A comparative study of innovation agglomeration and spatial evolution in Pearl River Delta and Yangtze River Delta

The increasingly intensive innovative activities enable a more straightforward tracing of knowledge spillover, by looking at the spatial agglomeration of innovative activities. Existing research looked at spatial patterns of innovative activities, but few examine the spatial evolution of the innovative agglomeration. This paper employs location quotient, Moran's Index and the method of standard deviation ellipse to analyse the spatial evolution of innovation agglomeration in the Pearl River Delta and Yangtze River Delta from 2004 to 2016. Research findings include: (i) high-tech industry agglomeration fosters a local knowledge production network which promotes local knowledge spillover; but an existing local knowledge production network is not the precondition of agglomeration; (ii) local knowledge spillover occurs in a "core" area and there is a tendency that the core area is shrinking and being more distinctive; but individual cities' breakthrough on innovation promotion triggered by intercity competition contributes to breaking the shrinking process; and (iii) the evolving of spatial patterns of innovative activities follows that of high-tech industries agglomeration in general, but some mismatches between them suggest the spatial distribution of high-tech industries cannot take full advantage of the innovative resources and facilities in the two regions. Accordingly, it is recommended that local policies would encourage better connections between high-tech industries with other local facilities and resources of innovation, to strengthen the spatial connections of innovative activities between cities within the region to provide sustained support to innovation in the region, but maintain the intercity competition mechanisms to secure the sources of breakthrough development on innovation. Las actividades innovadoras cada vez más intensivas permiten un rastreo más directo del spillover del conocimiento, mediante la observación de la aglomeración espacial de las actividades innovadoras. La investigación existente ha estudiado los patrones espaciales de las actividades innovadoras, pero pocos estudios examinan la evolución espacial de la aglomeración de la innovación. En este artículo se emplea el cociente de localización, el índice de Moran y el método de la elipse de desviación estándar para analizar la evolución espacial de la aglomeración de la innovación en el delta del río de las Perlas y el delta del río Yangtze desde 2004 hasta 2016. Entre los hallazgos de la investigación están: i) una elevada aglomeración de industrias tecnológicas fomenta una red de producción de conocimientos locales que promueve spillovers del conocimiento local, pero la existencia de una red de producción de conocimientos locales no es una condición previa de la aglomeración; ii) los spillovers del conocimiento local se producen en una zona "núcleo" y existe una tendencia a que la zona núcleo se reduzca y se haga más distintiva, pero los avances de las ciudades individuales en la promoción de la innovación impulsada por la competencia interurbana contribuye a destruir el proceso de reducción; y iii) la evolución de los patrones espaciales de las actividades innovadoras es parecido, en general, al de la aglomeración de las industrias de alta tecnología, pero algunos desajustes entre ellos sugieren que la distribución espacial de las industrias de alta tecnología no puede aprovechar plenamente los recursos e instalaciones innovadoras de las dos regiones. En consecuencia, se recomienda que las políticas locales fomenten una mejor conexión entre las industrias de alta tecnología y otras instalaciones y recursos locales de innovación, para fortalecer las conexiones espaciales de las actividades innovadoras entre las ciudades de la región, a fin de prestar un apoyo sostenido a la innovación en la región y mantener a la vez los mecanismos de competencia interurbana con los que asegurar las fuentes de desarrollo de los avances en materia de innovación. イノベーション的活動の空間的集積を見ることにより、イノベーション的活動の集中度が高まるにつれ、より直接的な知識のスピルオーバーの追跡が可能になる。既存研究では、イノベーション的活動の空間パターンは検討されているが、イノベーションの集積の空間的な進化を検討した研究はほとんどない。本稿では、立地係数、Moran指数および標準偏差楕円法を用いて、2004~2016年までの珠江デルタと長江デルタにおけるイノベーション集積の空間的発展を分析した。以下の知見が得られた。1)ハイテク産業の集積により、地域の知識のスピルオーバーを促進する地域の知識生産ネットワークが発達するが、既存の知識生産ネットワークは集積の前提条件ではない。2)中心(core)エリアでは、地域の知識のスピルオーバーが生じており、中心エリアが縮小し、より特徴的なものとなる傾向がある。しかし、都市間の競争をきっかけとした、各都市のイノベーションの促進に関するブレークスルーは、この縮小の抑止に寄与している。3)イノベーション的活動の空間パターンの進化は、通常はハイテク産業の集積の空間パターンの進化に従うが、それらのいくつかのミスマッチから、ハイテク産業の空間分布が2つの地域におけるイノベーションの資源と施設を十分に活用できないことが示唆される。したがって、ハイテク産業と他の地域の施設やイノベーションの資源とのより良い関係を促進し、地域におけるイノベーションへの持続的支援を提供するために地域内のイノベーション的活動の空間的な関係を強化しながらも、イノベーションの突破口となる発展の源を確保する都市間競争メカニズムを維持する地域政策の実施が推奨される。 Investigations on the knowledge spillover in industrial and urban agglomerations have a long history in the economic graphical traditions. Earlier research in the 20th century saw knowledge spillover as a critical but ambiguous element of externalities that, in the form of innovation, promotes the utilization efficiency of production factors and enhances the competitiveness and economic performance of firms in the agglomeration areas (e.g., Eberts & McMillen, 1999). With technology developed and industries re-structured and upgraded in the past decades however, it is not surprising to find that knowledge per se has increasingly become a pronounced production factor (Audretsch & Feldman, 2004; Knoben, 2009). Recent research focus shifts to the agglomeration of knowledge, high-tech or innovation activities, concerning how these agglomerations contribute to the performances of the firms, the industries, and the region. Research gaps are noticed, and to be filled by this paper, in two perspectives. First, existing empirical research on innovation agglomeration and knowledge spillover come up with various and often conflicting conclusions, which partially owe to the incomparability between studies based on different databases in different countries. So, this research investigates two urban agglomerations within similar political and institutional contexts and comparable statistic datasets, providing comparative analysis and more robust conclusions regarding how the knowledge externalities work in agglomerated regions in the given institutional context. Second, although the spatial locations of knowledge innovative activities are frequently discussed, few articles look at the evolving process of its spatial pattern. This paper fills this gap in the literature and contributes to the exploration of the spatial mechanism of knowledge spillover. The urban agglomeration of the Pearl River Delta (hereinafter referred to as the Pearl River Delta) and the urban agglomeration of the Yangtze River Delta (hereinafter referred to as the Yangtze River Delta) in China are the two regions to be investigated and cross-compared in this paper. They are now the primary locus for high-tech industrial clusters in China. Cities in these regions started their economic restructuring in the early 2000s and have been trying to attract high-tech industries for nearly two decades. The innovation agglomeration processes in the two regions were accelerated by China's innovation-driven strategy introduced in 2012, after which their advantages (over the rest parts of the country) on well-established infrastructures and institutional conditions for knowledge-based development become evident. Therefore, the spatial processes of the agglomeration of knowledge innovation in Yangtze River Delta and Pearl River Delta in the past decades became the ideal objects of this research, to investigate how the innovative agglomeration is come into being and how the knowledge spillover works in the processes. To be more specific, this paper answers the following questions: Statistical data of high-tech industries and patent applications of cities in the two regions (nine cities in the Pearl River Delta and 16 cities in the Yangtze River Delta) from 2004 to 2016 are used and are extracted from statistical yearbooks of those cities. Location quotient, Moran's Index and the method of standard deviation ellipse are employed to examine the degrees of the specialization of high-tech industries, spatial autocorrelations of innovation outputs, and the spatial characteristics of innovation outputs, respectively. The analysis led to three main findings regarding the spatial evolution of innovation agglomeration in the two regions. First, high-tech industries agglomeration fosters a local knowledge production network hence promotes local knowledge spillover; but an existing local knowledge production network is not the precondition of agglomeration. No evidence in the investigated period suggests that industry agglomeration may harm innovative performance. Second, local knowledge spillover occurs in a 'core' area and there is a tendency that the core area is shrinking and being more distinctive; but individual cities' breakthrough on innovation promotion triggered by intercity competition contributes to breaking the shrinking process. Third, the evolving of spatial patterns of innovative activities follows that of high-tech industries agglomeration in general, but some mismatches between them suggest the spatial distribution of high-tech industries cannot take full advantage of the innovative resources and facilities in the two regions, which typically include, for example, the universities and research centres. These findings provide more nuances to understand the spatial process of innovation agglomeration, regarding the relationships between the high-tech industries clustering, the spatial characteristics of innovative activities and that of knowledge spillover in the agglomeration. In addition, policy recommendations are generated for the two regions regarding their innovation-driven development strategies, that local policies would encourage better connections between high-tech industries with other local facilities and resources of innovation, to strengthen the spatial connections of innovative activities between cities within the region to provide sustained support to innovation in the region, but maintain the intercity competition mechanisms to secure the sources of breakthrough development on innovation. Research on the influences of knowledge spillover in an agglomerated area can be traced along with the enduring debates between Marshall-Arrow-Romer (MAR) externalities and Jacobs externalities, also known as localization or specification versus urbanization or diversification (Abdel-Rahman & Anas, 2004; Van der Panne, 2004). Supporters of the former defence for the benefits of externalities arise from knowledge transfers within an industry (e.g., Berliant, Reed, & Wang, 2006; Chen, 2003), while those of the latter argue for externalities that arise from knowledge transfers between industries (e.g., Keely, 2003). Henderson, Kuncoro, and Turner (1995) note that MAR externalities play a bigger role in traditional industries, while Jacobs externalities are more important in modern high-technology industries. Cheng and Liu's (2015) China-based research analysed the panel data of 285 prefecture-level cities in China Industry Business Performance Database and reaches similar conclusions. Harrison, Kelley, and Gant (1996) investigate the behaviours of individual innovative firms and conclude that urbanization is more important than localization in explaining spatial patterns of innovation and economic development. Paci and Usai's (2000) analysis based on the databank on innovation and production across the Italian local labour system note that both MAR externalities and Jacobs externalities work, and the dimension/scale of the local system and the types of industry in the local system (high or low tech sectors) matter. Capozza, Salomone, and Somma's (2018) study on Italian industrial policy intervention that fosters innovative start-ups re-confirms Paci and Usai's (2000) conclusions, and adds that other elements in the cities, such as the university and the urbanization process, affect the local benefits to innovative start-ups. In contrast, Knoben's (2009) study reveals that urbanization economies harm the innovative performance of firms, whereas localization economies generally have a positive effect, and that being located in the area with industrial agglomeration is not necessarily beneficial for the innovative performance of a firm. Shearmur's (2012) Montreal-based research supports Knoben's (2009) argument and suggests that the most intensive knowledge-intensive business services (KIBS) innovators tend to locate away from where employment and KIBS agglomerated. Nevertheless, Fang (2020) finds that the agglomeration of firms does actively select those with better innovative performance, for that non-innovators are less likely to survive in agglomerations. Wersching (2007) notes that young enterprises have a better incentive for agglomerating. Samusenko, Bukharova, Rutsky, and Maslodudov (2012) also find that entrepreneurship and innovation development trends relate to the organizing and functioning of the regional innovation system. That echoes Ortega-Argilés, Moreno, and Caralt's (2010) Spain-based conclusion that R&D decision-making tends to be increasingly conservative as it develops, and the conservative effects may spill over into the region in ways that reduce innovative practices. That may partially explain why small-firms are often the engine of innovative activities, while large firms, by working with public research funds and institutes, contribute to systematically organize and/or foster innovative activities in the region (Capozza et al., 2018; Cooke, 2010; Haynes & Shibusawa, 2009). In that sense, the urban agglomerations in China that start the innovation-driven strategy only recently and focus on emerging enterprises with front-edge high-techs are in the early and raising stage of the life circle of the innovation performance of agglomerations, and shall have sufficient engines of innovative activities. Furthermore, globalization is challenging conventional explanations about agglomeration. Krugman (1991) explains why geographical agglomeration remains necessary from the perspective of trades and transportation costs. Regarding knowledge spillover in the region that is both globally and locally connected, Glaeser and Ponzetto (2007) note that the developed communication and transportation technology weakens the advantages of firm agglomeration, but, by allowing the innovations to be used throughout the world, the improvement of transportation and communication technology could increase the returns of innovation. Hence, agglomerations of firms that produce ideas have far better performance than agglomerations of firms producing goods. Kekezi and Klaesson's (2020) research shows that the distance decay of spillovers is fast, and there are spatial competition effects between two groups of knowledge-intensive business services (KIBS) over a longer distance. Regarding the relations between innovation agglomeration and regional growth, Akhmedova and Zhogoleva's (2018) Russia-based research shows that the urban planning development policy of Samara–Togliatti agglomeration will guide the establishment and agglomeration of innovative cluster infrastructure in this region. That, however, is against Dieperink and Nijkamp's (1988) conclusion that no evidence supports a close correlation between R&D infrastructure and the spatial dispersion of innovation. Some China-based research is very much policy-driven or policy-based as well, aiming to find the policy approach of promoting the innovation-driven regional growth (e.g., Zhang, 2019; Zhou, Xu, & Wang, 2018). Choi (2020) admitted that the spatial agglomeration of emerging technology and innovative activities is not automatically determined by pre-existing conditions, but local universities and existing knowledge about the emerging technology have a significant and positive impact on firms' innovative activities. Given the various and complicated ways and perspectives of investigation, the existing literature has at least made one thing clear, that both the MAR externalities and the Jacobs externalities work to knowledge innovation in urban agglomerations, but with different roles to play, and the effect of knowledge externalities depends on the relative locations in an agglomeration of firms, as well as the types of firms agglomerated. Varga, Pontikakis, and Chorafakis (2014) further note that agglomeration and scientific networking are neither substitutes nor complement but operate at distinct parts of the knowledge production process. So, the question remains is how the knowledge production network works in the agglomeration, and what is its role in the regional growth. The various and often conflicting conclusions among existing research, however, reveal one of the weaknesses of current empirical research on knowledge innovation agglomeration, that is the incomparability between studies, especially between investigations based on different databases in different countries. This paper draws on the same types of statistic data collected from the same type of databases generated from the two regions with the same national political context and similar cultural environment, hence it contributes to providing comparative analyses of knowledge externalities in two urban agglomerations. Also, although it has been widely noted that the spatial location in the agglomeration area matters to knowledge innovative activities, few articles look at the evolving process of its spatial pattern. Yang, Sun, and Liang (2009) analysed the concentration degrees and change tendency of China's high-tech industries, using H-index, space Gini coefficient and EG coefficient. Long, Zhu, and Cao (2014) use standard deviational ellipse to analyse the GDP density of 88 counties in Guizhou Province and find the existence of polarization effect in Guizhou Province. This paper moves further to conduct a quantitative comparative analysis on the spatial evolution process of innovation agglomeration in the Pearl River Delta and the Yangtze River Delta, and raises policy suggestions for the two regions that further promote the role of innovation agglomeration in regional growth. As illustrated above, this paper aims to investigate the influence of knowledge externalities on innovative activities in the region and hence the regional growth. The two regions investigated—the Pearl River Delta and the Yangtze River Delta in China—are gradually shifting towards the knowledge-based economies in the past decades, with high-tech industries attracted and fostered, and favoured policies to attract talents and significant investments on innovative facilities. Therefore, they provide the two typical cases for this research to investigate whether and how the knowledge spillover works in the current urbanization processes that high-tech industries play an increasingly more significant role in city and regional economic growth. The comparison between analyses of the regions also strengthens the robustness of results. The analysis framework of this research incorporates three parts. The first step is to examine whether and to what degree the high-tech industries agglomerated in the two regions. Location Quotient is employed here to examine the degree of the specialization of high-tech industries in these regions against the national average. It is worth noting here that cities with a higher degree of the specialization of high-tech industries identified in the two investigated regions cannot be understood at the economies with specialized industry clusters; on the contrary, the emerging of specialized high-tech industries adds to the diversification of industries in the cities. So, the knowledge externalities discussed in this paper are Jacob externalities rather than the MAR ones. Second, spatial autocorrelations of innovation outputs within each region are analysed, to see if they are spatially interrelated and agglomerated, which leads to discussions on the emerging/establishing of local knowledge production network within the region. Finally, the spatial characteristics of innovation outputs as well as their evolution are pictured through standard deviational ellipse analysis, identifying the spatial routines of the innovation output agglomeration in the two regions. Comparing the spatial routines of innovation outputs agglomeration with the evolving spatial patterns of high-tech industries agglomeration, the interactions/mutual influences between the industrial agglomeration and innovative activities agglomeration are analysed, through which the influence of knowledge spillover is discussed. This paper primarily focuses on the spatial patterns of innovation agglomeration, contributing to spatially describe and explain the knowledge externalities, but does not conduct underlying causal analysis in the institutional contexts. More regressions should be implemented in further research to investigate and testify the causality of the spillovers. Several sets of data are employed in the analysis described above, and all of them are obtained from the National Statistical Yearbooks and statistical yearbooks of the individual cities and provinces in the Pearl River Delta and Yangtze River Delta released between 2005 and 2016. Namely, they are the Statistical Yearbook of Guangdong Province, Statistical Yearbook of Shanghai Municipality, Statistical Yearbook of Jiangsu Province, Statistical Yearbook of Zhejiang Province, as well as that of the nine cities in the Pearl River Delta and the 16 cities in the Yangtze River Delta. With the statistical data released from 2005 and 2016, the investigated period of this research is between 2004 and 2015. First, the high-tech industries discussed in this paper refer to the industries of electronic and information technology, mechatronics technology, biotechnology, new materials science and technology, and efficient and energy-saving technology and environmental protection technology. The output values of these industries altogether constitute the output values of high-tech industries. The proportions that the output value of high-tech industries in each of the two regions account for in the national total output value of high-tech industries are calculated to examine the significance of the high-tech industries in the economic growth in the two regions. Then, the "R&D personnel" of the "scientific and technological activities of industrial enterprises above designated size" in each city of the two regions is employed for the location quotient calculation. Second, the amounts of patent applications and that of patent applications granted each year are used as indicators of innovation outputs. The patent applications data is used in a spatial autocorrelation test, while the patent applications granted data is used in spatial pattern analysis, employing the standard deviational ellipse method. Moran's I values usually vary from −1 to 1; when Moran's I approaches 0, it is considered that the observed values are random or have no spatial autocorrelation; when Moran's I approaches 1 or −1, it indicates that the observed values are clustered, with positive correlation or negative correlation. The semi-major axis of the ellipse represents the direction of data distribution, while the semi-minor axis represents the range of data distribution. The results can be interpreted as follows: Figure 1 shows the proportions of the output value of high-tech industries of the two regions take in the national totals between 2004 and 2015. The total output value of the two regions accounts for over 40% of the national total throughout the investigated period, and it exceeded 50% in 2009, 2010 and 2011. The proportions that the output value of the Yangtze River Delta took during the investigated period went up between 2004 and 2012 and then down afterwards; while that of the Pearl River Delta has a contrast fluctuation, that went down first and then up. Both curves have their inflection points in 2012. Figure 2 shows the location quotient of cities in the Pearl River Delta during the investigated period. The curves of these cities have similar fluctuation in general, hitting a peak in 2008, declining in 2009 and remaining stable in the rest of the years. Shenzhen has the greatest LQ (over 2 in 2007 and 2008 and around 1.9 in the other years) among these cities and has a large gap with other cities, suggesting its much higher degree of specialization of high-tech industries than not only the national average but also all the other cities in the Pearl River Delta. The other metropolitan city, Guangzhou, however, show a low degree of specialization high-tech industries, with LQ closing to 0.5; the same as Zhaoqing, a less economically developed city in this region. That of the rest of the cities fluctuates around 1, roughly equivalent to the national average. Figure 3 shows the location quotient of high-tech agglomeration in the cities of the Yangtze River Delta. At the top of the figure, there are three cities—Shanghai, Hangzhou and Wuxi—whose LQ significantly greater than the others, at the level around 1.75, though the LQ of Hangzhou and Wuxi consistently decreased and Wuxi declined faster than Hangzhou. In the middle, there are six cities—Jiaxing, Huzhou, Changzhou, Ningbo, Taizhou and Shaoxing—whose LQs are sustained as the level above 1, regardless of some fluctuations. While the other seven cities have LQs lower than 1, suggesting the lower degrees of specification than the national average. The high-tech industries in both regions are agglomerated in one or a few "core" cities, and both regions have cities with the lower-than-average degree of specification of high-tech industries and those around or slightly above average. The spatial distributions of high-tech industries in cities within each of the individual regions make them the suitable urban agglomerations to investigate the knowledge spillover, and the similarity and slightly difference described make them comparable and worth to be compared. Table 1 lists the Moran's I indexes of the Pearl River Delta from 2004 to 2015, based on annual patent application amounts. The index increased rapidly from 0.1189 in 2004 to 0.3001 in 2010, suggesting the innovation activities in this region had been increasingly interrelated between cities. Then there are fluctuations afterwards and decreased to 0.2635 in 2015, which might be related to the recent industrial restructuring and upgrading within the Pearl River Delta. Nevertheless, the spatial correlation of technological innovation pictures the signs of knowledge spillover in this region. Table 2 presents the Moran's I indexes of the Yangtze River Delta during the investigated period. The index increased −0.0197 in 2004 to 0.2700 in 2010, decreased to 0.1171 in 2013 and 0.1241 in 2014, and then back to 0.2018 in 2015. The index of this region at the start of the investigated period is much lower than that of the Pearl River Delta. The results suggest the innovative activities in the cities in Yangtze River Delta between 2004 and 2006 were hardly interrelated or mutual influenced. The spatial correlation of innovation outputs increased significantly between 2007 and 2010, suggesting the tendency of spatial agglomeration of innovative activities. The decline between 2011 and 2013 might be the results of industrial restructuring and upgrading of some cities in Jiangsu Province, making them less correlated with the others in the region; and the upturn in 2014 and 2015 might benefit from the establishment of Shanghai Free Trade Zone, whose catchment area covers many cities in the region. Figure 4 shows five standard deviational ellipses of the geographical locations and the innovation outputs—based on annual patent application granted data in 2004, 2008, 2012 and 2016—in the map of the Pearl River Delta, respectively; Table 3 lists the statistical data for them. Looking at the angle of rotations of them, the four ellipses describing the spatial distribution of the innovation outputs are in the similar 'east–west' direction as the geographical distributions of the nine cities. All the ellipses of innovative outputs have larger angles (99.65°, 94.92°, 99.49° and 99.34°, respectively) than that of the geographic locations (92.34°), suggesting the innovative outputs from city/cities in the southeast of this region are more significant than its/their geographic locations in this region; while the city/cities in the northwest have weaker-than-average roles in the spatial pattern of innovative outputs in the region. Combining with the city-based analysis above, Shenzhen—the metropolitan city with high-tech industries agglomerated, in the southeast of the region—is the engine of innovative activities in this region, and the core area of innovative activities are to the southeast of the geographic core. The central points (focuses) of the innovation ellipses are to the east of that of the geographical ellipse, and they are consistently moving eastwards during the investigated periods, indicating the increasing significance of the southeast cities, that is, Shenzhen and those around Shenzhen, in the knowledge production network of this region. In addition, the central point of the innovation ellipses of 2004 is to the northeast of that of the geographical ellipse, and it turns northwards during 2012 and 2016, suggesting the significant and accelerated knowledge production in northern cities – Foshan and Guangzhou – in this region. The major axis of the innovation ellipses has been consistently shortened while the minor axis extended during the investigated period. The major axis reduced by 1.62 km during 2004–2008, 2.75 km during 2008–2012 and 0.83 km during 2012–2015, indicating the polarization phenomenon in the "east–west" direction of the region. The minor axis significantly increased by 0.32 km during 2004–08, 1.32 km during 2008–12 and 1.38 km during 2012–16, indicating dispersion phenomenon in the "south–north" direction of the region. That means: (i) a core area of innovation activities is come into being and has increasing gravitational force to the innovative activities in the region; and (ii) the emerging of the core innovative area means the existence of innovation spillover between major cities in the core area, but innovative spillover predominantly happens within the core area, with little influence on cities outside the core. Figure 5 shows the standard deviational ellipse of the geographical location of the 16 cities in the Yangtze River Delta and that of the innovation outputs in the four investigated years in this region. Table 4 lists the corresponding statistic data of them. The rotation angles of innovative ellipses vary between 142° and 156°, around that of the geographic ellipse, suggesting the spatial distribution of innovative activities in this region are generally similar to its geographic layout. A significant fluctuation occurred between 2004 and 2008, where the rotation angle increased from 142.69°to 156.49°, suggesting some cities in the north and south parts of the region—Suzhou, Ningbo and Shaoxing, according to the city-based analyses above—had significant improvement on the technology innovation activities; but latter other cities come up, hence the rotation angle moves backwards a little bit to 153.15° in 2012 and 151.46° in 2016. The central points (focuses) of all the innovative ellipses, especially that of the one in 2004, are in the east of that of the geographic ellipse, meaning the city in the east of the region, that is, Shanghai, have distinctive innovative performance. Nevertheless, the central point of the innovative ellipse moves 9.01 km southwestward between 2004 and 2008, 22.11 km northwestward between 2008 and 2012, and then 12.37 km northwestward between 2012 and 2016. The fluctuation tells two things: (i) a few core cities other than Shanghai had promoted their innovative performance significantly than shifted the spatial characteristics of the distribution pattern of innovation outputs, moving the central point westwards; and (ii) the growth of the innovative capability in some core cities (Wuxi, Nantong and Suzhou) are not consistent, especially in the period between 2008 and 2015, probably due to inconsistent investments. The major and minor axes of the innovative ellipses fluctuate in an accordant way throughout the investigated period, with gradually increased differences between them. The major axis reduced by 1.71 km during 2004–08 and 0.99 km during 2008–12, indicating the polarization phenomenon in "southeast-northwest" direction of the region, and it increased by 13.88 km during 2012–16 with the largest amplitude of fluctuation, indicating the dispersion phenomenon in "southeast-northwest" direction of the region. The minor axis reduced by 10.32 km during 2004–08, by 6.12 km during 2008–12, indicating the polarization phenomenon in "northeast-southeast", then it increased by 5.01 km during 2012–16, indicating the dispersion phenomenon in "northeast-southwest" direction of the region. Combining with the results of spatial autocorrelation analysis above, it is clear that the innovative activities in the Yangtze River Delta were already agglomerated in a few core cities in 2004, before there were clear signs of the emerging of intercity networks of knowledge production. The increased spatial relations of the innovative activities in this region, however, did witness a clear evolvement tendency of the spatial characteristics of innovative activities—that they are further agglomerated around the core area, and the core area is evolving with more emerging core cities along the north–south axis of this region, renewing the spatial pattern. Altogether, these results show that the innovative activities in the Yangtze River Delta are increasingly agglomerating to the core cities, and that the "core area" is extending from Shanghai to an area including other cities to its north, suggesting the impact of knowledge spillover. A shared political characteristic of the two investigated regions is the absence of regional assemblies that can strategically determine integrated regional development. That means there were no effective region-wide policy interventions that arranged the spatial distribution of knowledge production and innovative activities; the observed spatial patterns are the accumulated consequences of efforts made by individual cities in each of the regions, with both intercity competition and intercity co-operation exist. In that context, this paper finds that, although in both regions there are clear differences among "core" cities of innovation agglomeration, cities with the average degree of agglomeration and those below the average, the spatial characteristics of innovation outputs distributions vary to a limited degree from the characteristics of geographical relations between cities in each region. That means the knowledge externalities observed in the two regions do not, or not yet, dissimilate the spatial patterns of innovative activities from the geographic patterns. Comparing the results of standard deviational ellipse tests of innovative outputs against that of the location quotient analysis of high-tech industries agglomeration, it can be concluded that the core areas of innovative activities are come into being around the critical cities with the highest specialization degree of high-tech industries in the investigated period, which are the Shenzhen in the Pearl River Delta and Shanghai in the Yangtze River Delta. That suggests, at least in the early stages of the growth of the investigated high-tech industries, the industry agglomeration does bring the agglomeration of innovative activities, echoing Capozza et al.'s (2018) and Choi's (2020) arguments that existing knowledge about the emerging technology has a significant and positive impact on innovative activities. However, the agglomeration of high-tech industries alone cannot fully explain the spatial patterns of innovative activities. Most notably, Guangzhou gets lower-than-average location quotients (around 0.5) almost throughout the investigated period (exceptions are in 2007 and 2008), suggesting a low level of high-tech industries agglomeration, but it can be recognized through the northward turns of the focal points of innovation ellipses (presented in Figure 4) between 2012 and 2016 that innovative activities in the area around Guangzhou was boosted without evidence about a high-tech agglomeration observed. Drawing on Capozza et al.'s (2018) and Choi's (2020) conclusions, the distributions of universities and research centres might add to the explanations—Guangzhou does have a much larger number of top universities than other cities in the region; and cities with neither agglomerated universities nor agglomerated high-tech industries are not in the core areas in both regions. It is therefore recommended that the spatial distributions of high-tech industries and that of universities in both regions could be better matched, allowing universities and high-tech industries, or the R&D departments of them, to benefit from each other regarding the facilities, funding and talents needed for innovative activities, to further enhance the innovative performance and efficiency in the region. The results of spatial autocorrelation tests above show that, in both regions, the intercity connectivity of innovative activities increases significantly throughout the investigated period. At the beginning of the period investigated, the innovative activities in Yangtze River Delta show almost no (positive or negative) spatial correlations at all, though the high-tech industries agglomeration in some critical cities was evident. Therefore, the existence of a knowledge production network is not a determinant precondition of high-tech industries agglomeration, but the latter does contribute to foster the former in the long-term. Meanwhile, knowledge spillover effects are observed, but the spillover within the core areas discussed above are more notable than in marginal areas. In both regions, the analysis above shows that innovative activities are increasingly concentrated in an area around the core cities identified, which delineates the core areas, suggesting the limited catchment area of knowledge spillover from the core cities. Furthermore, as the region grows, there is a tendency that the core area is shrinking and with clearer boundary, suggesting the increasing differentiation between cities in and outside the core area regarding their innovative performances. Nevertheless, the breakthrough of some cities—especially those in the second-tier of knowledge production in the region, as the result of intercity competition in building high-tech hubs and attracting innovative talents—alters the spatial patterns as well as the core area of innovative activities in the region. The knowledge production network is therefore restructured and the catchments of knowledge spillover in the regions are influenced. On the one hand, there is the need to strengthen spatial connections between innovative activities in different cities in the region, to enhance the stability of the knowledge production network, so the innovative activities around the region are sustainably supported and the development of the region could be better planned. On the other hand, individual cities' breakthrough on innovation promotion is the critical force extending the catchment areas of knowledge spillover in the region. This paper investigates and compares the spatial evolution process of innovation agglomeration in the Pearl River Delta and the Yangtze River Delta, analysing the degree of high-tech industries agglomeration, the spatial autocorrelations among cities within the same region, and the evolving spatial characteristics of innovative activities in the two regions the between 2004 and 2016. Three critical conclusions are made. First, the high-tech industries agglomeration fosters the local knowledge production network and hence promotes local knowledge spillover; but an existing local knowledge production network is not the precondition of high-tech industries agglomeration. No evidence in the period investigated suggests that industry agglomeration may harm innovative performance. Second, local knowledge spillover occurred in a "core" area and, without individual cities' breakthrough on innovation promotion, there is a tendency that the core area is shrinking and the differentiation between cities in and outside the core are more distinctive. Third, the evolving of spatial patterns of innovative activities follows that of high-tech industries agglomeration in general, but some mismatches between them suggest the spatial distribution of high-tech industries cannot take full advantage of the innovative resources and facilities in the two regions, which typically include, for example, the universities and research centres. Accordingly, three recommendations are generated for the two regions regarding their innovation-driven development strategies. First, it is recommended that local innovative policies to encourage high-tech industries to have better linkages with university and/or research centres, as well as other local facilities and resources relating to innovative activities. Second, the spatial connections of innovative activities between cities within the region should be strengthened, to secure the stable support of local knowledge production network. Third, the political power and innovative policy-making power of the local authorities of individual cities, as well as the intercity competition mechanism, should be sustained while a collaborative regional network comes into being, to enable the consistent renew of innovative policies and the long-lasting active communications between innovative activities among cities across the region. This research notes two directions for further studies. First, this research has the limitation that both regions investigated have relatively short histories of high-tech industry development, for approximately twenty years. The findings speak for the spatial evolution of innovative agglomeration in the early stages of high-tech industries agglomeration only. Follow-up research with a longer-term investigation is expected. Second, focusing on the evolving spatial pattern of agglomeration, this paper reveals the co-evolvement of innovation agglomeration and knowledge spillover within the regions, as well as the spatial characteristics of them, as a part of the spatial mechanism of the knowledge spillover. Further investigations on the causal relations between them and the underlying institutional or economic mechanisms are still in need. This research receives funds from the Ministry of Education of China (Major Project of Social Science Base of the Ministry of Education of China, Grant/Award Numbers: 16JJD790040, 12JJD790031) and China Scholarship Council (Project Supported by China Scholarship Council, Grant/Award Number: 201708440491).