A Study of Knowledge Spillovers within Chinese Mega-Economic Zones

2.1 JJJ

The JJJ region, known as Beijing–Tianjin–Hebei, is China’s capital economic area, in which Beijing, Tianjin, and Shijiazhuang are the core areas. The JJJ region is located in the heart of Bohai Sea in northeast China. It is the largest and most dynamic area in northern China. In 2019, the GDP of JJJ region in China reached 8458.008 billion-yuan, accounting for 8.5% of the nation.

1. Beijing, the capital of China, is the national political, cultural, and innovation center.

2. Over the past years, JJJ did poor in realizing coordinated development, and the gap between Beijing and its neighboring cities within JJJ has been obviously widening. Some reasons are:

   1. Serious administrative barriers between Beijing and other cities, and the development of Beijing is always at the cost of the others.

   2. A great amount of resources (financial resources, human resources, energy, etc.) actively or passively flow into Beijing;

   3. Poor industry connection and huge regional gap allows little cooperation between Beijing and other cities.

Based on this, the coordinated development of Beijing–Tianjin–Hebei has become the national strategy nowadays. Some political measures^[1] (in terms of innovation) include:

1. Build JJJ innovation system where Beijing acts as the origin of innovation, Tianjin conducts innovation transformation, and Hebei applies and popularizes it;

2. Share the innovation resources and technological achievements, strengthen the training and exchange of R&D personnel, and encourage innovation communication and cooperation. For example, set up the JJJ Alliance of Higher Education that encourages Beijing and other cities to cooperatively run universities and construct disciplines; encourages Beijing and other cities to cooperatively build innovation zones.

2.2 YRD

The YRD is a triangle-shaped metropolitan region generally comprising of Shanghai, southern Jiangsu province, and northern Zhejiang province. With less than 4% of its territory, the YRD has created nearly one-fourth of China’s GDP and one-third of its trade volume. The YRD region is one of the most active, open, and innovative regions in China.

1. Shanghai: an international center for economic, finance, trade, and shipping, and striving for global innovation center nowadays.

2. Over the past years, YRD did well in realizing coordinated development and Shanghai well pushes the development of its neighboring cities.

YRD has always been the important business and trade centers since olden days. The cities here, as well as Shanghai and its neighbors, are willing to and capable of keeping close contact with each other in various aspects (technology, human capital, finance, etc.), due to their historical links, open and equal atmosphere, good industry connection, and more importantly, the impetus of market economy. In other word, Shanghai pulls the development of its neighboring cities, and in return, the cities within YRD support improvement of Shanghai.

1. Some political measures towards a regional convergence are:

   1. Speed up the development of modern service industry with Shanghai as the international shipping center and international financial center, and strive for an industry structure with service industry as core (The Guidance of State Council on Further Promoting the Reform and Opening up and Economic and Social Development in the Yangtze River Delta, approved in 2008).

   2. Give full play to the core role of Shanghai as the international center for economic, finance, trade, and shipping, and speed up its development of modern service industry and high-end manufacturing industry, and other cities should well take advantage of this in order to coordinately promote the industrial upgrading, technological innovation, and further development (Regional Plan for Yangtze River Delta, approved in 2010).

2.3 PRD

The PRD metropolitan region is in the south-central part of Guangdong province, consisting of nine cities. The total area of PRD accounts for less than one-third of Guangdong province, while it has 53.35% of the population and 79.67% of the GDP of Guangdong province. It is one of the biggest urban agglomerations with the strongest innovation ability in China. Today, as a national strategy, Guangdong–Hong Kong–Macao Greater Bay Area is formed in this area.

1. There are two core cities in PRD: Guangzhou and Shenzhen. Guangzhou, the capital of Guangdong, works for the concentration of high-end elements, technological innovation, cultural leading, and providing comprehensive service. Shenzhen is the national economic center and the national innovation city.^[2]

2. Like YRD, PRD did well in realizing coordinated development in the past. The close links between cities, and core cities to its neighbors is also mainly driven by the market economy (without so much political issues). There are many specific cases demonstrating this positive cooperation between core cities and its neighbors. For example, Shenzhen benefits its cities nearby by the transfer of industries, technology, talents, and so on.

3. During 2016–2020, a regional innovation system (“1 + 1 + 7”) has been set up where the leading roles of Guangzhou and Shenzhen were more prominent, the mechanism of regional innovation cooperation has been greatly improved and an efficient division-cooperation system has been established (The Plan for the Construction of National Independent Innovation Demonstration Zone of Pearl River Delta (2016—2020), approved in 2016)

3.1 Specification

In order to investigate the spillover effect within three MEZs, we develop a knowledge production function rooted in models of endogenous economic growth in open economies (e.g., Grossman & Helpman, 1990):

Patents in city i ( P i t ) depends on recent own research effort ( E i t − 1 α ) (R&D expenditures), available knowledge stock ( ( K ⁎ ) i t − 1 β ), and controls ( A i t − 1 ) including total factor productivity in R&D, local specificities, and other confounding effects. The available knowledge stock ( K ⁎ ) is given by:

The available knowledge stock consists of own knowledge developed in the past ( K i t − 1 β 0 ), knowledge that spilled over from other cities ( K j t − 1 β i j ), knowledge that spilled over from foreign investors ( FDI i t − 1 γ F ), and knowledge that spilled over from international trade partners ( TR i t − 1 γ T ). Inter-city weights are measured with geographical distance adopting the inverse exponential distances between city pairs ( w i j G = e − δ D i j ) , and technological distance (dissimilarity in technological specialization) using uncentered correlation of sectorial R&D expenditures across city pairs.

Taking the logarithmic form of production function, we setup a following spatial lag model with time–space recursive form like Anselin (2007), in which the dependence relies on a weighted average of neighboring and the value of itself in time t−1.

where the dependent variable P i t denotes the innovation output of city i in year t measured as the number of total patent applications filed in each of the 270 regions in 7 years. W is the spatial weight matrix and ρ is the spatial autocorrelation coefficient. The one-lagged dependent variable is included in the specification in order to capture the lagged effect of innovative activity. Other controls include patent stock (Bilgin, Lau, & Karabulut, 2012), R&D expenditure (Shanga, Poonb, & Yuec, 2012), trade volume, FDI stock (Hanley, Liu, & Vaona, 2015), employees (Gogokhia & Berulava, 2021), and GDP per capita. Time fixed effects and individual fixed effects are also introduced in the specification. Table 1 describes the key variables in detail.

3.2 Data

This study adopts a municipal-level panel dataset over the period of 2007–2012. Since the patent data are not available from China Statistical Book, we manually collected the patent data from the statistical bulletins released by the prefectural governments of the three MEZs through their websites. 270 prefecture-level cities are included in the sample.^[3] The data of other variables are acquired from China Statistical Book. Table 2 defines the variables of main interest. Table 3 reports the statistical descriptions of key variables.

3.3 Method

For the sake of choosing a suitable empirical method, we compute the spatial autocorrelation statistics using Moran’s I index to test the spatial correlation of dependent variable. The Moran’s I index is given by:

where S ² is the sample variance, W _ij is the spatial weights matrix. The value of I is between 0 and 1, when the value is greater than 0, there is a positive spatial correlation, and when the value is less than 0, there is a negative spatial correlation.

Table 4 reports the Moran’s I index by year and different distances. As seen in this table, innovation activity measured by patent or patent stock in all years shows positive significant spatial dependence. And this spatial correlation increases with year. However, the values of Moran index are very low, suggesting that the spatial correlation is weak. The reason why the values of Moran Index are low lies in that the technical development across China is extremely uneven, which leads to the low average spatial technical dependence. Considering the existence of spatial correlation of dependent variable, we think that Spatial autoregressive (SAR) model outperforms other spatial models.

Furthermore, we check the spatial effects by executing several LM tests including LMERR, LMLAG, and robust R-LMERR and R-LMLAG. Table 5 reports the test results. According to Anselin et al. (1997), if LMLAG is more significant than LMERR in statistics, and R-LMLAG is significant while R-LMERR is not, it can be concluded that spatial lag model is the suitable model. Otherwise, spatial error model is more appropriate.

As shown in Table 5, the statistical values of LMLAG and R-LMLAG are greater than that of LMERR and R-LMERR, and R-LMLAG is significant, while R-LMERR is not. The test result suggests that spatial lag model is a better option.

4.1 Spillovers within All Regions

We first look at the spillovers within all regions as presented in Table 6. Columns 1 and 3 report within estimators of dynamic panel data model estimated by system generalized method of moments (GMM) for reference. Column 2 reports the estimation results using geographic distance as spatial weights matrix. Column 4 employs technological distance as spatial weights matrix. As shown in this table, the spatial autoregressive parameters ρ (rho) in columns 2 and 4 are significant at 1% level, suggesting there are remarkable spillovers between cities in China. However, the magnitude of the spatial parameter weighted by geographic distance is much lower than that weighted by technological distance, which means that technological similarity plays a more important role on the spillovers than geographic proximity and the spillovers between cities with higher technological similarity are stronger. One reason is that cities with lower technological distance have higher absorptive ability, which promotes the spillover effect. The results confirm the findings of previous studies on absorptive ability.

Concerning other controls, the 1-year lagged patent, patent stock, GDP per capita, and R&D expenditure are found to have significant impact on the number of patent applications. While the FDI, trade volume, and the number of employees seem to have no significant relationship with regional innovation.

4.2 Spillovers within Three MEZs

In the next step, the spillovers within three MEZs are examined in this section as demonstrated in Table 7. Columns 1, 3, and 5 report the estimation results using geographic distance as spatial weights and the rest uses technological distance. As reported in this table, the spatial autoregressive parameters ρ (rho) in all columns are found to be significant, which means that spillovers exist across all these three metropolitan regions. More importantly, the magnitude of spatial coefficient (rho) of the PRD is the largest, followed by the YRD, and the JJJ. A comparison of the coefficient magnitudes reveals that spillovers between cites in the PRD are the greatest among the three economic zones. This result may be explained as follows. First, there is higher concentration of manufacturing industries in the PRD, which gives birth to more innovation activities, because manufacturing activities are inherently associated with more innovations. Second, the PRD is composed of cities from the same province, which means a higher proximity in geography, institution, and technology. Higher proximity is definitely helpful to the spillovers between cities. Finally, there are two metropolises in the PRD including Guangzhou and Shenzhen. Guangzhou is a cultural and political center where lots of universities are located. Shenzhen is an immigrant city and the manufacturing base. Different orientations of Guangzhou and Shenzhen lead to complementarity rather than competition. The complementarity will increase innovation cooperation and spillovers in this area. On the contrary, a counter-intuitive fact is that the spillovers among JJJ are the lowest. One possible explanation is that Beijing, the national capital, attracts the innovation resources and talents from neighboring cities like a vortex because of its political and educational resources, which reduces the absorptive ability of neighboring cities.^[5] Furthermore, looking at results within groups, we also find that spatial coefficients weighted by technological distance are larger than that weighted by geographic distance in all three economic zones.

4.3 Spillovers from Three Core Cities

Moreover, we continue to investigate the spillovers from three core cities with the most patents including Beijing, Shanghai, and Shenzhen, which can help us to explore the mechanism of the heterogeneous regional spillovers studied in last section. To this end, we setup an empirical model as given by:

where ∑ j = 1 j ≠ i N w i j ln P j t − 1 denotes the spillovers from three core cities to others containing four items as described in Table 8. ksgdc represents knowledge spillover by geographical distance from three core cities to rest of the cities. kstdc indicates knowledge spillover by technological distance from three core cities to other cities. nksgdc captures knowledge spillover by geographical distance from three core cities to its neighbors within the same economic zone. It is defined as the interaction term of neighbor dummy and ksgdc . Similarly, nkstdc means knowledge spillover by technological distance from three core cities to its neighbors within the same economic zone. It is also the interaction term of neighbor dummy and kstdc .

The specification is a dynamic panel model rather than spatial model. The coefficient α measures the direction and importance of spillover like a normal coefficient rather than spatial autoregressive parameter. The dependent variable and other controls are defined as spatial model. System GMM method is implemented to estimate the coefficients and fixed effect model is adopted as suggested by the Hausman tests.

Table 9 reports estimation results of the spillovers from three core cities. Columns 1, 3, and 5 report the estimation results weighted by geographic distance. Columns 2, 4, and 6 report the estimation results weighted by technological distance. As illustrated in this table, the spillover effect of Beijing to the rest is found to be insignificant when using geographic distance as weights, but it is significant when taking technological similarity into account as suggested by the estimation results in columns 1 and 2. Moreover, an unexpected finding is that there are significantly negative spillovers from Beijing to neighbors weighted by geographic distance, but they turn to be significantly positive when weighted by technological distance. The negative spillovers of Beijing to neighbors could explain the lowest spillovers to great extent within JJJ found in Section 4.2, which can be ascribed to the high cluster of innovation resources in Beijing. As expected, the spillovers from both Shanghai and Shenzhen to the rest of the cities or their neighbors are found to be significant and positive. In addition, the spillovers from Shenzhen are the strongest among three core cities, which are consistent with biggest spillovers within the PRD.