The Transportation Infrastructure and Regional Economic Growth—Evidence from Dongguan Humen Bridge

Transportation infrastructure is an important factor driving regional economic growth, and scientifically evaluating the economic growth effects of key transportation infrastructure is an important theoretical issue of development economics. This paper takes the opening of Dongguan Humen Bridge as a quasi-natural experiment, and quantitatively identifies the regional growth effect it brings. This paper uses panel data from 21 prefecture-level cities in Guangdong Province from 1991 to 2007, and based on the synthetic control method, assigns different weights to the cities in the control group to construct the “counterfactual” state of Dongguan where the Humen Bridge is not opened, so as to compare the difference in economic development between “real Dongguan City” and “Synthetic Dongguan City”. The results show that the opening of the Humen Bridge in 1997 made Dongguan’s GDP and GDP per capita exceed the theoretical level that could be achieved when the Humen Bridge was not opened. The opening of the Humen Bridge significantly changed the economic growth trend of Dongguan. With the passage of time, the opening of the Humen Bridge has continuously strengthened the promotion of Dongguan’s economic growth, indicating that the transportation infrastructure has a long-term effect on the promotion of regional economic growth. This work advances the research on the relationship between transportation infrastructure and regional economic growth.


Introduction
The role of transportation infrastructure in economic growth and regional coor-How to cite this paper: Lai, Q. M. (2020). The Transportation Infrastructure and Regional Economic Growth-Evidence from Gate Bridge, the Panama Canal and so on. The opening of the transportation infrastructure in these important geographical locations not only directly improves the efficiency of the exchange of people and the flow of goods, but also creates huge revenue for the local area every year. The economic benefits are considerable.
Therefore, on the basis of learning from previous literature studies, this paper will be based on a brand-new perspective, with transportation hubs in important locations as the entry point, using the Synthetic Control Method (SCM) in policy evaluation to study key transportation foundations The impact of the opening of facilities on the development of the regional economy in turn gives a relatively stable result. This article believes that the Dongguan Humen Bridge is undoubtedly a key transportation infrastructure in China's regional economy, and it provides us with an excellent quasi-natural experimental scenario for analyzing the impact of the opening of transportation infrastructure on regional economic growth. This paper uses panel data from 21 prefecture-level cities in Guangdong Province from 1991 to 2007. The rest of the paper is organized as follows: the second part is literature review and analysis; the third part is estimation methods and data description; the fourth part is empirical results and analysis; the fifth part is the robustness test; the sixth part is the conclusion and policy recommendations.

Literature review and analysis
The upsurge of empirical research on transportation infrastructure by foreign economists began in the late 1980s. The pioneering research came from Aschauer (1989), who included transportation infrastructure into public infrastructure and considered the economic growth benefits of the overall infrastructure. He proved that the "core" infrastructure promotes economic growth, with significant output elasticity estimates of 0.24. Barro (1990) based on modern endogenous growth theory, pointed out that public infrastructure can overrate the long-term economic growth rate. However, some scholars pointed out that the estimation results using the time series method may overestimate the output elasticity of infrastructure (Tatom, 1991). Krugman (1991) used the research methods of regional economics and geographic economics and found that the emergence of the "core-periphery pattern" depends on transportation costs, and the reduction of transportation costs will make the industry appear at the same time. When the balance between the centripetal force and the centrifugal force is broken, the industry will have a "core-periphery" structural distribution. Venables (1996) studied the position changes of vertically connected industries in the face of changes in transportation costs. He found that when transportation costs are reduced from high to low, the first thing that appears is the concentration of economic structure and income, and the gap in regional economic structure and per capita income has widened. But further reductions may disrupt agglomeration and bring convergence. Fedderke and Bogetic (2009)  impact of infrastructure on South Africa's manufacturing industry and found that infrastructure has a direct impact on labor productivity and infrastructure has an indirect impact on total factor productivity. Roberts et al. (2012) adopted a counterfactual method based on the hybrid estimation correction of the "New Structural Economic Geography" model to evaluate China's expressway network expansion plan. They believed that the expressway network seems to reinforce the existing pattern of spatial inequality. But over time, these spatial development inequalities will decrease due to increased migration. In addition, convenient transportation infrastructure is also one of the important basic conditions for promoting regional trade development (Bougheas et al., 1999;Limao & Venables, 2001;Francois & Manchin, 2013). Furthermore, some scholars have studied the connection between the individual transportation infrastructure and international trade, and concluded that the development of transportation infrastructure and the improvement of transportation efficiency have a positive and positive effect on trade (Nordas & Piermartini, 2004;Clark et al., 2004;Fujimura & Edmonds, 2006).
But not all studies tend to hold that there is a direct or indirect link between transportation infrastructure and economic growth. For example, Holtz-Eakin and Schwartz (1995) believe that the relationship between transportation infrastructure and economic growth is not obvious. Later, after exploring the impact of infrastructure on economic growth, some scholars came to conclusions similar to those of Holtz-Eakin and Schwartz (Evans & Karras, 1994a, 1994bChandra & Thompson, 2000).
In China's research, some documents focus on the impact of China's transportation infrastructure on economic growth. Huang and Li (2006) put forward the viewpoint of expanding the market scale by further improving the transportation infrastructure of provinces and cities in the Mainland, so as to give play to the role of market scale in determining economic growth performance. Wang and Wang (2007) concluded that infrastructure plays a dominant role in economic growth. Zhang (2007Zhang ( , 2012 separated transportation infrastructure as a separate form from capital. His research found that China's transportation infrastructure investment and economic growth showed strong spatial agglomeration characteristics, and the spatial spillover effect on inter-regional economic growth is very significant. Liu and Hu's (2010) study found that transportation infrastructure has a significant positive role in promoting China's economic growth. Liu (2010) studied the spatial spillover effect of the stock of fixed capital of road and water transportation on China's economic growth and found that the stock of fixed capital of road and water transportation on the whole has a positive effect on regional economic growth. Liu and Hu (2011) pointed out that the improvement of transportation infrastructure has had a significant positive impact on China's regional trade, and the improvement of transportation infrastructure has promoted regional economic integration. Zhou and Zheng (2012) used the DID method to investigate the impact of railway acceleration on eco- Furthermore, some scholars had begun to study the relationship between transportation infrastructure and total factor productivity (TFP). They found that transportation infrastructure has a positive impact on total factor productivity Zhang & Yi, 2012). In addition, some scholars believe that investment in transportation infrastructure construction also plays an important role in promoting regional industrial clusters (Ren & Zhang, 2010). Sheng et al. (2011) found that infrastructures have a positive effect on the export of Chinese enterprises. Ruan (2017) studied the investment and economic effects of transportation infrastructure planning in the Silk Road Economic Belt and found that the impact of railways and highways on import and export trade was significantly positive.
There are also researchers who explore the impact of transportation infrastructure on the micro-level of corporate behavior or residents' behavior from a micro perspective. Li and Li (2009) believed that the construction of high-grade roads significantly reduced the inventory capital occupation of Chinese manufacturing enterprises, while the investment of ordinary roads and railways had no significant impact on the inventory level of enterprises. Shi et al. (2018) found that the railway acceleration had played a positive role in technological progress and efficiency improvement of enterprises and promoted the growth of total factor productivity. From the perspective of corporate inventory, Zhang et al. (2018) explored the mechanism behind transportation infrastructure to promote economic growth under a unified framework and verified that transportation infrastructure plays an important role in expanding enterprise scale, improving enterprise operating efficiency and enhancing market vitality. Guo et al. (2019) found that road infrastructure not only has a "multiplier effect" that can directly promote economic development, but also can indirectly promote economic growth through the residents' "consumption effect".
A review of the previous research literature shows that the research on the relationship between transportation infrastructure and economic development has made great progress. Domestic and foreign literature is mostly empirical research, and most of them focus on the aspects of transportation infrastructure promoting changes in the production field. It is believed that the investment and construction of transportation infrastructure mainly affects economic development by affecting the flow of factors in the production field.
In research methods, the cost function method and production function method are mainly used to measure the impact of transportation infrastructure on economic development. In terms of the data, time series data is mainly used for analysis in the early stage, and the output elasticity obtained based on this method is relatively high; In recent years, mainly based on panel data analysis, the In terms of research content, the research literature has gradually expanded from exploring the relationship between transportation infrastructure and economic growth to specific areas, such as the impact of transportation infrastructure on regional industrial structure, import and export trade, location selection of industry enterprises, and individual behavior decisions. The selection of research indicators is still relatively single. Early research mainly focused on the stock of transportation infrastructure, and later gradually evolved to extract specific levels of transportation infrastructure from the overall for individual research; with the research boom of new economic geography With the rise, spatial and geographic factors are gradually being considered, and then the spillover effect of transportation infrastructure is discussed. However, it is rare to study the impact of a specific infrastructure on regional economic growth, which also limits our understanding of regional economic development by building transportation infrastructure in special locations between regions.
In fact, the economic development of many regions and the changes in the geo-spatial pattern of regional and even national economic development are largely dependent on the impact of the completion and opening of key infrastructure. Based on this important background with both theoretical value and practical significance, this paper selects the opening of the Dongguan Humen Bridge as a quasi-natural experiment, and uses the cutting-edge analysis methods of policy evaluation to explore its impact on economic growth, which undoubtedly forms an important supplement to the existing literature and existing research boundaries.

Background
The transportation infrastructure of Dongguan is the hardware foundation that supports and leads the development of Dongguan and surrounding areas. In order to strengthen exchanges and ties with the west bank of the Pearl River, Humen Automobile Ferry Port was completed and put into operation in 1991, but shortly there was a problem of insufficient capacity. As a result, Guangdong Provincial Government began to promote the Humen Bridge construction project. On June 9, 1997, the Humen Bridge was officially opened for operation.
In this context, for this article we regards the opening of the Humen Bridge as a policy implemented by the Guangdong Provincial Government and uses the method of policy evaluation to evaluate the impact of the opening of the Humen Bridge on the regional economic development. The methods for evaluating policy items mainly include difference-in-differences (DID), propensity score matching method (PSM) and composite control method (SCM). However, it is worth noting that, according to the construction history of Humen Bridge, the thod and the PSM method, this paper will choose the data-driven method SCM as the main analytical method. The next thing to do is to introduce the synthetic control method in detail, and then defines and explains the main variables and sample data.

Model Framework
The Synthetic Control Method (SCM) was first proposed by Abadie and Gardeazabal (2003). They used it to study the economic impact of terrorist activities in the Basque Country in Spain. Since then, this method has been widely used in the field of policy evaluation (Abadie et al., 2010;Ando, 2015;Wang & Nie, 2010;Liu & Fan, 2013;Liu & Wu, 2017;Liu & Zeng, 2018). The characteristic of the synthetic control method is that it is driven by data and selects suitable individuals from the control group to synthesize the "counterfactual" processing group. The basic idea of this method is to select a reasonable control group and predictor variables, synthesize a "counterfactual" treatment group based on the control group and the existing predictor variables, and then the difference between the "synthetic treatment group" and the "real treatment group" was compared to assess the impact of the policy. One of the prerequisites is that the sum of the weights of the control group areas participating in the synthesis "processing group" is 1, in order to avoid excessive extrapolation in the synthesis control. The following will briefly introduce the basic principles and models of the SCM.
Assuming that the observable region is J + 1, the time span of the sample is [1, T]. During the period of T0 (June 1997), individual 1 had an important traffic infrastructure opening (corresponding to the opening of Humen Bridge in the paper), in which 1 ≤ T 0 < T; Other J regions serve as potential control groups for region 1; After time T 0 , area 1 was constantly affected by the opening of the bridge. We use N it Y to represent the economic development of region I that was not affected by the opening of Humen Bridge at time T, and I it Y represent the economic development of region I that was affected by the opening of Humen Bridge at time T, in which α represents region I with the effect of Humen Bridge opening, and the following model is set: ; when 0 t T > , the economic effect of the humen bridge opening on region 1 at time t is 1 1 1 α is the treatment effect to be estimated in this paper. However, 1 I t Y region i can be observed after the bridge is opened. But the potential results of 1 N t Y unaffected by the bridge opening after T 0 cannot be observed. Therefore, referring to the practice of Abadie et al. (2010), the following factor model was used to estimate the "counterfactual result 1 N t Y " that region I was not affected by the bridge opening after T 0 .
δ is a fixed time effect with the same influence on all sample units; i Z is a (r × 1)-dimensional vector that contains observable variables that are not affected by the opening of the bridge in area i , and t θ is an unknown parameter is a (1 × F)-dimensional unobservable common factor vector; it ε is an error term, representing a temporary shock that cannot be observed in each sample area, and the mean is zero.
Here, the weight is set to non-negative and the limit is 1, the purpose is to avoid excessive extrapolation, so that the control group can synthesize a feasible synthetic experimental group within the observable data range. Each specific value of vector W represents the composite control for the first region, which is a weighted average of all regions within the control group. By weighting the variable values of each control group region, we can get: Suppose there is a vector group is a nonsingular matrix, then we can get: It can be proved that formula (4) tends to 0 under general conditions. Therefore, when can be used as the counterfactual result of the processing group (area 1), thus the estimated value of policy  it α can be obtained: The key to solving  it α is to find the weight vector W* that makes the above equation hold. We can determine the optimal weight vector W* by minimizing the distance between X 1 and X 0 W. X 1 represents the (k × 1)-dimensional feature vector Z of the target city before the Humen Bridge opening. X 0 is the (k × J) matrix containing k characteristic variables of the selected J control groups, and the characteristic vector Z is the main factor affecting regional economic growth or any linear combination of economic growth variables, which is the predictor variable mentioned above. In general, the distance function: where V is a (k × k)'s dimensions symmetric positive semidefinite matrices. Although the derivation process here is valid for any V, in fact the choice of V will affect the estimated mean square error. Here, the program developed by Abadie et al. is used to calculate V, so that the economic growth trajectory of the composite area is the same as that before the opening of the transportation infrastructure. The economic growth trajectory of the treatment group is similar. According to the idea of SCM, it is necessary to select a series of predictive variables it X to make the fitting effect and stability effect between the "synthetic treatment group" and the "real treatment group" better. With reference to the research of Liu and Hu (2010), Zhang (2012 and considering the availability of data, we determined the following predictor variables: The data is based on 1990 and adjusted according to the GDP index and per capita GDP index of each city to obtain the actual GDP and per capita actual GDP of each city. Except for RVGDP, RVPGDP and GOV, we took the logarithm of the rest of the variables. Table 1 shows the descriptive statistical results of the variables used in the empirical part.

Empirical Results on the Impact of GDP Growth
First of all, we verified the impact of the opening of Humen Bridge on the GDP of Dongguan. For the two indexes of lnrGDP and RVGDP of Dongguan before the opening of Humen Bridge, the weight of the city calculated by the synthetic control method is shown in Table 2 and Table 3. Weights reported in Table 2 indicate that rGDP trends in Dongguan are best reproduced by a combination of

Empirical Results on the Impact of per Capita GDP Growth
Next, our article verified the impact of the opening of the Humen Bridge on the per capita GDP of Dongguan. Table 4 and Table 5 show the city weights calculated by using the synthetic control method to synthesize the lnrPGDP and RVPGDP in Dongguan. Weights reported in Table 4 and Table 5     Further, by observing Figure 1 to Figure 4, we can see that these four indicators are in a steadily rising trend, and the gap with the "counterfactual" synthesis of Dongguan is gradually widening, which indicated that that the opening of Humen Bridge has a sustained growth effect on Dongguan's economic over time. There are also research articles showing that the opening of transportation infrastructure has a lagging period for regional economic benefits (Liu, 2010;Zhou & Zheng, 2012;Liu & Wang, 2017).

Permutation Test
In order to verify the robustness of the previous empirical results, learned from the practices of Abadie and Gardeazabal (2003) and Abadie et al. (2010), we use the Permutation Test to perform robustness testing. The idea of the test is to repeatedly use the synthetic control method to evaluate the impact of the policy implementation (the opening of the Humen Bridge) on the economic development of other control groups and assumed that all the control groups indepen- We calculate the estimated effect associated with each placebo trial. This iterative process provides us with the estimated gap distribution of the state without intervention, and then compares the actual effect found in the treatment group with the effect produced by the randomly selected control group area. If the difference between the actual effect and the random effect is large enough, the opening of the transportation infrastructure will have a significant effect on the economic growth of the treatment group. The ranking test method can visually show whether there are other areas that are similar to the real treatment group, and the probability of such a situation.
The Permutation Test requires that the synthetic groups and the real treatment group have a good fit before the opening of the transportation infrastructure. Therefore, in this placebo test method, if the calculated RMSPE value during the iterative process is too large, it means the sample's fitting effect of the time period before the opening of the transportation infrastructure is not good enough, so even if the difference of the explained variable of the sample is large in the later period, it cannot convincingly reflect the effect of the event. Therefore, if a sample has an unsatisfactory fitting effect between the synthetic object and itself before the event happened, we will no longer analyze the permutation test of this city. We excluded cities that had a RMSPE of more than 2 times the RMSPE of Dongguan. Figures 5-8 display the results for the placebo test, the black line represents Dongguan, and the dotted lines represent the cities of the control group.
Taking predictive variables lnrGDP and rvGDP as examples, it can be seen that before 1996, there was not much difference between Dongguan and the cities in the control group. However, after 1996, the gap between Dongguan and the control cities began to widen, and Dongguan's development path was located outside compare with other cities. The greater the difference value, the stronger the promotion effect brought by the opening of Humen Bridge. That indicated the opening of Humen Bridge had changed the GDP growth trend of Dongguan and brought positive effects. Because these two figures included 17 control cities, the probability of estimating a gap of the magnitude of the gap for Dongguan under a random permutation of the intervention in our data is 5.6%, a test level typically used in conventional tests of statistical significance. Similarly, it can also be proved that the growth trend of per capita GDP in Dongguan also showed positive and significant effects after 1996, the results are shown in Figure 7 and Figure 8. The results of the placebo test are basically consistent with the results of the empirical analysis in part IV 1 . 1 When analyzing the predictive variables of lnrGDP, the four cities of Guangzhou, Heyua, Jieyang and Yunfu with poor fitting degree were removed, leaving 17 sample cities. When analyzing the predictive variables of RVGDP, the four cities of Guangzhou, Shenzhen, Foshan and Yunfu with poor fitting degree were removed, leaving 17 sample cities. When analyzing the predictive variables of lnrPGDP, the four cities of Shenzhen, Zhuhai, Heyuan and Jieyang with poor fitting degree were removed，leaving 17 sample cities. When analyzing the predictive variables of RVPGDP, the two cities of Shenzhen and Zhuhai with poor fitting degree were removed, leaving 19 sample cities.

DID for Further Testing
Some researchers were all used the DID method as robustness test when studied the impact of policy effects on the economy (Liu & Fan, 2013;Liu & Wu, 2017;Liu & Zeng, 2018). Therefore, based on their approach, in order to ensure the robustness of the analysis, we used the DID with fixed effects to test the robustness of the predictor variables. The econometric model is set as follows: Equation (8)  = 1, before the opening of the bridge = 0). The interaction term represents the effect of the treatment group after the bridge was opened, and the coefficient β is the treatment effect that the model focuses on. In addition, it X is the set of control variables, i δ is the regional fixed effect, t γ is the annual fixed effect, and it ε is the random error term. Table 6 reports the estimated results of the Humen Bridge opening on Dongguan's lnrGDP and RVGDP. Columns 1 and 3 use Equation (8)   Furthermore, this article replaces the indicators to measure regional economic growth with GDP growth rate (g 1 ) and per capita GDP growth rate (g 2 ). Columns 5 and 6 in Table 6 and Table 7 report the estimation results. The coefficient of the interaction term reflects the impact of the completion of Humen Bridge on the GDP growth rate and per capita GDP growth rate of Dongguan.
The interaction term's coefficients obtained are all significantly positive, indicating that the opening of Humen Bridge has a positive and significant impact on the GDP growth rate and per capita GDP growth rate of Dongguan city.

Conclusions and Policy Recommendations
Most of the existing literature proved that transportation infrastructure had significant effects in promoting economic growth, but most of the studies are based 2 A similar situation had occurred when Liu Naiquan and Wu You (2017) used PSM-DID for robustness analysis. Therefore, in further analysis, we eliminated Guangzhou and Shenzhen, two first-tier cities with excellent economic development, to avoid the impact of special cities. Then we still use the DID with fixed effect to estimate, and the result is that the interaction term is 0.18 and is significant at the 1% significance level. Therefore, the test results estimated by the DID method verify the robustness of the above research conclusions.   From the perspective of overall regional development, it is necessary to coordinate relatively balanced development between regions. Imperfect transportation infrastructure may have a negative effect on the economic development of underdeveloped areas. Because as the flow cost decreases, the production factors of underdeveloped areas will flow to relatively developed areas, which makes the economic agglomeration effect of developed areas more obvious. Therefore, it is necessary to prevent the oversaturation of the central area caused by the "siphon effect" 3 of the transportation infrastructure, and to prevent the excessive loss of economic factors in the surrounding areas. Reasonably coordinate the layout of the transportation network and use the external functions of the transportation infrastructure to help upgrade the "central-peripheral" industry.
Although we have argued in the preceding article that the opening of Humen Bridge has a positive impact on the economic growth of Dongguan. It is worth noting that there are still some deficiencies worth discussing in this paper. One caveat of the article is that the control cities we found were only prefecture-level cities in Guangdong province. Due to the huge development gap between PRD (the Pearl River delta) and non-PRD, these control cities were not good enough to give weights to the synthetic city. Selecting the control cities with similar economic development status to the experimental group, and estimating the 3 Siphon Effect: originally refers to the physical phenomenon that liquid will flow from one side with high pressure to the other side with low pressure due to the existence of gravity and potential energy difference between liquid molecules. The "siphon effect" in economics refers to the fact that with the continuous development of a central city, when it grows into a large city or even a first-tier super-large city, it will attract talents, investment, population, information and other high-quality resources from disadvantaged areas due to the high concentration of high-quality medical care, education, infrastructure resources and abundant capital. For big cities, this will lead to further improvement of competitiveness and further expansion of scale.

Conflicts of Interest
The author declares no conflicts of interest regarding the publication of this paper.