The Impacts of High-Speed Railway on Urban GDP and Its Agglomeration: Evidence from China

,


Introduction
High-speed railways (HSR) have earned widespread recognition as a fast, safe, and comfortable intercity transportation mode [1].Tis transit form signifcantly improved the overall travel satisfaction for passengers [2,3].Over the past decade, the mainland of China has witnessed a notable surge in the development of HSR.As of the end of 2019, China possesses the world's most extensive HSR network, encompassing an operational distance exceeding 35,000 billion kilometers [4].
Concurrently, the rapid expansion of HSR has ignited a surge of scholarly investigations worldwide into the spillover efects generated by this transportation mode [5][6][7].A profound understanding of the economic externalities associated with HSR is paramount, as it empowers us to conduct a comprehensive assessment of its spillover efects on the economy.Moreover, these economic externalities provide invaluable insights for the formulation of policies and regulations concerning HSR.For instance, understanding how HSR infuences land use and property values can guide decisions related to zoning and urban development policies.Terefore, grasping the exogenous factors that infuence HSR is of paramount importance for policymakers endeavoring to devise a more pragmatic strategy for HSR development [8][9][10].
Investigating the spillover efects of various modes of transportation has been widely studied by the scholars, and an excellent study system has been established.Econometric model plays an irreplaceable role in such studies [11][12][13][14].Currently, the primary models used for analyzing the spillover efects of rail transit are the panel data model [2,15], spatial econometric model [6,[16][17][18][19][20], and DID model [21][22][23][24].Table 1 lists some of the research on economic externalities associated with HSR and urban rail conducted using these three models.Panel data include  [14] Rail transit is able to enhance the housing prices of the surrounding houses Combining hedonic model and panel data model Tian et al. [2] Te improvement in network position of HSR inhibits service industry agglomeration in peripheral regions Constructing framework of complex network analysis and panel regression methods

DID model
Aslund et al. [1] Commuter train access has little impacts on the employment development Using the introduction of a local commuter train in Sweden Tian et al. [25] HSR and service-sector agglomeration have positive correlation Applying hypothetical counterfactuals to eliminate the exogeneity problem of dependent variables Li et al. [26] Te opening of HSR has a signifcant threshold efect on improving the efciency of the service industry Heterogeneity analysis of the impact of HSR on the service industry Zhu et al. [27] Te opening of HSR has a positive impact on urban land expansion Analyzing the time lag impact of HSR Tang et al. [28] HSR signifcantly promotes regional innovation Investigating the impacts of urban form on the correlation between HSR and regional innovation

Spatial econometric model
Zheng et al. [6] Urban rail transit can signifcantly improve urban air quality Using several methods to test the robustness of the regression results

Huang and Xu
[29] Te construction of HSR narrows the regional diferences in daily accessibility, but it will expand the diferences in potential accessibility and location accessibility Divide accessibility into location accessibility, potential accessibility, and daily accessibility observational data for the same group of subjects at diferent time points and/or diferent locations, encompassing both spatial and temporal dimensions.Tus, the panel data model aids in identifying the efects of time trends, observed individual diferences, and endogeneity, thereby providing a more comprehensive understanding and quantifcation of the impact of projects such as the economic externalities of HSR [30,31].Failure to capture the impact of policies at a certain time node is a drawback of the panel data model [15,32].Spatial econometric models focus on the interrelationships of infuencing factors within geographic space [6,35].While the DID model allows for causal inference, i.e., it enables the determination of the impact of the opening and operation of HSR on urban economies by examining the changes in various economic indicators in cities before and after the HSR is introduced.Te DID model is employed in this paper.Our research introduces two noteworthy modeling innovations.First, we utilize multiple robustness testing methods to enhance the credibility of the causal relationships derived from the DID model.Tese methods strengthen the basis for drawing meaningful conclusions.Second, within the factors afecting the DID model, we introduce a virtual HSR operation time to control for unobservable factors that might otherwise afect the urban economy.Tese refnements contribute to a more robust analysis of the spillover efects of HSR on urban economies.However, due to the diferent stages of HSR development and variations in urban economic foundations, the impact of HSR on economy varies across cities [34].Many studies indicate that the economic efects of transportation infrastructure are positive for highly developed cities, as the accessibility and connectivity of HSR can promote economic growth in these cities.However, for some smaller cities, the agglomeration and suction efects of HSR may lead to talent outfow [15].Previous studies on heterogeneity have primarily focused on a geographical perspective [27,35], while our paper will approach the economic development heterogeneity brought about by HSR from the standpoint of a city's development level, providing insights for maximizing the positive externalities of HSR.Specifcally, we will explore the regional heterogeneity of HSR from two aspects: the impact of HSR operation on cities at diferent economic levels and which economic level of cities is more sensitive to HSR operation.
Most studies investigating the spillover efects of HSR or other modes of transportation on urban economy are usually empirical, meaning that although the positive externalities of HSR have been identifed, the underlying mechanism of these externalities has not been fully understood yet [36,37].In addressing this issue, we have incorporated the fow of population, urban industrial structure, and scientifc research level as intermediate variables in the models.Tese intermediate variables are introduced to shed light on the underlying mechanisms of the spillover efects of HSR on urban economies.Urban population serves as a critical indicator of the scale and dynamism of a city, which could infuence how HSR impacts labor markets, demand for goods and services, and overall economic growth.City industrial structure refects the composition of economic activities within a city, and variations in this structure may mediate the efects of HSR by determining which industries beneft the most from improved connectivity.Scientifc research level represents the knowledge and innovation capacities of a city, which may be afected by enhanced accessibility through HSR and, in turn, infuence the city's economic development.By introducing these intermediate variables, this study aims to delve deeper into the intricate mechanisms that drive the spillover efects of HSR on urban economies.It is anticipated that the analysis of these mediating variables will contribute to a more comprehensive understanding of the multifaceted dynamics in the relationship between HSR and urban economic development.
In summary, this work has three main innovations.First, we develop three robustness methods to verify the causal relationship derived from DID model.Second, the heterogeneity analysis is conducted to explore the impact of HSR operation on cities at diferent economic levels and which economic level of cities is more sensitive to HSR operation.Finally, unlike other empirical studies, intermediate variables are introduced to shed light on the underlying mechanisms of the spillover efects of high-speed rail (HSR) on urban economies.
To conduct a comprehensive analysis mentioned previously, this study puts forward three hypotheses of the relationship between HSR and urban gross domestic product (GDP), as shown in Figure 1, and verifes these hypotheses by statistical data analysis and modeling.Tese hypotheses, elucidating the impact of HSR on urban GDP and its underlying mechanisms, are progressively layered.
Hypothesis 1. Te operation of HSR can directly contribute to the growth of urban GDP.
On the one hand, the introduction of HSR in a city facilitates more convenient intercity travel for residents, signifcantly amplifying the quantity and scale of economic activities.Tus, the overall economic situation will be improved.However, due to the siphoning efect of HSR, talent and markets may relocate to other areas with more developed economies.Similar to the exogeneities of HSR on urban air quality [38,39], the spillover efects of HSR on GDP may exist heterogeneity for cities with diferent development levels.Terefore, it is necessary to investigate whether the efects of HSR on cities at diferent developmental stages difer.Hypothesis 2. Te urban GDP promoted by HSR exists heterogeneities among diferent cities.
On the other hand, cities may exhibit varied responses to the construction and expansion of HSR.For cities with more developed economic level, HSR may have a greater efect.

Hypothesis 3. Te operation of HSR can promote the development of urban GDP by afecting several intermediate variables.
Furthermore, the operation of HSR may have indirect efects on urban GDP.As shown in Figure 1, HSR can accelerate the fow of population to the city where the Journal of Advanced Transportation high-speed rail operates, promote the upgrading of urban industrial structure, improve the level of urban scientifc research, and thus promote the development of urban GDP.Te indirect efects are often ignored in several relevant studies; hence, the positive efects of HSR have been underestimated.
Te remainder of this paper is organized as follows.Te statistical data are collected and preliminary analyzed in Section 2. Section 3 introduces common DID model and establishes two-stage DID model to study the direct and indirect efects of HSR on urban GDP.Section 4 presents the regression results of the models, and the robustness of the results is tested.Te heterogeneities of HSR on cities with diferent development level are further analyzed.Te conclusions are presented, and relevant policy suggestions are put forward in Section 5.

Data Description
To mitigate the impact of regional heterogeneity on the estimation results, thirty cities on the mainland of China which operated HSR by the end of 2019 are selected as study sites.Figure 2 depicts the spatial distributions of these cities.Te selection of studied cities is based on two criteria: First, their distribution aligns with that of the HSR network, with a concentration of cities in the eastern region.Second, we prioritize provincial capitals and municipalities in the selection process.However, certain provincial capitals like Hohhot were excluded due to insufcient data.Given the rapid development of HSR in mainland China over the past decades, our research spans from 2010 to 2019 to capture this transformative period.
Te economic variables refecting the development level of the city are the explained variables of interests in this study.In order to measure the economic level of the city, we select real gross domestic product (GDP) instead of nominal GDP.Compared with nominal GDP, real GDP can more accurately refect the quantity of goods that citizens can buy with their income, making it a more accurate indicator of the urban GDP's level.Real GDP agglomerates spatially, which could refect the extent of economic agglomeration in a city, as illustrated by equation ( 1) and Table 1.
where A it denotes the area of city i in year t.HSR variables are the main independent variable in this study.According to Chang et al. [4], we use 0-1 dummy variables (i.e., HSR it ) to describe whether city i introduces HSR in year t.Furthermore, a number of cities connected by HSR for city i in year t (i.e., HSR line it ) are introduced to measure the service level of HSR in the city more accurately, as shown in Table 2.
In order to examine the indirect efects of HSR on urban GDP, three mediating variables are taken into account in this study.We select resident population instead of registered population to measure the fow of population.Compared with resident population, registered population is more sensitive to policy, resulting in estimation bias.When a city is sufciently attractive, its resident population will immediately increase signifcantly.Nevertheless, the change of registered residence population has a certain lag.Consequently, studying the efect of HSR on the urban GDP via resident population is a more accurate method.We use Ind calculated through equation ( 2) to measure the industrial structure of city i in year t.
where RS denotes the proportion of added value of secondary industry in GDP.RT represents the proportion of added value of tertiary industry in GDP.Te greater the value of Ind, the more industrial structure is transferred to the tertiary industry.STE is introduced to refect urban scientifc research level.Generally speaking, the more the government invests in the science and technology industry, the more developed the science and technology industry.Moreover, several control variables are chosen to avoid estimate bias, as indicated in Table 1.Due to the lack of data, some data are flled through expectation maximization (EM) algorithm.To mitigate the potential impact of multicollinearity on the estimation results, variables that exhibit strong correlations are excluded from the analysis.Figure 3 presents the Pearson coefcients for all variables evaluated on the basis of equation (3).Te darker color block indicates  Journal of Advanced Transportation a higher linear correlation.Te elements on the diagonal are all black, because the collinearity of the variable and itself is one.Figure 3 displays a strong relationship among HSR, GDP, and the mediating variables.However, the causal relationships need to be further studied by establishing DID models after removing interference caused by other factors.
where X and Y denote two variables that are selected to calculate their correlation.

Modeling Study
Taking Zhang [24] as a reference, we initially establish a multiphase DID model to explore the economic impact of HSR, as follows: where GDP it denotes the economic development level of city i in year t.X it is a vector composed of control variables illustrated in Section 2. According to Tao et al. [20], when determining the spillover efects of transportation modes in city level in China, μ i ought to be regarded as fxed efect instead of random efect.Terefore, μ i represents individual fxed efect that difers across cities, and λ t denotes time fxed efect which measures the change of years in this research.ε it presents random interference term.α 1 measures the causal elastic relationship between HSR it and GDP it .However, due to the endogenous problems of independent variables and confounding dynamics, the robustness of estimate results regressed by equation (4) needs to be tested.Tis study employs three methods to verify the robustness of the model.If strong spatial correlations exist among the studied cities, spatial econometric models are better appropriate for the data than DID models.Consequently, the frst method is to calculate the Moran's I index and Geary's C index of each variable, as given in equations ( 5) and ( 6), in order to measure the spatial autocorrelation degree of studied variables.Te value of Moran's I is between −1.000 and 1.000.When Moran's I > 0.000, a positive spatial association exists between the studied variables.Te larger the absolute value of Moran's I, the more obvious the spatial correlation of the variables.Geary's C typically ranges between 0.000 and 2.000, with Geary's C < 1 indicating a spatial positive correlation for the variable.Te second method is changing the explained variable to GDP it .If the relationship and the elastic relationship are both positive and signifcant, the causal relationship is verifed.Te third method is setting virtual HSR open years.Te regression results of equation ( 1) cannot be accepted even though they are signifcant if GDP it is improved by other invisible factors afecting GDP it at the same time of HSR.Te main idea of the third method is to set virtual HSR open years to avoid such a situation, as shown in equation (7).If α 2 is insignifcant, the causal relationship between HSR and urban GDP is verifed, indicating that HSR, instead of other invisible factors, afects urban GDP.While if α 2 remains positive and signifcant, it is other infuencing factors instead of HSR that promote the development of urban GDP.
Equations ( 8) and ( 9) show the two-stage DID model that can be regressed using 2SLS method.M it is the mediating variables introduced in Section 2. Equation ( 8) represents the frst-stage regression that measures the efects of HSR on mediating variables.Equation ( 9) is the secondstage regression investigating the impacts of mediating variables to urban GDP.In case α 3 and α 4 are all positive and signifcant, HSR afects mediating variables and then improves urban GDP.Te indirect efects of HSR can be found and proved in this way.

Regression Results
In this section, the DID models are frst applied to investigate the spillover efects of HSR on urban GDP and its agglomeration.Tereafter, the three methods mentioned above are employed to test the robustness of the regression results.Ten, the heterogeneities of HSR on cities with diferent development levels are further analyzed.Finally, the indirect impacts illustrated in Figure 1 and the mechanism of HSR promoting urban economic development are identifed.

General Results
. Tables 3 and 4 present the estimate results of equation (6).Table 3 shows the efects of HSR on urban GDP.Te values in brackets are the p value of each coefcient.Te values in brackets in the subsequent tables have the same meaning.Te main explanatory variable is HSR in columns 1, 3, and 5, and it changes to HSR line in columns 2, 4, and 6.Column s1 and 2 do not introduce control variables.Columns 3 and 4 add control variables into the model, and columns 5 and 6 add individual fxed efects and time fxed efects.Te coefcients of HSR and HSR line are all positive and signifcant, indicating that the operation and expansion of HSR can efectively promote the development of urban GDP.After adding the fxed efects and the control variables into the model, the R square signifcantly improves, indicating that it is essential to take them into consideration.Te values of HSR and HSR line decrease after adding control variables, implying that if the efects of other factors and the infuences of individual and time are ignored, the efects of HSR on urban GDP will be overestimated.In Table 3, all the fxed efects and the control variables are put into the regression progress.In columns 5 and 6, the coefcients of HSR and HSR line are all positive and signifcant at 0.01 level, which demonstrates that the construction of HSR is able to promote urban GDP remarkably.HSR can increase the GDP of a city by 14.200%, and for each additional HSR line, the urban GDP will increase by 7.200%.Te opening and operation of HSR makes the intercity travel more convenient for citizens, hence increasing economic activities between cities and promoting economic growth.Te exogeneity of HSR on urban economic agglomeration is presented in Table 4.Each column in Table 4 has the same meaning as shown in Table 3, except for the explained variable ln GDPag.Columns 5 and 6 clarify that the operation of HSR can signifcantly promote the economic agglomeration of a city.Te opening of HSR can increase the economic agglomeration level of the city by 14.200%.Te economic agglomeration level of the city will increase 7.200% on average when a new HSR line is constructed in the city.In the past decade, the majority of Chinese cities have developed at a high speed.
Te land area of the city has expanded rapidly.Despite this, HSR still contributes positively to the development of GDP, even when urbanization is considered.HSR can signifcantly and comprehensively promote the development of urban GDP.
From Tables 3 and 4, it can be concluded that HSR may efciently and considerably enhance both the city's economic development and its agglomeration.Section 1's Hypothesis 1 is proven.However, as demonstrated in Section 3, the results cannot be accepted if they are not robust, even if the coefcients are signifcant and positive.Terefore, it is necessary to assess the robustness of the estimate fndings regressed by DID models to confrm their availability.

Robustness Tests.
Tree methods are applied to verify the robustness of estimate results obtained in Section 4.1.Table 5 presents the results of spatial correlation analysis of each variable of interest.All variables' Moran's I index is close to 0.000 and Geary's C index is close to 1.000.Furthermore, the coefcient of each index is not statistically signifcant at 0.01 level, which indicates that the spatial correlations of the studied variables are insignifcant.It is appropriate to employ DID model for regression.
Te regression results of the second and third methods are presented in Table 6.Columns 1 and 2 replace the explained variable with GDP, while columns 3 and 4 replace the explained variable with GDPag.Te coefcients of HSR and HSR line remain positive and signifcant at 0.01 level, confrming the robustness of the results regressed in Section 4.1.Column 5 in Table 5 has the same meaning as column 5 Journal of Advanced Transportation in Table 2. Column 6 changes HSR to HSR virtual .From the comparison of column 5 and column it can be found that the coefcient of HSR turns insignifcant when replacing HSR with HSR virtual , indicating that HSR, not other infuencing factors, promotes the growth of the urban GDP.
In this section, three methods are used to confrm the robustness of the regression results, and it can be concluded that HSR has injected vitality into the urban GDP and agglomeration.However, diferent cities may beneft differently from the opening of HSR [40].Terefore, the heterogeneities of HSR on cities with diferent development levels ought to be investigated.

Heterogeneity Analysis.
City level is a measurement standard indicating the complete development level of a city on the Chinese mainland.Hence, this study selects city level to divide the studied cities.Table 7 presents the heterogeneity analysis of HSR.Columns 1 and 2 show the economic efects of HSR on frst-tier cities, columns 3 and 4 denote the economic efects of HSR on second-tier cities, and the economic efects of HSR on third-tier cities are shown in columns 5 and 6.
Diferent from Hypothesis 2 presented in Section 1, although heterogeneities exist in the ability of HSR promoting urban GDP among diferent cities, these abilities are largest in the third-tier cities, followed by the frst-tier cities and the second-tier cities. Tis phenomenon is strongly related to the indirect efects of HSR on urban GDP.Compared with the frst-tier cities and the second-tier cities, the populations of the third-tier cities are relatively fewer, and the urban infrastructure is better developed and suitable for every citizen.In addition, the industrial structure of the  third-tier cities has more room for development.As a consequence, when HSR opens in the third-tier city, it improve the economic development of city in a more efcient way.However, this does not mean that priority should be given to the development of HSR systems in the third-tier cities, as the construction of HSR network should not be divorced from the basic principle of passenger demand.
From heterogeneity analysis, we derived several interest conclusions and had some new problems.To explore the full potential of HSR, it is vital to investigate the fundamental mechanisms of HSR's contribution to the urban GDP.4.4.Indirect Impacts.Equations ( 7) and ( 8) provide an effcient method for determining how HSR infuences the urban GDP.Te indirect efects of high-speed rail on metropolitan economies are shown in Table 8.In columns 1, 2, and 3, urban population serves as the mediating variable.In columns 4, 5, and 6, the industrial structure is introduced as the mediating variable.In columns 7, 8, and 9, expenditure on science and technology is used to examine indirect efects.
Te regression results presented in Table 8 demonstrate that HSR has the potential to facilitate population migration, expedite the transformation of urban industrial structure, and enhance investments in science and technology within cities. First, urbanization is closely associated with increased economic activity, and as economic activity expands, it inevitably contributes to the growth of the urban GDP.Second, the transformation of the industrial structure leads to a higher proportion of the tertiary industry.Te expansion of the tertiary industry, particularly the service sector, not only generates signifcant economic activity but also stimulates growth in other industries.Tird, innovation serves as a driving force for urban economic development.Te city's investment in science and technology plays a crucial role in enhancing its economic vitality.HSR indirectly supports the growth of the urban GDP through these three factors.Terefore, Hypothesis 3 has been confrmed.
In order to maximize the positive impact of high-speed rail on urban economic development, cities must make adequate preparations for the following aspects before opening or expanding high-speed rail: (1) ensure that urban infrastructure is sufcient to accommodate expected population growth, (2) prepare for industrial restructuring, and (3) the government should provide sufcient funds for technological innovation.Tese measures will enable highspeed rail to play a more comprehensive role in the economic development of cities. Te introduction of HSR systems may have negative consequences on cities with inadequate infrastructure.Tis is due to the siphon efect, Journal of Advanced Transportation   where more people migrate from these cities to other locations, resulting in a redistribution of production resources.Tis serves as a warning to policymakers.Terefore, when expanding a country's HSR network, it is crucial to consider the development level of each city rather than blindly connecting every city on the map.A wellthought-out HSR development plan should aim to maximize the positive externalities of HSR while minimizing its negative efects to the greatest extent possible.

Conclusions
Tis study develops DID models to assess the efects of HSR operation on the urban GDP.Te panel data consist of thirty cities operating HSR by the end of 2019, covering the period from 2010 to 2019.Panel data model is applied to investigate the causal elastic relationship between HSR and urban GDP and agglomeration, respectively.Te regression results are evaluated using three methods.In addition, the ability of HSR promoting urban GDP exists heterogeneities among cities with diferent development levels.Also, mediating variables play an important part in the efects of HSR on urban GDP.Te innovation of this study can be summarized into the following three aspects.
(1) HSR Economic Externality Clarifcation.We emphasize the signifcance of understanding the economic externality of HSR in assessing its impact on the economy and its role in informing policy decisions (2) Comprehensive Econometric Modeling.Our study employs robust econometric models, including panel data and spatial autoregressive models, aiming to mitigate bias and enhance the accuracy of estimations (3) Mechanistic Insights into HSR Efects.We propose and test three progressive hypotheses regarding the relationship between HSR and urban GDP, contributing to a deeper understanding of the mechanisms through which HSR infuences urban economies Indeed, HSR can have signifcant impacts on urban GDP and agglomeration.First, HSR can contribute to the growth of urban GDP.Te strongest impact is observed in third-tier cities, followed by frst-tier and second-tier cities. Tis pattern suggests that HSR can bring more opportunities for economic activities, such as increased business interactions, trade, tourism, and investment, to these cities.As a result, the urban economies in these areas can experience considerable growth.Second, HSR facilitates the migration of people to urban areas.Improved connectivity and reduced travel time provided by HSR make urban areas more attractive for both individuals and businesses.Tis infux of people can stimulate urban development and modernization.Additionally, HSR can encourage the upgrading of urban industrial structures by attracting new industries, promoting innovation, and fostering economic diversifcation.Finally, the presence of HSR contributes to enhancing the quality of urban scientifc research.By connecting cities and improving accessibility, HSR enables researchers and academics to collaborate more easily across diferent urban areas.Tis collaboration facilitates the exchange of knowledge, expertise, and resources, thereby improving scientifc research and innovation in urban centers.
To optimize the function of HSR and maximize its societal benefts, policymakers should consider the following measures.Investing in urban science and technology is crucial to harness the potential of HSR.Tis includes supporting research institutions, promoting innovation hubs, and fostering a culture of scientifc inquiry.By focusing on science and technology, cities can attract talent, foster entrepreneurship, and drive economic growth.In addition, policymakers should encourage the upgrading of urban industrial structures.Tis involves attracting industries with high value addition, promoting sustainable practices, and supporting the development of knowledgeintensive sectors.By diversifying and modernizing their industrial base, cities can improve productivity, create employment opportunities, and enhance their competitiveness.Finally, given the population upsurge resulting from HSR, investments in urban infrastructure are imperative to meet the escalating needs of residents and businesses.Tis includes expanding transportation networks, developing residential areas, improving public services, and ensuring access to amenities.Adequate infrastructure development ensures that urban areas can accommodate the increased demand and provide a high quality of life for their residents.In summary, HSR can indeed promote urban GDP and agglomeration.By focusing on urban science and technology development, upgrading industrial structures, and constructing necessary infrastructure, policymakers can optimize the benefts of HSR and support sustainable urban development.However, it is important to note that, due to the lack of data, the direct efects of HSR are not investigated in this research, which needs to be studied in the future work.

Figure 1 :Figure 2 :
Figure 1: Te mechanism of the efects of HSR on urban GDP.

Table 1 :
Summary of relevant researches on the exogeneities of transportation.

Table 3 :
Te causal elastic relationship between HSR and urban GDP.

Table 4 :
Te causal elastic relationship between HSR and urban economic agglomeration.

Table 5 :
Te results of the frst method.

Table 6 :
Te regression results of the second and third methods.

Table 7 :
Heterogeneity analysis between HSR and urban GDP.