM)6 could be considered for our empirical analysis:

y ¼ qWy þ Xb þ hWX þ aln þ e

ð3Þ

where q is the spatial autocorrelation coefficient, W is the spatial weight matrix, X is the matrix of control variables (including labor, private capital, public capital and transport infrastructure), ln denotes an n  1 vector of ones, a, h and b are vectors of regression coefficient estimates, and e is the error term. The SDM includes a spatial lag of the dependent variable (Wy) as well as spatial lagged explanatory variables (WX). An implication of this is that a change in the dependent variable for a single region may affect the dependent variable in all other regions by the network effect; meanwhile a change in the explanatory variable for a single observation can potentially affect the dependent variable in all other observations. 5 Here we have used the standard Moran’s index (Moran’s I), as an indicator of spatial autocorrelation. The results of the spatial autocorrelation analysis indicate that there is significant spatial autocorrelation in these models, thus it is necessary to introduce the spatial factors when we calculate the contributions of transport infrastructure on the regional economic growth. The details of calculation process are available from the authors. 6 The SDM is a general spatial model, which, in a restricted form, can be interpreted as a spatial autoregressive model (SAR) or spatial error model (SEM). The choice of this unconstrained specification was driven by LM tests and LR tests. Our result shows that the SDM is to be adopted against SAR and SEM. A Hausman test is calculated to select between fixed and random effects, and the value is -80.95, thus the random effects is more suitable for our spatial panel model at the national level. We also did the calculation process for each small panel, however, we just provide the final results because of the words limitation. But the details are available from the authors.

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66

59

Fig. 2. Four macro regions in China.

Combining Eqs. (2) and (3), our key empirical model for the estimation of spatial spillover effects in China can be constructed as:

ln Y i;t ¼ q

N X wij ln Y j;t þ b0 þ b1 ln Li;t þ b2 ln Kci;t þ b3 ln Kg i;t j¼1

þ b4 ln TIi;t þ h1

N N X X wij ln Lj;t þ h2 wij ln Kcj;t j¼1

þ h3

N X j¼1

wij ln Kg j;t þ h4

j¼1 N X

wij ln TIj;t þ ei;t

j¼1

ð4Þ

where Y is real gross domestic product; i and t are the indices of province and year respectively; j represents nearby provinces (j – i); wij is each of the elements in the spatial weights matrix W that describes the spatial arrangement of the different regions, and other variables are defined as before. In a SDM context, the regional variation in GDP levels is modeled to depend on the GDP levels from the neighboring provinces captured by the spatial lag vector Wy, as well as the factors input (including L, Kc, Kg and TI) of neighboring provinces represented by WX. In this study, a binary contiguity matrix (wbin) is used to construct the spatial weighted matrix (wij), which assumes only contiguous provinces can influence each other:

 wij ¼

1 if the province i has a border with province j 0 otherwise

P and Nj¼1 wij ¼ 1. Here we choose the ‘provincial borders’ to define the ‘spatial geographic unit’ for this study, because they are the containers of the data we need for our calculations, while the governments in these units (provinces) surrounded by the provincial borders have the power to decide on public investments, which is essential for the policy application of our empirical findings. For wbin, we can get a symmetric spatial matrix of the 29 Chinese provinces,7 as Fig. 3 shows. In order to better evaluate spatial spillovers, following LeSage and Pace (2009), the direct, indirect and total impacts can be calculated based on the estimators of SDM. These measures capture the accumulative effect in the Chinese regions of changes in the independent variables (including transport infrastructure), which lead to a change in the long-run steady-state equilibrium. The purpose was to verify whether the positive effect of an increase in a region’s transport infrastructure is accompanied by a negative spillover effect from other regions. What’s worthy to note is that the direct effect of independent variables is different with the coefficient b, Fig. 3. Symmetric spatial matrix of the 29 Chinese provinces.

7

We give the explanation of the provinces selection in the data collection part.

60

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66

since b also contains the feedback effect, representing the effect of the impacts spilling over to the neighboring regions and back to the region itself. According to LeSage and Pace (2009) and Del Bo and Florio (2012), the SDM specification contains spatially lagged values of both the dependent and the explanatory variables. LeSage and Pace (2009) provided the theoretical framework to interpret these direct and indirect effects, by transforming the spatial weight matrix and by considering the role of off and on diagonal elements. Formally, the SDM can be re-written as:

ðIn  qWÞy ¼ Xb þ WXh þ ln a þ e y¼

ð5Þ

k X Sr ðWÞxr þ VðWÞln a þ VðWÞe

ð6Þ

r¼1

where Sr ðWÞ ¼ VðWÞðIn br þ Whr Þ and VðWÞ ¼ ðIn  qWÞ1 . If we expand Eq. (6) from one region to n regions and transfer to the matrix form, we can get:

2

y1 y2  yn

¼

Sr ðWÞ11

k 6 X 6 Sr ðWÞ21 6 .. 6 . r¼1 4

Sr ðWÞn1

Sr ðWÞ12 Sr ðWÞ22 .. . 

   Sr ðWÞ1n

32

x1r

3

7 6    Sr ðWÞ2n 7 76 x2r 7 76 . 7 þ VðWÞe .. .. 76 . 7 54 . 5 . . xnr    Sr ðWÞnn

2010). The spillover effects to sub-national growth in China will be estimated with data of Chinese provincial governments except Hainan province, which is a separate island without any geographic neighbors. The data of Chongqing is combined with those of Sichuan province in this paper, because Chongqing used to be a city in Sichuan province until 1997. Some investment data during 1966–1974 are unavailable from official sources due to political issues during 1974–1978, and consequently we use data from a panel of 29 Chinese provinces for the period 1978–2009 for which data is available on real GDP, private sector investment, employed population (labor input), transport infrastructure investment and public investment. The separate investment data in transport infrastructure cannot be found in various sources and we have to adopt the data on ‘‘investment in transport infrastructure and postal service’’ from the Statistical Yearbook of provinces and municipalities instead of the transport investment. We calculate the transport infrastructure capital stock, private sector capital stock and public capital stock based on investment data, according to the perpetual inventory method (Goldsmith, 1951).8 3.2. Results and discussion

ð7Þ Here, average direct impacts can be obtained as the average of the diagonal elements of matrix Sr(W), the average total impacts could be calculated by averaging over all regions the sum of the rows (or columns) of matrix Sr(W) and average indirect impacts (spillover effects) were obtained as a difference between the total and direct impacts. Formally: 0

MðrÞtotal ¼ n1 ln Sr ðWÞln MðrÞdirect ¼ n1 trðSr ðWÞÞ MðrÞindirect ¼ MðrÞtotal  MðrÞdirect In this way, the sign and magnitude of direct and indirect impacts of the explanative variables can be calculated. Our methodology differs from the previous studies in two ways: firstly, our model could consider the spillovers from all the regions, not limited to the first round neighbors as the previous literatures did (Boarnet, 1996; Berechman et al., 2006; Ozbay et al., 2007; Moreno and Lopez-Bazo, 2007); secondly, our study uses a more general spatial specification developed recently, both considering the spatial autoregressive model and spatial error model, which could provide a more complete and accurate picture of the spillover effects than the existed studies (Hu and Liu, 2009; Zhang and Yi, 2012), especially with respect to transport infrastructure. To deal with the endogeneity in our model, we apply Maximum Likehood (ML) procedures in estimating spatial panel data models as implemented in the MATLAB. The spatial panel model and the direct and indirect effect can be computed by the spatial econometrics library for MATLAB provided by LeSage. Although the Generalized Method of Moments (GMMs) could be regarded as an alternative, GMM usually has to include spatially lagged independent variables, a requirement that would not allow us to test the influence of spatial spillovers (Del Bo and Florio, 2012). 3.1.2. Data collection The data used in this research are collected from a number of different official Chinese sources, including the China Statistical Yearbook (1982–2010), the Statistical Yearbook of Chinese provinces, municipalities and PRC’s Statistical Series of 60 Years (SSB,

In order to compare the changes of the spillover effects over time, we also ran the spillover effects model (Eq. (4)) for three sub-periods, 1978–1990, 1991–2000 and 2001–2009, respectively. The key results at the national and regional levels are presented in Tables 2–4. The spatial autocorrelation coefficient q is positive and significant in all panels, indicating that the Chinese provinces are characterized by a positive and significant level of spatial correlation, with an estimated coefficient value ranging from 0.15 to 0.39. 3.2.1. Spillover effects at the national level Table 2 reports the results of the estimation of the SDM, and we can find that the coefficients of the labor, private capital, public capital and transport infrastructure are positive and significant. In terms of the spatial lagged independent variables, the national output is a positive function of private capital, public capital and transport infrastructure endowment in the neighboring provinces, while the spillover effects of labor is negative, but not significant. This finding is in line with Zhang and Yi (2012), which concluded that the spillovers of labor in China were mainly limited at the municipality and county level because of its huge distance between areas. However, these estimators just provide an idea of interactions among provinces, thus we provide the sign and magnitude of the direct and indirect impacts in order to provide the accurate spillover effects, especially associated with transport infrastructure in Table 3. Our empirical findings show that total effects of private capital and public capital variables have positive signs and are significant at the 1% level for the entire observation period. Moreover, these two capitals have a similar magnitude of output elasticity (the coefficients are 0.24 and 0.22), which means that private capital and public capital have almost the same contribution to the output. In our view, this could be because of the fact that after three decades of economic reform, the market factors have played an essential role in the Chinese economy, even though the

8 Private sector capital stock was computed using perpetual inventory method as follows:, where K is the private capital stock in year t; I is the real private investment in fixed assets in constant 1978 prices; d is the depreciation rate (we assume the depreciation rate of the private capital is 9.6% according to Zhang et al. (2004)). The same method was applied to the calculation of other infrastructure capital stock and transport capital stock. The process of the calculation is not provided here due to the word limitation, but available from the authors.

61

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66 Table 2 Estimation results of SDM at the national level. Variable Constant L Kc Kg TI

q WL W  Kc W  Kg W  TI Adj. R2 Log likelihood

1978–2009

Period1 ***

1.203(9.23) 0.572(19.89)*** 0.141(12.46)*** 0.186(13.59)*** 0.114(16.53)*** 0.231(4.25)*** 0.283(1.65) 0.073(6.37)*** 0.022(11.24)*** 0.045(7.32)*** 0.796 177.42

Period2 ***

Period3 ***

0.691(6.92) 0.471(14.39)*** 0.069(17.04)*** 0.259(13.37)*** 0.256(15.61)*** 0.307(3.16)*** 0.156(0.31) 0.046(2.32)** 0.002(5.16)*** 0.019(12.49)*** 0.519 153.43

1.304(4.20)*** 0.597(15.58)*** 0.148(15.16)*** 0.124(12.23)*** 0.036(5.36)*** 0.264(8.53)*** 0.176(1.32) 0.082(10.24)*** 0.047(2.27)** 0.072(13.24)*** 0.764 148.35

1.521(14.24) 0.566(13.52)*** 0.100(13.62)*** 0.161(13.38)*** 0.170(15.35)*** 0.279(13.01)*** 0.188(1.37) 0.061(7.57)*** 0.030(9.23)*** 0.041(7.35)*** 0.877 164.36

Note: t-statistics are in parentheses. Numbers of observations equals to numbers of years in each period multiplied by 29 provinces. Statistical significance at the 10% level. Statistical significance at the 5% level. *** Statistical significance at the 1% level. 

**

Table 3 The direct and indirect effects of explanative variables. Variables

1978–2009

Period 1 

Period 2

Period 3

Labor

Direct effect Indirect effect Total

0.554(23.14) 0.123(1.26) 0.531(9.45)

0.465(14.67) 0.114(1.24) 0.451(14.43)

0.537(6.36) 0.144(0.37) 0.493(4.75)

0.585(26.41) 0.140(1.47) 0.545(13.28)

Private capital

Direct effect Indirect effect Total

0.149(6.03) 0.087(15.35) 0.236(21.75)

0.074(14.65) 0.061(9.31) 0.135(4.67)

0.103(21.24) 0.079(4.25) 0.182(10.46)

0.154(14.37) 0.100(6.32) 0.254(13.46)

Public capital

Direct effect Indirect effect Total

0.192(19.24) 0.031(7.34) 0.223(13.04)

0.261(3.65) 0.003(1.97) 0.264(11.69)

0.166(11.46) 0.037(8.24) 0.203(12.65)

0.129(2.23) 0.054(7.25) 0.183(10.53)

Transport infrastructure

Direct effect Indirect effect

0.119(17.21) 0.054(15.17)

0.258(13.46) 0.027(13.01)

0.173(19.47) 0.051 (2.49)

0.036(1.99) 0.084(12.16)

Total

0.173(12.36)

0.285(26.35)

0.224(2.16)

0.120(7.47)

governments could still partly control the allocation of social resources through economic policies. The results also provide a reasonable estimate for the labor coefficient (0.53), which indicates that labor input growth has the largest impact on Chinese real GDP growth; the coefficient value is very much in line with findings from earlier studies and growth-accounting analyses for Asia (Zhang, 2009; Sahoo et al., 2010). Considering the impact of transport infrastructure, we find that transport infrastructure has a positive total impact on national growth (the coefficient is 0.17), but this impact decreases over time since the direct impact declines during the different periods (the coefficients are 0.26, 0.17 and 0.04 in the periods 1978–1990, 1991–2000 and 2001–2009, respectively). The impact of transport infrastructure declines over time, which may be because since the economic reform, investment in transport projects has continuously increased, and after some time the marginal returns began to decline. These empirical findings are mainly in line with previous studies for the case of China (Zhang, 2009; Liu, 2010), but providing some difference in the magnitude of elasticities of these inputs since our study has been unfolded in a spatial context, assuming both of the productivity and factor inputs in the neighboring provinces could affect the local economy. For China, as Table 3 shows, the spillover effect (indirect effect) of transport infrastructure is 0.05 (the coefficient is 0.05 and statistically significant), which means transport stock does not only contribute to GDP directly but also indirectly through regional spillover effects. This finding is consistent with Zhang (2009) and Hu and Liu (2009). Meanwhile, we find that these spatial spillovers are significantly positive in each period and increase over time: the coefficients are 0.03 for the period 1978–1990, 0.05 for 1991–2000,





and 0.08 for the years 2001–2009, which are statistically different based on a t-test.9 This finding implies that the spillover effects played a more and more important part in promoting economic growth (the coefficients of spillovers increase over time) because of transport network expansion. This expansion helps to reduce transportation costs among regions and also brings indirect social externalities due to the improvement of transport network accessibility. The declining transportation cost is propitious to enlarge the domestic market and to facilitate the development of foreign trade, which could stimulate economic growth (Mao and Sheng, 2011). 3.2.2. Spillover effects at the regional level Focusing mainly on the indirect effect of transport infrastructure (represented by l), as can be seen clearly in Table 4, the elasticities of the spillovers vary considerably among regions in the entire period under study (the coefficients are 0.14, 0.04, 0.05 and 0.06 for the eastern, northeastern, central and western regions, respectively). The neighboring transport investment will lead to positive effects in the eastern region, and the output elasticity is very high, 0.14, which means the GDP of the eastern region will increase by 0.14% if the transport stock in the neighboring region increases by 1%. The spillover effect in the northeastern region is also significant and positive (the coefficient is 0.04), but statistically lower than the one of the eastern region.10 However, for the 9 The results from the t-test indicate that the elasticities of transport infrastructure in various periods are statistically different. 10 The t-test results verify that the spillovers in the eastern region and the northeastern region are statistically different.

62

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66

Table 4 Estimation results of SDM at the regional level (East, Northeast, Central and West). Regions Eastern region

Variables L Kc Kg TI

q WL W  Kc W  Kg W  TI

l Adj. R2 Log likelihood Northeastern region

L Kc Kg TI

q WL W  Kc W  Kg W  TI

l Adj. R2 Log likelihood Central region

L Kc Kg TI

q WL W  Kc W  Kg W  TI

l Adj. R2 Log likelihood Western region

L Kc Kg TI

q WL W  Kc W  Kg W  TI

l Adj. R2 Log likelihood

1978–2009

Period 1 ***

Period 2 **

Period 3 **

0.616(15.61) 0.167(20.05)*** 0.132(16.36)*** 0.091(2.25)** 0.394(13.54)*** 1.423(1.36) 0.101(15.21)*** 0.031(3.25)*** 0.124(13.56)*** 0.141(16.32)*** 0.645 149.23

0.495(13.12) 0.092(21.14)*** 0.136(10.42)*** 0.203(2.59)** 0.325(6.47)*** 1.065(1.44) 0.093(12.21)*** 0.023(1.92)* 0.035(16.34)*** 0.062(6.87)*** 0.832 136.65

0.597(11.90) 0.150(17.98)*** 0.139(18.35)*** 0.100(2.34)** 0.293(15.35)*** 1.710(0.79) 0.064(1.99)* 0.021(7.37)*** 0.151(14.74)*** 0.167 (7.12)*** 0.535 104.36

0.668(14.32)*** 0.283(15.45)*** 0.085(19.31)*** 0.088(0.41) 0.343(21.35)*** 1.431(1.22) 0.125(2.30)** 0.035(12.53)*** 0.103(8.42)*** 0.139(14.55)*** 0.677 157.64

0.537(12.09)*** 0.099(10.02)*** 0.179(6.63)*** 0.141(4.03)*** 0.214(24.68)*** 0.194(0.47) 0.061(14.62)*** 0.056(8.49)*** 0.021(12.42)*** 0.039(9.26)*** 0.783 178.57

0.575(16.03) *** 0.051(1.98)* 0.214(10.35)*** 0.221(21.56)*** 0.269(3.74)*** 0.114(2.62)** 0.053(1.87)* 0.046(0.91) 0.121(0.64) 0.106(0.47) 0.832 103.24

0.506(2.93)** 0.108(7.61)*** 0.166(2.34)** 0.193(1.86)* 0.196(12.71)*** 0.107(0.74) 0.104(0.39) 0.055(6.74)*** 0.074(2.36)** 0.057(2.14)* 0.663 121.01

0.563(10.31)*** 0.109(9.05)*** 0.173(15.29)*** 0.110(2.90)** 0.286(6.74)*** 0.083(1.29) 0.086(16.31)*** 0.082(24.26)*** 0.022(2.50)** 0.030(2.79)** 0.675 177.36

0.535(6.36)*** 0.105(11.20)*** 0.169(10.84)*** 0.194(15.75)*** 0.262(5.32)*** 0.097(1.09) 0.054(16.32)*** 0.037(6.84)*** 0.071(16.42)*** 0.054(8.09)*** 0.705 132.53

0.566(1.86)* 0.058(2.33)** 0.181(10.24)*** 0.171(2.25)** 0.305(7.15)*** 0.106(0.73) 0.062(6.31)*** 0.071(1.04) 0.033(7.58)*** 0.015(6.23)*** 0.635 144.89

0.521(2.03)* 0.081(9.71)*** 0.200(9.45)*** 0.163(17.34)*** 0.192(15.26)*** 0.093(1.37) 0.056(14.28)*** 0.102(2.35)** 0.134(1.89)* 0.122(2.21)** 0.463 161.26

0.573(2.74)** 0.104(7.89)*** 0.143(6.36)*** 0.209(10.60)*** 0.254(3.86)*** 0.165(1.93)* 0.094(14.63)*** 0.089(1.42) 0.075(2.61)*** 0.050(2.28)** 0.562 105.85

0.462(14.23)*** 0.126(12.63)*** 0.228(13.25)*** 0.073(13.02)*** 0.235(7.47)*** 0.0521(1.36) 0.037(13.63)*** 0.021(5.17)*** 0.084(4.25)*** 0.061 (2.22)** 0.547 162.54

0.403(32.90) 0.080(2.89)** 0.193(15.36)*** 0.082(2.77)** 0.169(16.73)*** 0.131(0.14) 0.030(5.62)*** 0.010(1.98)* 0.107(0.16) 0.070(0.88) 0.495 181.26

0.345(2.78)** 0.061(10.67)*** 0.250(2.35)** 0.081(2.06)* 0.145(21.63)*** 0.067(2.20)** 0.051(2.36)** 0.043(2.04)* 0.135(8.47)*** 0.102(10.15)*** 0.537 165.52

0.387(2.48)** 0.133(21.47)*** 0.201(12.35)*** 0.029(0.65) 0.271(15.36)*** 0.133(1.27) 0.036(4.74)*** 0.039(2.59)** 0.092(2.43)** 0.059(2.44)** 0.512 174.27

Note: t-statistics are given in parenthesis. Period 1, Period 2 and Period 3 represent 1978–1990, 1990–2000 and 2001–2009, respectively. Numbers of observations equals to numbers of provinces in each region multiplied by analysis period. Here, we calculated and reported the indirect effect (spillover effects) of transport infrastructure for each region in different periods, represented by l. * Statistical significance at the 10% level. ** Statistical significance at the 5% level. *** Statistical significance at the 1% level.

central and western regions, the increase of investment in neighboring transport infrastructure may hold back the local economy (the coefficients have a negative sign). When we compare our results for the three sub-periods, we can see that the changes in spillovers vary considerably among these regions: (1) For the eastern region, the transport stock in the neighboring region has a positive external impact during the whole period. The regression results illustrate that the output elasticities of neighboring transport infrastructures for the three sub-periods are significant and positive (the coefficients are 0.06, 0.17 and 0.14).

(2) For the northeastern region, no significant spillovers can be found in period 1, but negative spillovers can be observed in the second period (the coefficient is 0.06). In the last period, positive externalities can be found (the coefficient is 0.03). (3) In the central region, the estimated coefficients of spillovers are 0.02 during 1978–1990, 0.12 during 1991–2000, 0.05 during 2001–2009, which means that the growth of the transport stock in neighboring regions actually had a negative impact on economic growth in the central region all the time. (4) For the western region, the negative spillovers can be captured in the last two sub-periods (the coefficients are

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66

0.10 and 0.06). We do not find spillover effects for the first period: according to our estimations, the estimated coefficients are not statistically significant in period 1. Our empirical findings are partly in line with the previous studies (Zhang, 2009; Liu, 2010; Hu and Liu, 2009) at the national level, but show some contradictory results comparing to Liu (2010) at the regional level. That may be because: (1) Our paper adopted an advanced spatial Durbin Model, considering both the spatial lagged dependent and independent variables; meanwhile the spatial spillovers from all the regions were measured in our study, which could make our estimators are more accurate and convincing. (2) Only the highway and waterway capital stock have been considered in his paper, while we adopt a broader selection of transport infrastructure data, including railway and aviation investment. We believe the railway constructions have been emphasized by the central government in recent years (several large high speed train projects). Railway networks are supposed to have significant spillover effects. Thus, the incorporation of railway investment data may yield changes in the results. (3) The different definitions of regions may also cause the conflicting results. In order to underline the spatial factors, four macro regions are classified considering the geographic position, instead of the traditional classification (east, central and west regions) according to the economic development level, which would make our estimate results of the spatial spillovers more realistic. To summarize, the empirical results from this study confirm the existence of spillover effects of transport infrastructure for the case of China. More specifically, changes in the spillovers between Chinese regions over time can be observed. For the purpose of an indepth analysis on the regional difference in spatial spillovers, we will next investigate how the spillovers of transport infrastructure work in China. 4. Analysis on the changes of spillover effects among Chinese regions Different from the previous studies on the estimation of transport spillover effects from a macro view (Zhang, 2009; Liu et al., 2007), we analyze the sources of spillovers in this paper: two types of spillover effects can be classified. On the one hand, positive spillovers can be caused by productivity leakages because of the connectivity characteristics of transport facilities (Munnell, 1992). On the other hand, spillovers of transport infrastructure may arise from the migration of production factors: mobile factors of production migrate to places with better transport stock. That migration results in output gains in places with well-developed transport capital stocks and output losses elsewhere, which has been theoretically verified in previous studies (Boarnet, 1996, 1998; Moreno and Lopez-Bazo, 2007). Thus, in our study, we try to explain the changes in spillover effects by understanding the nature of the spillovers, interpreting our empirical findings according to the actual situation in China. 4.1. Spillover effects arising from network characteristics According to Banister and Berechman (2001), the increase in transport investment in one region could improve the network accessibility of this region and therefore enlarge its market scale. Adam Smith proposed the ‘extent of the market’ hypothesis: as the size of the market expands, this makes possible a greater division of labor and hence specialization, and this in turn would allow the economy to expand further and the growth in output and productivity would cumulate. Krugman (1991) and Fujita et al. (1999) re-formulated Smith’s argument from a viewpoint of external

63

economies of scale and increasing returns to scale. Thus, we can argue that the transport network expansion could stimulate the economy in both the area where the investment happens and the neighboring areas because of the growing market. In other words, the positive network spillover effects could occur when infrastructure investments in one state benefit people in other states through the transport network (Munnell, 1992). For the case of China, Mao and Sheng (2011) concluded that both economic opening and regional integration have demonstrably positive effects on China’s provincial TFP (total factor productivity); Huang and Li (2006) found that the market scale expansion had a positive impact on economic growth adopting the New Economic Geography methodology. So, we can argue that for China, the rapid enlargement of the market scale induced by transport improvement brings many economic benefits. Thus, it is reasonable that we find the existence of positive spillovers in China at the national level for the entire period under study and for different sub-periods since the transport network could facilitate the productivity leakages among provinces. However Chinese regions show that transport stock spillovers change over timeperiods, probably because of negative spillovers from factor migration, which may counteract positive externalities from market size changes in some sub-state areas. 4.2. Spillover effects arising from mobile factors Spillovers from factor migration are positive for regions of destination, but negative for regions of origin (Boarnet, 1996, 1998). Changes in inter-regional migration flows play an essential role in explaining the changes of spillover effects among regions. In the period of 1978–1990, the centralized decision-making structure still applied to spatial distribution of the capitals since the economic reform just started, and factor migration was limited. In the second period (1991–2000), China started to implement a market-oriented economy and the exchange of production factors and commodities grew much easier because of lower transport costs and enhanced accessibility. The production factors began to transfer from the poor West, and the intermediate Northeast and Center to the well-developed East. In 2000s, the eastern region has witnessed a dramatic development and its productivity overflow was expected to benefit the other regions by the way of industrial redistribution. To understand what has happened, we consider how the migration of production factors has changed over time across the four regions in Table 5. The table reports the migration of production factors among regions in different periods. From Table 5, we can see the total trend of the net migration value of production factors (capital and labor). Obviously, the central region (the connecting areas) is the victim of those migrations in 1978–2000, when a substantial outward shift of production activities to the eastern region occurred. Remarkably, the northeastern region and western region do not lose as we expected, possibly because of their substantial geographic distance from the well developed eastern region. In the last decade, the production factors have started to shift from the well-developed eastern region to the under-developed regions. A large amount of capital shifted to the western region possibly because of the ‘Western Development Strategy’ started in 1999. Since then the central government has invested a lot in the West and also provided favorable policies to attract other outside investments to that region. These findings seem totally in line with the actual situation we analyzed before. 4.3. A saldo matrix for the spillover change in different periods In view of the two sources of spillovers of transport infrastructure, we will construct a balanced matrix to show what happened in Chinese regions. Here, we suppose that transport spillovers

64

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66

Table 5 The migration of production factors among regions in various periods. East

Northeast

Center

West

Capital

Labor

Trend

Capital

Labor

Trend

Capital

Labor

Trend

Capital

Labor

Trend

Period 1 1978–1990

+330

+391

+

39

32



238

175



52

78



Period 2 1991–2000

+7536

+2222

++

1763

383



3735

1331



2200

503



Period 3 2001–2008

712

239



+210

+68

+

+287

+97

+

+455

+65

+

Note: ‘+’ means shift-in and ‘’ means shift-out. The data of capital migration are collected from Peng (2008) (unit is RMB one hundred million yuan). The data of labor migration are gathered from Wang et al. (2010) (unit is ten thousand people) and the National Statistical Compilation of Transient population (Ministry of Public Security, 2010).

Table 6 Spillover changes of Chinese regions in different periods. Spillovers

Source 1 Source 2 Saldo

East

Northeast

Center

West

Period1

Period2

Period3

Period1

Period2

Period3

Period1

Period2

Period3

Period1

Period2

Period3

+ + ++

+ ++ +++

+  0

+  0

+  

+ + ++

+  0

+  

+ + ++

+  0

+  

+ + ++

arising from mobile production factors show the same trend as factor migration, which has been theoretically verified in Moreno and Lopez-Bazo (2007). Meanwhile, we assume that the spillovers hinging on the network characteristics of transport facilities are always positive (+) since our empirical findings show the autoregressive coefficient (q) is positive and significant for each case. Consequently, we can construct a saldo matrix for the spillover change in different periods in these four regions, as shown in Table 6. In 1978–2000, the eastern region has undergone a great capital and labor influx, and the spillovers there from factor migration have been positive. Thus, the saldo matrix shows that the two types of effects are of the same sign, which can explain why the spillover effects from the regression findings are positive all the time. In period 3, our empirical findings show that the spillovers lightly decline but remain positive, which conflicts with the saldo matrix (the spillover should be zero). For the northeastern region, from a balance perspective, the saldo matrix shows that in period 1, the negative spillover accruing from factor migration from the central region to the eastern region counteracted the positive spillover caused by the connectivity characteristic of transport facilities; in period 2, the negative spillover also exceeded the positive one and the saldo remained negative; in period 3, these two types of spillover have the same sign (positive), therefore a positive spillover was found. That is because in the last period, the equipment manufacturing industry has concentrated in the northeastern provinces because of lower transport cost (Liu et al., 2011). These findings can help explain the changes in spillover effects with respect to transport infrastructure from our empirical estimate results. For the central and western regions, the saldo shows the same trend as for the northeastern region. But in the period of 2001– 2009, our empirical results indicate that negative spillovers exist in the central and western region. However we can see that the saldo of spillovers should be positive in the last period, which contradicts the data. Indeed, in the last decade, the improvement in the transport network did not result in industrial expansion in the eastern region, which may disappoint some government officials who expect that transport infrastructure construction could reduce the gap between regions by realizing the ‘industrial gradient transfer’. As the previous literature (Krugman, 1991; Fujita et al., 1999; Banister

and Berechman, 2001) pointed out, in cases where labor migration is limited, the price of labor would increase with the agglomeration of industries (economic activities), and therefore increase the production cost. Some enterprises may choose to relocate to peripheral areas if production costs exceed savings in exchange costs. In this way, the peripheral regions may benefit from the productivity overflow from the core region. However in China, the gradual industrial redistribution may not yet have happened because of a seemingly endless supply of cheap manual labor. China has a very substantial rural labor-surplus, about 120 million persons in 2009 (Yi and Ying, 2011). Meanwhile, apart from the transport cost and production cost, the economic policies (such as tax policy) played a key part in the industrial redistribution in China. Thus, in the last period, the technology-intensive industries still moved to the eastern region (Liu et al., 2011) due to agglomeration effects, caused by lower transport costs. The in-migration of economic activities in the last period in the central and western regions (visible in Table 5) occurs because some resource-intensive industries in the eastern region depend on resource exploitation in the central and western regions. Consequently these industries, such as agriculture, petroleum and coal extraction, and metal smelting, have to move out of the eastern region into the resource-rich regions in order to meet the growing demand of raw materials for the export industries in the last decade (Liu et al., 2011). But this type of industrial redistribution (production factor migration) happened in the central and western regions had nothing to do with transport costs. That is why we can see the positive sign in the last period from our saldo but negative spillovers can be found from our empirical study, in the central and western regions. For the same reason, the economic activities (production factors) migrated from the eastern part in the last period, but it was not because of the transport network. The factor migration caused by the transport infrastructure still went into the eastern coastal provinces from the other regions. Thus, it makes sense that we can still find positive spillovers in the eastern region over the 2001–2009 period. 5. Conclusion and policy implications Much of the evidence on transport infrastructure spillovers has been reported for the states and counties in the developed

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66

countries, such as United States and Spain, where there may be no severe lack of infrastructure endowment. Here, in contrast, we provide evidence on the spillover effect of transport stock in the Chinese provinces, some of which were characterized by a low level of economic development and also by a short supply of transport infrastructure in most of the periods under analysis. Therefore, some lessons for emerging economies, which are also having a large working population, can be derived from our results. Based on this study, transport capital is associated with increased output within a region, positive network spillovers, and negative (or positive) output spillovers. The positive spillovers exist at the all-China level, but the Chinese regions have considerable variance in their spatial spillovers across the different periods under analysis. Economic growth gains from transport infrastructure in the same region may come at the expense of other regions as there is clear evidence of negative spillovers from mobile production factors. In terms of policy implications, the following conclusions are possible. (1) Based on the empirical results in Sections 3.2 and 4.1, we suggest that the investment policy should give priority to the development of cross-regional transport networks instead of intra-regional construction. The existence of spatial externalities emerging from the contribution of transport infrastructure to regional growth implies that the decision for the provision of transport infrastructure should be made within a ‘‘supra-regional’’ perspective. Due to the network characteristics of transport facilities, the central government should pay special attention to the regional coordination of transport construction among lower administrative units, such as provinces, in order to avoid regionoriented investment modes. By altering investment patterns in transport infrastructure relative to those of the neighboring regions, each region has the ability to modify the size of its transport stock at the expense or to the benefit of its neighbor (Lall, 2007). Thus, the central government should give guidelines and constraints for decision-making by local government on their investment patterns. (2) According to the analysis in Sections 3.2, 4.2 and 4.3, we believe that at the regional level, relevant industrial policies for the lagging regions are urgently needed due to the existence of negative spillovers. The industrial agglomeration effects induced by transport development will lead to an increased transfer of industrial activity from western, northeastern, central China to eastern China, especially the technology-intensive industries. Since labor costs have not yet hindered the economic development in the eastern provinces, transport infrastructure development cannot generate industrial expansion there. Thus we can deduce that in the coming years, the technology-intensive industries in the other regions would still transfer to the eastern region due to agglomeration effects. Due to the possible presence of negative spillovers of transport infrastructure arising from these factor migrations, local governments in underdeveloped areas should alter their industrial policies to avoid redistribution of economic activities. Targeted region-specific industrial policies are needed, such as favorable tax policies and lower interest rates for loans for investments in local labor-intensive and technology-intensive industries (Liu et al., 2011).

Acknowledgements We thank the anonymous reviewer and the editor for their constructive comments and valuable suggestions. We express our

65

appreciation for the support of the project entitled ‘Public Policy Simulation Facing Complex Environment’ granted National Nature and Science Foundation at China (No. 71073037), and Short Term Visiting Study Funding of Harbin Institute of Technology, China. We thank Mr. Yang Yong for his contributions on the map making.

References Álvarez, A., Arias, C., Orea, L., 2006. Econometric testing of spatial productivity spillovers from public capital. Spanish Treasury 178, 9–21. Anselin, L., 2001. Spatial Econometrics: A Companion to Theoretical Econometrics. Basil Blackwell, Oxford. Aschauer, D.A., 1989. Is public expenditure productive? Journal of Monetary Economics 23 (2), 177–200. Banister, D., Berechman, J., 2001. Transport investment and the promotion of economic growth. Journal of Transport Geography 9 (3), 209–218. Berechman, J., Ozmen, D., Ozbay, K., 2006. Empirical analysis of transportation investment and economic development at state, county and municipality levels. Transportation 33 (6), 537–551. Boarnet, M.G., 1995. Transportation infrastructure, economic productivity, and geographic scale: aggregate growth versus spatial redistribution. Working Paper No. 255. University of California Transportation Center, Berkeley, California. Boarnet, M.G., 1996. The direct and indirect economic effects of transportation infrastructure. Working Paper No.340, The Transportation Center, University of California, Berkeley, CA. Boarnet, G.M., 1998. Spillovers and the locational effects of public infrastructure. Journal of Regional Science 38 (3), 381–400. Canning, D., Bennathan, E., 2007. The rate of return to transportation infrastructure. World Bank Working Paper No. 3293. Cantos, P., Gumbau-Albert, M., Maudos, J., 2005. Transport infrastructure, spillover effects and regional growth: evidence of the Spanish case. Transport Reviews 25 (1), 25–50. Cohen, J.P., Paul, C.J.M., 2004. Public infrastructure investment, interstate spatial spillovers, and manufacturing costs. The Review of Economics and Statistics 86 (2), 551–560. Del Bo, C., Florio, M., 2012. Infrastructure and growth in a spatial framework: evidence from the EU regions. European Planning Studies 20 (8), 1393–1414. Fujita, M., Krugman, P., Venables, A.J., 1999. The Spatial Economy, Cities, Regions and International Trade. The MIT Press, Cambridge. Goldsmith, S.B., 1951. A Perpetual Inventory of National Wealth, Studies in Income and Wealth, vol. 14. NBER, New York. Gomez-Antonio, M., Fingleton, B., 2009. Analysing the impact of public capital stock using the NEG wage equation: a panel data approach. Spatial Economics Research Centre Discussion Paper 24. Holtz-Eakin, D., 1994. Public-sector capital and the productivity puzzle. Review of Economics and Statistics 76, 12–21. Holtz-Eakin, D., Schwartz, E.A., 1995. Spatial productivity spillovers from public infrastructure: evidence from state highways. National Bureau of Economic Research Working Paper No. 5004. Hu, A., Liu, S., 2009. Transportation, economic growth and spillovers: conclusion based on spatial econometrics. China Industrial Economics 5, 5–14. Huang, J., Li, K., 2006. Export openness, regional market size and economic growth. Economic Research 6, 27–38. Krugman, P., 1991. Increasing returns and economic geography. Journal of Political Economy 99, 483–499. Lall, V.S., 2007. Infrastructure and regional growth, growth dynamics and policy relevance for India. The Annals of Regional Science 41 (3), 581–599. LeSage, J., Pace, R.K., 2009. Introduction to Spatial Econometrics. CRC Press, Boca Raton, FL. Liu, Y., 2010. Transport infrastructure investment, regional economic growth and the spatial spillover effects-based on highway and marine’s panel data analysis. China Industrial Economics 12, 37–46. Liu, N., Chen, Y., Zhou, Q., 2007. Spatial spillover effects of transport infrastructure on regional economic growth. Journal of Southeast University (English Edition) 23, 33–39. Liu, H., Liu, W., Liu, Z., 2011. The quantitative study on inter-regional industry transfer. China Industrial Economics 9, 79–87. Mao, Q., Sheng, B., 2011. Economic opening, regional market integration and total factor productivity. China Economic Quarterly 11 (1), 181–210. Ministry of Public Security, 2010. National Statistical Compilation of Transient population. China Statistical Publishing House, Beijing. Ministry of Transportation, 1984–2010. China Transportation Yearbook. China Statistical Publishing House, Beijing. Moreno, R., Lopez-Bazo, E., 2007. Returns to local and transport infrastructure under regional spillovers. International Regional Science Review 30 (1), 47–71. Munnell, A.H., 1990. How does public infrastructure affect regional economic performance. New England Economic Review (September–October), 11–32. Munnell, A.H., 1992. Policy watch: infrastructure investment and economic growth. Journal of Economic Perspectives 6 (4), 189–198. Ozbay, Ozmen-Ertekin, D., Berechman, J., 2003. Empirical analysis of the relationship between accessibility and economic development. Journal of Urban Planning and Development 129 (2), 97–119.

66

N. Yu et al. / Journal of Transport Geography 28 (2013) 56–66

Ozbay, K., Ozmen, D., Berechman, J., 2007. Contribution of transportation investments to county output. Transport Policy 14 (4), 317–329. Peng, W., 2008. The influence of capital flow on Chinese three regional economic gap. PhD. Thesis. Fudan University. Sahoo, P., Dash, K.R., Nataraj, G., 2010. Infrastructure development and economic growth in China. Institute of Developing Economies Discussion Paper No. 261. State Statistical Bureau (SSB), 1982–2010. China Statistical Yearbook. China Statistical Publishing House, Beijing. State Statistical Bureau, 2010. People’s Republic of China’s Statistical Series of 60 Years. China Statistical Publishing House, Beijing. Wang, F., Guo, S., Jiang, Q., 2010. Labor migration and regional development in China: a regional CGE analysis. .

Yi, Z., Ying, Q., 2011. The regional labor shortage in the background of industries transfer in China. Financial Industry 4, 21–53. Yu, N., de Jong, M., Storm, S., Mi, J., 2012. Transport infrastructure, spatial clusters and regional economic growth in China. Transport Reviews 32 (1), 3–28. Zhang, X., 2009. Transport Infrastructure, Spillover, and Regional Economic Growth. Nanjing University Press, Nanjing. Zhang, H., Yi, B., 2012. Infrastructure, spatial spillovers and regional total factor productivity. Economics 2, 61–67. Zhang, J., Wu, G., Zhang, J., 2004. The estimation of China’s provincial capital stock: 1952-2000. Economic Research 10, 35–44. Zhang, J., Gao, Y., Fu, Y., Zhang, H., 2007. Why does China enjoy so much better physical infrastructure? Economic Research 4 (3), 4–19.