blogbanner

Moderator:  Calla Wiemer (calla.wiemer@acaes.us)

The Geography of Innovation in China

Co-authors: Fushu Luan; Ming He; and Donghyun Park 

This post draws on a paper presented at the Allied Social Science Association Annual Meeting in a session sponsored by the American Committee on Asian Economic Studies on "Economics of Innovation in Asia", 3 January 2021, video here.   

Economic progress is propelled by new innovation building on the legacy of the past. Thus the more widely innovation is diffused across geographical space, the greater the potential for new innovation to proliferate.

We examine the geographic pattern of innovation diffusion in China using data on patent citations for the period 1985 to 2015. More than 3.6 million patents were published during this period containing nearly 5.1 million citations. Our analysis of the geographic reach of citation networks employs descriptive techniques at the province level and regression techniques at both province and city levels.

We expect innovation to be concentrated geographically in connection with R&D capacity as undergirded by academic, government, corporate, and financial resources. Beijing is China's preeminent seat of academic and government research. The country's major corporations are concentrated in Guangdong province and the Yantze delta region. Shanghai is the leading financial center with Guandong home to a secondary center and Hong Kong in close proximity.

 Table 1:  Citation Rank by Province, 1985-2015
  Citation Source Citation Recipient
    Total Inter-Province %   Total Inter-Province %
1 Jiangsu 655,327    70.7 Beijing 714,068    66.2
2 Beijing 643,029    62.4 Guangdong 650,552    61.6
3 Guangdong 611,341    59.1 Jiangsu 559,138    65.6
4 Zhejiang 323,533    75.4 Shanghai 398,204    77.6
5 Shanghai 308,372    71.0 Zhejiang 306,031    74.0
6 Shandong 291,486    77.1 Shandong 270,196    75.3
7 Anhui 257,100    79.2 Liaoning 165,951    84.0
8 Sichuan 167,359    81.7 Sichuan 150,027    79.6
           
30 Qinghai 4,734    83.1 Qinghai 4,510    82.2
31 Tibet 1,223    85.5 Tibet 1,202    85.3
 

Figure 1:  Patent Citation Matrix by Province, 1985-2015

The influence of this economic geography is evident in the patent citation rankings shown in Table 1. The top ranked source of citations is Jiangsu, followed by Beijing, Guangdong, Zhejiang, and Shanghai. The same five provinces are the top ranked recipients of citations, with some shuffling of the order to put Beijing first. The remote and less economically developed provinces of Qinghai and Tibet lie at the bottom of the rankings. The citation numbers drop off sharply moving down the list, the 8th ranked province of Sichuan having only a quarter or so of the citations given and received as the top ranked provinces.

The share of citations that extends across provincial boundaries lies generally in the range of 60-85%. This share tends to be lower for the most active provinces.

A detailed depiction of inter-provincial citations is presented in the heat matrix of Figure 1. Citing provinces are given in the rows with shares by recipient provinces in the columns such that totals along each row sum to one. The prominence of the diagonal indicates a high concentration of citations received from within provinces. Echoing Table 1, Beijing, Guangdong, and Jiangsu received the highest shares of citations. Beijing and Jiangsu received consistently high shares originating from across all provinces. Guangdong received discernibly higher shares from eastern and southern provinces indicating that its influence on innovation is more localized. To the extent that the rarely cited provinces of Qinghai and Tibet are cited at all, they are cited mostly by each other as neighbors. Geography thus matters for innovation activity.

We examine the importance of geographic networks by adapting the methodology of Acemoglu et al. (2016), originally applied to linkages across technology classes. The premise for Acemoglu et al. was that innovations in upstream technology classes would seed innovations in downstream classes (for example, innovations in "electrical measuring & testing technology" leading to innovations in "nuclear & X-ray technology"). Similarly, we reason that innovations in research hubs will provide the impetus for innovations stretching out through integrated geographic networks.

For our purposes, the methodology involves applying past patterns of citations among cities, as captured in citation matrices like that of Figure 1, to upstream patent numbers to generate predictions of downstream patents throughout the geographic network, then using regression analysis to test for the significance of the predicted values in explaining actual values. The predictions are based on a 10-year lag structure where, for example, a vector of predicted patents by city in 2012 is generated based on summing the products of vectors of actual patents for each year from 2001 to 2011 times their respective matrices of cumulative citation shares (e.g., the term for the lag at 2009 is given by the vector of citations in 2009 times the matrix of citation shares accumulated through years 2009-2011).

The sample consists of 339 cities for years 2005-2015 incorporating lags for prediction purposes back to 1995. For the baseline regression, patent prediction is based on the complete sample of cities including each city as a predictor of its own downstream patents. 

Table 2:
Downstream Patent Prediction 

log predicted patents  0.856***
(0.004)
 0.537***
(0.020)
constant  0.263***
(0.026)
 1.411***
(0.081)
fixed effects no yes
R-squared  0.841  0.885
observations  3,729  3,729
The dependent variable is the log of actual downstream patents. Fixed effects are by both city and year. *** indicates significance at the 1% level. Robust standard errors are in parentheses. 

Baseline regression resutls reported in Table 2 suggest that past patents traced through a geographic network serve well to predict downstream patenting. A one percent increase in predicted patents is associated with a 0.856 percent increase in actual patents in a model with no fixed effects and a 0.537 percent increase in a model that includes fixed effects by year and city. In both models, the estimates are statistically significant at the one percent level. When included, fixed effects absorb a substantial portion of the explanatory power suggesting consistent patterns of citation by city and year are influential in outcomes. Nevertheless, when these patterns are controlled for, deviations are explained to a significant degree by geographic networking effects.  

In our paper for the ASSA meetings, we expand the regression analysis to incorporate more refined measures of geographic networking. We also elaborate geographic patterns of citation using network mapping techniques.

Our research shows innovation in China to be highly concentrated geographically and given to strong inter-city network effects. The work by Acemoglu et al., on which our analysis is modeled, similarly finds strong network effects, but these are based on linkages rooted in technology. Linkages based on geography would in principle seem more fluid and expansive since they depend essentially on flows of information across space, although technological specializations may well be geographically concentrated such that an Acemoglu effect applies. To the extent that impediments exist to the geographic propogation of knowledge, easing the flows could be a boon to progress. New innovation builds on past innovation, hence the more broadly advances are shared at every stage, the better for the accumulation of knowledge.

____________________

Co-Author Fushu Luan is Lecturer in Nanjing Audit University.

Co-Author Ming He is Lecturer at the International Business School, Xi'an Jiaotong-Liverpool University, Suzhou, China.

Co-Author Donghyun Park is Principal Economist at the Asian Development Bank, Metro Manila, Philippines.

_________________________

Related posts from the ACAES session on the Economics of Innovation in Asia at the 2021 Allied Social Science Association Meeting:

Also relevant is the Review of Keun Lee, The Art of Economic Catch-Up

 

PAST Webinar on Online Platforms to Catalyze Inclu...
Fabrication vs Knowledge Activities in Global Valu...

Related Posts

 

By accepting you will be accessing a service provided by a third-party external to http://acaes.us/