TECHNOLOGICAL INNOVATION OF CHINESE FIRMS: INDIGENOUS R&D, FOREIGN DIRECT INVESTMENT, AND MARKETS

Transcription

1 TECHNOLOGICAL INNOVATION OF CHINESE FIRMS: INDIGENOUS R&D, FOREIGN DIRECT INVESTMENT, AND MARKETS A Dissertation Presented to The Academic Faculty By Jingjing Zhang In Partial Fulfillment Of the Requirements for the Degree Doctor of Philosophy in the School of Public Policy Georgia Institute of Technology August 2006 Copyright 2006 by Jingjing Zhang

2 TECHNOLOGICAL INNOVATION OF CHINESE FIRMS: INDIGENOUS R&D, FOREIGN DIRECT INVESTMENT, AND MARKETS Approved by: Dr. Juan Rogers, Advisor School of Public Policy Georgia Institute of Technology Dr. Marco Castillo School of Public Policy Georgia Institute of Technology Dr. Diana Hicks School of Public Policy Georgia Institute of Technology Dr. Haizheng Li School of Economics Georgia Institute of Technology Dr. Philip Shapira School of Public Policy Georgia Institute of Technology Date Approved: April 25, 2006

3 [For my Mother and Father]

4 ACKNOWLEDGEMENTS I am particularly indebted to my advisor Professor Juan Rogers, for his mentorship, guidance, and encouragement during the two years of research and writing of this dissertation. I would also express my appreciation to the members of my dissertation committee: Professor Marco Castillo, Professor Haizheng Li, Professor Diana Hicks, and Professor Philip Shapira for their constructive comments on the earlier versions of the dissertation. I would like to thank Professor Zhengfeng Li and Professor Guoping Zeng of Tsinghua University in Beijing for their financial support and assistance in collecting Chinese patent data. I am also grateful to many faculty members, students, and staffs in School of Public Policy at Georgia Tech. I would express my warmest appreciation to my parents who provided invaluable support throughout the whole process of planning and writing this dissertation.

5 TABLE OF CONTENTS ACKNOWLEDGEMENTS... IV LIST OF TABLES... VIII LIST OF FIGURES... X LIST OF SYMBOLS AND ABBREVIATIONS... XI SUMMARY...XII CHAPTER 1: INTRODUCTION AND OVERVIEW Two Models of Chinese Industrial Development Chinese Domestic Patent Data Policy Implications... 7 CHAPTER 2: LITERATURE REVIEW Use Patent Data to Measure Technological Innovation Four Indicators that Measure Innovation Patent Data: Limitations, Remedies, and Previous Studies Determinants of a Firm s Technological Innovation Capability Learning from the Outside: The Bottom-up Model Development with the Country: The Top-Down Model Technological and Regional Differences CHAPTER 3: RESEARCH MODEL Research Model Research Questions and Hypotheses CHAPTER 4: CONSTRUCTION OF A DATASET FROM SIPO PATENT ABSTRACT DATABASE AND THE CHINESE STATISTICAL YEARBOOKS v

6 4.1 The Overall Provincial Level Panel Data Set Constructing Patent counts from Chinese Patent Abstracts CHAPTER 5: CHINESE PATENTING ACTIVITIES: TRENDS AND DISTRIBUTIONS Chinese Domestic and Foreign Patents Chinese Firm, University, Research Institution, and Individual Patents Chinese Domestic Patents and Indigenous R&D, FDI, and Markets Technology Differences in Patenting Activity Regional Distribution of Patenting Activity CHAPTER 6: THE DETERMINANTS OF TECHNOLOGICAL INNOVATION OF CHINESE DOMESTIC FIRMS Search for the Appropriate Regression Model Hypotheses on the Regression Results Pooled Regression Model Could FDI Be Endogenous to Patenting? Do Patents Hamper Firms Efforts and Other Factors? Possible Effects of the Interaction Between FDI and R&D Expenditures Differences in Firm Patenting Across Regions Differences in Firm Patenting over the Years Innovation Performance by Technology Sector Discussion the Internal Validity of the Model CHAPTER 7: THE COMPLEX PICTURE OF CHINA S TECHNOLOGICAL INNOVATION AND POLICY IMPLICATIONS vi

7 7.1 Summary of the Main Findings Patenting Trends Determinants of Technological Innovation of Chinese Firms The Complex Picture of China s Industrial Technology Innovation and Policy Implications Limitations and Further Studies APPENDIX A: STATISTICS OF THE FIGURES APPENDIX B: STATA REGRESSION OUTPUTS REFERENCES VITA vii

8 LIST OF TABLES Page Table 1. A comparison of Four Indicators that Measure Technological Innovation Table 2. Concepts, Symbols, and Indicators Table 3. Summary of the Legal Framework of China s Patent System Table 4. Chinese Patent Abstract: an Example Table 5. Three Types of Patent counts of Chinese Invention Patents ( ) Table 6. Top 20 Countries that Own Chinese Patents ( ) Table 7. Top 20 Patent Assignees ( ) Table 8. Top 10 Patent Holders of U.S. Patents to China ( ) Table 9. Top 10 Chinese Domestic Patent Holders ( ) Table 10. Top 10 Patenting Chinese Domestic Firms ( ) Table 11. Top 5 Firm Patent Holders in the Electricity and Electronics Sector (Before and After 1997) Table 12. Top 5 Firm Patent Holders in the Chemical and Pharmaceutical Sector Table 13. Top 5 Firm Patent Holders in the Processing Engineering Sector Table 14. Number of Domestic Firm Patents for 30 Provinces by Application Year ( ) Table 15. R&D and Economic Indicators of 30 Chinese Provinces: Average Value between 1989 and Table 16. Descriptive Statistics Table 17. Regression Results of OLS, Poisson, and Negative Binomial Models a,b viii

9 Table 18. Regression Results of OLS, Poisson, and Negative Binomial Models without GDP and Population Table 19. Regression Results of OLS, Poisson, and Negative Binomial Models with Time Dummies a Table 20. Two-Stage Regression Table 21. Regression with Lagged Variables a Table 22. Stepwise Regressions Table 23. Interaction Effects a Table 24. Regional Dynamics Table 25. Time Dynamics Table 26. Regression of Firm Patents in Six Technological Fields Table 27. Check for Hidden Bias ix

10 LIST OF FIGURES Page Figure 1. Research Model Figure 2. Count Patents by Geographical Locations, Types of Institution, and Technology Sectors Figure 3. Chinese Invention Patents to Domestic and Foreign Assignees ( ).. 58 Figure 4. Chinese Domestic Invention Patent by Assignee Type ( ) Figure 5. Chinese Domestic Firm Patent and R&D Performed by Firms and Universities & Research Institutes ( ) Figure 6. Chinese Domestic Firm Patent and Foreign Direct Investment ( ) Figure 7. Chinese Domestic Firm Patent and Domestic Consumption and Exports Figure 8. Chinese Domestic Firm Patent by Sectors ( ) Figure 9. Histogram of the Firm Patent counts Figure 10. Number of Domestic Firm Patents ( ) x

11 LIST OF SYMBOLS AND ABBREVIATIONS CAS EPO FDI GDP IT JPO IPC LIBO MNCs NIEs OBD ODM OECD OEM OLS PC R&D SCI SIPO SOE TFP USPTO WIPO Chinese Academy of Sciences European Patent Office foreign direct investment gross domestic product information technology Japan Patent Office international patent classification literature-based innovation output multinational corporations newly industrializing economies original brand manufacturing original design manufacturing Organization for Economic Co-operation and Development original equipment manufacturing ordinary least squares personal computer research and development science citation index State Intellectual Property Office of Peoples Republic of China state-owned enterprise total factor productivity United States Patent and Trademark Office World International Property Organization xi

12 SUMMARY What are the factors behind the recent development of industrial technology in China? Does China follow the path of learning technology from outside through direct foreign investment and international trade as other Asian newly industrialized economies, or imitate the U.S. model that develop science and technology within the country based on the strong domestic research capacity? This study examines these questions using a comprehensive research model and a new Chinese patent dataset. The research model applied in this study integrates the international factors, including inward foreign direct investment and export, and the domestic ingredients, such as domestic firms R&D, local academic research, and domestic market demands. The patent statistics in this study are created based on more than 120 thousand granted invention patent abstracts in China between 1985 and Compared with the Chinese patent data used in prior studies, this dataset allows the identification of the technological sector in which the patents were classified and distinguishes firm patents from universities and research institutes that also were awarded patents. The dependent variable for regression analysis is the technological innovation performance of Chinese domestic firms as measured by the number of patents awarded to firms in 30 Chinese provinces from 1989 to The final panel data for regression analysis were completed with other provincial indicators for the same years on research and development (R&D) expenditures by firms and public institutions, foreign direct investment (FDI), domestic consumption, and foreign exports. xii

13 The analysis of the patenting trends of Chinese firms reveals that strengthened large state-owned enterprise (SOEs), growing new domestic technology enterprises, and burgeoning international joint-ventures are the sources of the continuously increasing firm patenting activity in China since The results of count data fixed effect regression approaches show that the efforts of firms, measured by industrial R&D expenditures, spillovers from R&D activities conducted at universities and public institutions in the same region, and demand driven mainly by foreign exports are the most prominent positive factors in the domestic firms technological innovation performance, measured by the number of patents they were awarded. Of these factors, FDI plays a complicated role. When its effect is studied in the regression model, the net impact of FDI on domestic firms patenting activity is mostly insignificant and sometimes negative. However, a closer look of the top domestic patenting firms in the electricity and electronics sector reveals significant knowledge diffusion from foreign multinational corporations (MNCs ) Chinese subsidiaries to Chinese domestic firms. This study also points out that China s national innovation system has improved over years as indicated by substantial academic research knowledge spillovers and considerable demand pull effects, and the inequality distribution of firm patenting activity in China has been strongly determined by unbalanced regional innovation system. xiii

14 CHAPTER 1: INTRODUCTION AND OVERVIEW 1.1 Two Models of Chinese Industrial Development Since the inception of the reform and open door policy of China in 1978, the country has undergone tremendous economic growth. From 1978 to 2003, the average annual growth rate of the gross domestic product (GDP) was as high as 9%, and the share of agriculture in the GDP dropped sharply from 31.2% in 1979 to 14.6% in This unprecedented growth cannot be solely explained by investments in physical and human capital. Many studies (Li 1997; Otsuka, Liu et al. 1998; Tong 2001; Yan and Yudong 2003) have exhibited robust results that indicate that along with the GDP growth, the total factor productivity (TFP) increased considerably, which contributed significantly to overall economic growth in China. During the same period, China became a manufacturing powerhouse of not only labor-intensive products but also technology-intensive industries that did not exist in China before the reform era. In 2001, technology products accounted for 19.4% of China s export, quadrupling since the early 1990 s. In 2002, China overtook Japan and Taiwan to become the world's second largest information technology (IT) hardware producer. In 2005, China s largest personal computer (PC) maker, Lenovo, acquired IBM s PC business for $1.75 billion. These are remarkable achievements for a country that had almost no IT sector 20 years ago (Research Group on Development and Strategy of Science and Technology of China 2003). 1

15 Such empirical evidence indicates that technological innovation has played an increasingly important role in driving recent economic growth in China. However, researchers have not reached a consensus on what the dominant factors are behind this fast development of industrial technology in China. The literature on the development of industrial technology in China offers two main explanations for its growth in the last several decades. One involves the application of the bottom-up model, suggested by the studies of successful stories of Asian newly industrializing economies (NIEs) (Kim 1997; Hobday 1995). This model characterizes technological development in developing countries as a backward learning process, starting from OEM (original equipment manufacturing), through ODM (original design manufacturing), to OBM (original brand manufacturing). Movement from OEM to OBM, according to Kim and Hobday, is determined by aggressive learning of the local firms, international knowledge spillovers from multinational corporations (MNCs), and access to foreign markets. Since the establishment of the open door policy, a large amount of foreign direct investment (FDI) has entered China. From 1978 to 2004, China attracting a total of $550 billion in FDI, which played a vital role in recent export growth in China, channeled most of the money into labor-intensive manufacturing industries (Fu 2000). The World bank appraised this as one of the most important achievements of China s reform era (World Bank 1997). Recently, the presence of MNCs in China has triggered a shift from simply Made in China to R&D in China. They have started to apply for more Chinese patents and built research and development (R&D) centers in China (Walsh 2003). Due to increasing inward FDI, together with growing international trade, escalating foreign 2

16 patents, and burgeoning MNCs R&D centers, to some degree, Chinese domestic firms have been presented with great opportunities to acquire advanced technologies and master management skills embodied in foreign goods, intellectual property, as well as physical and human capital. Various empirical studies have shown that Chinese domestic firms benefited from international technology transfer. Using case study approaches, Xie (2004) found that successful technological upgrades in China s color television industry were the result of effectively absorbing technologies contained in the imported assembly lines and really understanding domestic market demands. Mu and Lee (2005) suggested that labor turnover from Bell s Chinese subsidiaries in Shanghai, imitation, and reverse engineering were behind the impressive technological development of Huawei Technologies Co Ltd., the leading telecommunication equipment producers in China. Among Chinese scholars, the explanation for the success of such development strategies was the application of the top-down model. Based on detailed case studies of four successful information technology companies in Beijing, Lu (2000) suggested that Chinese firms that followed this model were able to skip the stages of OEM and ODM, and leapfrog to the establishment of their own brand name early in the domestic market, which occurred because of the large and fast growing domestic demand with specific local user needs and strong scientific research capabilities. China s scientific capability has increased sharply in the past 10 years. On the input side, China s annual R&D investment rose from US$6.2 billion (0.64% of the GDP) in 1997 to $156.5 billion (1.23% of the GDP) in By 2003, China had surpassed Germany, becoming the third country worldwide in terms of R&D investment, sharing 9% of total world R&D, behind only the United States and Japan (OECD 2004). 3

17 On the output side, Chinese scientists and engineers, in 2002, published 77,395 papers that were included in the Science Citation Index (SCI), ranking 6 th worldwide (China Statistical Yearbook 2003). This highly capable scientific research system, along with the large pool of well-trained science and technology labor force and a huge domestic market indicates that unlike firms in most other developing countries, those in China are in closer geographic proximity to advanced scientific knowledge, high quality human capital, and potential markets. To improve development within the country, China still needs to overcome the organizational barriers erected during the planned economy period. Under the planned economy, domestic firms, following orders from the central or local government, were merely production units without financing, marketing or R&D functions. R&D, on the other side, was conducted in various public research institutes and universities. As a result, domestic firms lacked the capabilities to innovate while public research institutes and universities lacked the incentives to commercialize their inventions (Gu 1998; Suttmeier 1997). Various government policies were enacted to promote, even force the technology transfer from local universities and research institutes to domestic firms, including merging R&D institutions with existing enterprises, spinning-off new technology enterprises from R&D institutions and universities and establishing national level High and Emerging Technology Industry Development Zones (Gu 1998). These resulted in clusters of burgeoning high and new technology industries in the coastal, frontier, border and inland cities throughout the country (Segal 2003). Therefore, in contrast to the bottom-up model, the top-down model stresses in-house innovation by domestic firms, knowledge spillovers from local universities and public research institutes, and the growth induced by domestic market. 4

18 To shed light on the debate between the bottom-up and top-down models, the first contribution of this study is to present a comprehensive research model that includes key variables from both models technology push from both indigenous R&D and FDI, and demand-pull from both domestic and foreign market identifies and compare the spillover effects from both local universities/research institutes and MNCs, and examines the interaction effects between indigenous R&D and foreign direct investments. A detailed discussion of this research model will be addressed in chapter Chinese Domestic Patent Data The second contribution of this research is to provide a measure of China s industrial technological innovation with a new set of patent indicators that have never been used for quantitative academic research before. A majority of previous quantitative studies that have investigated determinants behind China s industrial innovation used total factor productivity (TFP) as the primary measurement of the innovative capability of domestic firms (Li 1997; Otsuka, Liu et al. 1998; Tong 2001). For example, Liu and Wang (2003) found that inward FDI, in-house R&D investment, and company size are the three main factors contributing to the recent TFP increases in China. Compared to TFP, which is a broad and indirect measurement of technological innovation at the firm level, patent data are relatively specific and direct indicators that record technological changes at the invention level. Thus, they are able to provide a close assessment of technology trajectories and industrial differences that cannot be accurately captured by TFP. Because of these advantages, patent data have begun to be used more frequently in current innovation studies of China. 5

19 Recently, two types of patent data have been widely used in innovation studies about China: (1) U.S. patents granted to Chinese inventors or Chinese assignees by the U.S. Patent and Trademark Office (USPTO) (Bhattacharya 2004, Mahmood and Singh 2003), and Chinese patents issued to Chinese assignees by the State Intellectual Property Office of the People s Republic of China (SIPO) (Sun 2000, 2003a; Cheung and Lin 2004). To distinguish between the two, this paper will refer to the former as Chinese U.S. patent and the latter as Chinese domestic patent. Because China, unlike Korean or Taiwan, is a continental economy with a huge domestic market and large pool of human resources, the use of Chinese U.S. patents to measure innovative capability of Chinese domestic firms leads to biased results by overestimating the impacts of foreign markets without much consideration for the important domestic factors. Local demand, indigenous R&D, and competition within the country are the main drivers of technological innovation of Chinese domestic firms, so many of them apply for patents only inside of China because domestic markets are their major market. As a result, a large proportion of their innovation activities are not visible through China s international patents. In the cases of other developing economies with large domestic markets, a similar measurement bias also results. For example, Albuquerque (2000) found that Brazilian U.S. patents systematically differ from Brazilian domestic patents. The second type of patent indicator is the Chinese domestic patent data used by Sun (2000, 2003) and Cheung and Lin (2004). These data are provincial-level patent counts 1 available in the China Statistical Yearbooks. Because they are domestic patents, they 1 Patent counts are the number of patents assigned over a certain period of time to firms, universities, industries, or countries. 6

20 represent a more accurate measurement of local innovative capabilities than Chinese U.S. patents. However, patent data in China Statistical Yearbooks are highly aggregate, providing only the total number of patent applications and patent grants at the national and provincial levels. Because of these data constraints, previous studies have not been capable of regrouping the patents by institution (i.e., firm or university) to differentiate between industrial invention and academic research, categorize the patents by various technology fields to observe different innovation behaviors in different sectors, or arrange issued patents by application years to better estimate when the patented inventions took place. To solve these problems, this study uses alternative Chinese domestic patent data based on the recent computerization of the Chinese patent abstract dataset (SIPO 2004). The Chinese patent abstract database (SIPO 2004) includes all public patent applications to the SIPO between 1985 and This study extracted a subset of 125,783 patent abstracts of all invention patents granted by the SIPO between 1985 and 2003 and then counted these patents by province, by industry sector, by firm ownership (domestic or foreign), and by type of institution (i.e., firm or university) to obtain a series of more accurate, less biased patent counts that provided a more detailed measurement of China s technological innovation than the patent indicators used in previous studies. 1.3 Policy Implications This study used a count data fixed effect regression model to examine the relationship between Chinese domestic firms technological innovation performance, measured by provincial patent counts, and a vector of independent variables: in-house R&D investment, R&D performed by local universities and public research institutes, 7

21 foreign direct investment (FDI), and domestic and foreign market demands. These independent variables are closely tied to the China s recent policies aiming to enhance the countries innovative capability. Such government policies included the following: (1) Increasing R&D investment, which jumped from $2 billion in 1991 to $12.7 billion in 2001 with an annual growth rate of 12.2%. (2) Reforming state-owned enterprises, which granted management and financial autonomy to enterprises in order to help them adapt to the market system and increase enterprises incentives to conduct innovation. (3) Encouraging spin-offs from universities and public research institutes and partnerships between universities, research institutes and domestic firms to commercialize strong R&D capability stored in universities and public research institutes. (4) Attracting inward foreign direct investment (FDI), hoping to capture the advanced technologies and management know-how embodied in the FDI. The results of this study provide additional empirical evidence from which current policies can be evaluated, and it presents useful insights for future improvement. The remainder of the paper is organized as follows: Chapter 2, a literature review, surveys prior studies on patent data and two industrial development models: the top-down model and the bottom-up model. Chapter 3 presents the research model of this study and lists five research questions and hypotheses. Chapter 4 describes the process of extracting patent data from the China patent abstract database and the construction of the provincial panel dataset. Chapter 5 examines the patenting trends of Chinese firms, universities, research institutes, individuals, and foreign MNCs. Chapter 6 summarizes the main findings of the regression analysis that explores the relationship between domestic firm 8

22 patenting activity and firms in-house R&D, academic research, FDI, and domestic and foreign market demands. Chapter 7 concludes the dissertation with a summary of the main findings, discussions on policy implications and limitations of the study, and suggestions for future research. 9

23 CHAPTER 2: LITERATURE REVIEW 2.1 Use Patent Data to Measure Technological Innovation Four Indicators that Measure Innovation The idea of measuring technological innovation stems from the discovery of a large residual of aggregate productivity growth that could not be explained by tangible and human capital accumulation (Abramovitz 1956; Solow 1957; Griliches 1996). Since then, many studies have attempted to quantify technological change using different indicators. One of these indicators is total factor productivity (TFP). Introduced by Solow (1957), TFP, which represents all the output variations that cannot be attributed to changes in human and capital input 2 (Nadiri 1970), is a comprehensive but, at the same time, indirect and vague measurement of technological innovation. Other financial and political factors that have little to do with technological change, such as economic depressions or wars, can also affect TFP. Besides TFP, Acs et al. (2002) categorized the direct measurement of technological innovation into the following three types of measures: (1) the measure of the input into the innovation process (i.e., R&D expenditures, R&D personnel), (2) the direct measure of innovative output (i.e., literature-based innovation output (LBIO)), and (3) the measure of inventions, the intermediate output of innovation (i.e., patents). Each 2 In a Cobb-Douglas production function: Y = A K L input (K), labor input (L), and total factor productivity (A). α 1 α, total output (Y) is a function of capital 10

24 of these three measures provides both advantages and disadvantages in direct measurement of technological innovation. The first measurement, R&D data, has been widely used to estimate technological innovation at the firm, industrial, regional, and national levels. As an input factor to the innovation process, R&D data provide a descriptive index that can effectively assess technological innovation in both private and public sectors. However R&D data do not represent a good dependent variable if the purpose of the study is to identify determinants behind the innovation. Another concern with using R&D data is that the definition of R&D is not rigorous enough to assure a consistent estimation of R&D expenditures across different industries, regions, and companies. Finally, R&D data are susceptible to manipulation because R&D indicators are usually provided by surveyed firms, not by disinterested third-party sources. The second measure, LBIO, generated by sampling the new product announcement sections of technical and trade journals (Coombs et al. 1996), overcomes the bias across the commercialization of diverse inventions by including only commercialized inventions. However, the lack of standard to decide which journals should be included in the data collection process affects the consistency of the LBIO data. In addition, collecting LBIO data is a very time-consuming task. According to Griliches (1990),the third measure, the patent, is a document, issued by an authorized governmental agency, granting the right to exclude anyone else from the production or use of a specific new device, apparatus, or process of a stated number of years. The stated purpose of the patent system is to encourage invention and technical progress both by providing a temporary monopoly for the inventor and by 11

25 forcing the early disclosure of the information necessary for the production of this item or the operation of the new process ( ). Patents, as computerized and publicly available data, are easy to access. Patent data are published by patent offices representing the objective decision by patent examiners, so they have little chance of being manipulated (Jeff 1998). Moreover, patent data are presented at the invention level, each patent representing an invention, which gives researchers more freedom to form customized datasets that fit their specific research questions than they would have by employing TFP and R&D data. For example, researchers can regroup patents by technological classification code assigned to each patent to closely examine the technology trajectories across different industries, in geographical regions, and over time. Compared to TFP, patent data are a more direct measurement of technological change, revealing technological, individual, and institutional elements behind inventions. Patent records, as highly structured documents contain detailed information on the innovation itself, the technological area to which it belongs, the inventors and the organization (if any) to which the inventors assign the patent property right (Jaffe and Trajtenberg 2002, 3).Another advantage of patent data over TFP is that they can be used to measure inventions that originated from the public sector, because universities, government labs, and other non-profit organizations also file patents. In addition, as patent data are flexible, they can be aggregated into and regrouped (1) by type of inventors or assignees, by single firms, or groups of firms, (2) by technology fields, (3) by countries or regions, and (4) by time periods (OECD 1994). Therefore, patent data can be manipulated to analyze technological changes at both macro (i.e., national and regional) and micro (i.e., firm and individual) levels. 12

26 A comparison of patent data with LBIO shows that whereas patent statistics, after they have been computerized, are easy to access, collecting LBIO is a more complex and time-consuming process, despite the fact that LBIO is a less biased measurement of innovation outputs that include only commercialized new products. A summary of the main characteristics of all three measurements and TFP is presented in Table 1. Table 1. A comparison of Four Indicators that Measure Technological Innovation Indicator TFP R&D literaturebased innovation output Patents Input /Output Final output as economic growth Input as investment in R&D expenditure and human resource Intermediate outputs as new products Intermediate output as new inventions Source: Created by Author The smallest unit of analysis Difficulty in collecting data Measure Technological innovation in public sector Company Level Somewhat No Company & institution level Somewhat Yes Product level Yes No Invention level No Yes Patent data, particularly Chinese domestic patent data, have not yet been widely used for academic research. Some representative earlier studies using Chinese domestic patent data are Sun (2000), who showed the spatial distribution of patenting activity in China using number of Chinese patent applications at the provincial level from ; Cheung and Lin (2004), who measured regional innovation using patent applications in China from 1995 to 2000; and Guan and Liu (2005), who assessed China s national innovation capacities with number of invention patents applications 13

27 from each province from The patent data used in these studies are provincial patent counts reported in the China Statistical Yearbooks. These highly aggregate patent counts do not differentiate patents by the various institutions and individuals that applied for them. Each province has only one overall patent counts (including patents applied by firms, universities, public research institutes, individuals). Furthermore, they do not provide information about the technological compositions of the patent applications or patent grants at the province level. For example, it is impossible to know how many of the 1,074 invention patents assigned to Beijing in 1999 belonged to biotechnology or mechanical engineering. Because of these inherent data constraints, previous studies were not be able to precisely measuring Chinese domestic firms patenting activities, examining knowledge spillovers from academic research to industrial innovation, and revealing the unique industrial development pattern for different technologies. The newly available China patent abstract database (SIPO 2004) provides a possible remedy for this situation The China patent abstract database includes abstracts of all published patent applications since 1985, the inception of China s patent system. Each patent abstract contains detailed information about the patent invention: the year of application, the year of publication, the grant year of the patent, the international patent classification (IPC), the inventors name(s), the name(s) and full address(es) of the assignees, and an abstract briefly describing the invention. This informative dataset allows researchers to generate patent counts not only by regions but also by institutions and industries. After comparing the most commonly used indicators that measure technological innovation and surveying the dataset used in prior studies on the technological innovation 14

28 of China, this study chooses China domestic patent data constructed on the China patent abstract database (SIPO 2004), a dataset that has not been used before, as a indicator to measure the technological innovations of China. A detailed process of how to extract patent data from the Chinese patent abstract dataset and how to construct the provincial panel dataset are described in Chapter Patent Data: Limitations, Remedies, and Previous Studies Using patent statistics to assess Chinese technological innovation poses the following four well-known drawbacks. First, they are not complete measurement of inventions because not all inventions are patented. Griliches (1990) pointed out that patent statistics are an intermediate quantitative indicator of inventive activity between R&D input and social and economic inventive outputs. The quality of patent data as an accurate measurement of invention largely depends on the actual number of patented inventions. In addition, the patentability of an invention largely depends on the industry to which the invention belongs, and the propensity to patent an invention could also vary across industries, regions, and institutions. Currently, no approach that completely solves this problem of self-selection is available except controlling for industries or narrowing research down to particular sectors or regions. Research using patent data, thus, has observed only a fraction of inventions for which firms have selected to file patent rights. Fortunately, many technologies, particularly those in high tech sectors, are patentable, and the propensity to patent increases over time with extensive and intense intellectual property competition. Patent data, after all, are still a valid and leading indicator of new technological competition (NSF 2005, 6-21). 15

29 Another drawback of patent data is their inconsistent quality. That is, patents differ greatly in their technical and economic significance, and the quality of the patented inventions can vary considerably. The technical and economic value of patents are highly skewed: only a few patents are highly innovative and have significant economic value, while the majority of the patents are for mediocre inventions with low value. Griliches (1990) showed that about half of all patents are not renewed within 10 years which implies that majority of patents are either of low value, or that their value depreciated rapidly. Patent counts, therefore, may not provide an accurate estimation of the variables technology strength or economic payback that we want to use patent data to measure. These measurement problems can often be overcome if patent citation indicators are used. One of the most important contributions that facilitate studies in using patent citation indicators is the US Patent Citation Data File 3 that was originated by Francis Narin in early 1970s, and was made widely available by the National Bureau of Economic Research in The approach, to weight each patent counts by the total number of citations received by the patents opened a new series of studies that use patent citations to quantify the importance or value of patents (Hall and Jaffe 2001), to measure the technological innovation of firms, regions, and countries (Narin 1999), and to trace knowledge flow among institutions, regions (Jaffe and Trajtenberg 1996), and countries (Hu and Jaffe 2003). Jaffe and Trajtenberg (2002) provide a comprehensive review of this method and early studies using patent citation indicators. Besides the NBER research, other influential studies using patent citation statistics includes Narin 3 See NBER U.S. Patent Citation Data File at 16

30 (1994, 1997), and works of European scholars, such as Schmoch (1993), Harhoff (1999), and Mayer (2000, 2001). Unfortunately, patent citation information is not included in the current Chinese patent abstract database, to use patent citations to correct possible bias in patent counts is currently not applicable to Chinese domestic patents. Thus, to reduce heterogeneity in patent quality, this study uses patent grants rather than patent applications as the indicator to measure the technological innovation of Chinese firms, because compared with patent applications, patent grants have passed the additional scrutiny of patent examiners for novelty and utility which filters out the low quality patent applications that do not meet those criteria, although granted patents is still highly variable. In addition, granted patent, compared with patent applications, has less measurement errors, as a indicator for innovation because granted patent is one step closer to the commercialized new products or processes than patent applications. A further disadvantage to patent data is that patents are also inconsistency across industries and fields, which vary considerably in their propensity to patent inventions. This inconsistency can be reduced by breaking down patent counts by industries. Instead of using one comprehensive patent counts per region, this paper divided the patents of each region into six broad industrial sectors: electricity and electronics, chemical and pharmaceutical, process engineering, mechanical engineering, consumer goods, and instruments 4. 4 The other technical difficulty, as Griliches (1990) pointed out, is how to match patent data that are organized by substantive patent classes with industrial classifications. There are several available concordances between the patent class and standard industrial classification. This study applies a broad match up provided by the OECD patent manual (OECD 1994). 17

31 A final drawback of using patent data to evaluate technological innovation is that patenting activities could be affected by changes in laws and regulations. For example, studies by Henderson et al. (1998), Mowery et al. (2002), and Thursby and Thursby (2003) show that the enforcement of the Bayh-Dole Act 5 that allows universities to patent in 1980 has been the determining factor behind the recent dramatic increase in U.S. university patents. Technically, in a regression analysis, changes in patent trends caused by new or modified policies can be captured by adding time dummy variables that indicate the time when a policy was in force in the regression model. Despite the drawbacks of patent statistics, patent counts are still a plausible measurement of technological innovation and have been widely used, for decades, in various empirical studies and policy reports to measure innovation capability and technology trajectories and as the dependent variable to identify the factors behind invention activities and technological changes. Science and Engineering Indicators (NSF 2005) dedicated a special section to the use of patent counts to measure inventive output over time for the U.S. as well as foreign countries. The results show that corporations from Japan, the U.S., and Germany are the three top owners of worldwide intellectual properties based on the data provided from the World Intellectual Property Organization (WIPO) and that technology developments in developing countries rely on technology diffusion from these three and other industrialized counties. 5 The Bayh-Dole Act allows for the transfer of exclusive control over many government-funded inventions to universities and businesses operating under federal contracts for the purpose of further development and commercialization ( 6 See 18

32 The Organization for Economic Co-operation and Development (OECD) established a patent project 7 to improve the development of, methodologies, and the collection and the accessibility of patent data. Its patent database includes patents applied to the European Patent Office (EPO) and the United States Patent and Trademark Office (USPTO). Recently, the OECD has also constructed triadic patent families that link patents from all three main patent offices: the EPO, the Japanese Patent Office (JPO), and the USPTO. Patents included in the triadic patent families generally have higher economic value than those that applied patent rights in only one country because the patent applicants of the former are willing to pay extra application fees to protect the inventions worldwide. The details of this new database are provided in Dernis and Khan (2004) 8. In addition to this information, the OECD (1994) 9 presents examples on how to use various types of patent counts for cross-firm, cross-sector, and cross-country comparison studies. The report also provides a useful concordance to match IPC with specific technology groups. Dernis and Guellec (2001) 10 summarized the three sets of information presented in a patent document: (1) technical features, which include the technical classification to which the invention belongs, the list of claims that defines the field of coverage of the patent, and a list of patents and scientific papers cited; (2) information that relates to the development of the invention, including the name(s) and address(es) of the inventor(s) that, name(s) and address(es) of the applicant(s); and (3) a history of the patent application, including the date of application, the date of publication, the date of denial or 7 See 8 See 9 See 10 See 19

33 withdrawal, the date of the grant, the date of termination, and the priority date (the date of first filing worldwide). Based on this information, patents can be summed up by countries of inventors or applicants, by technological classifications, and by organizations to which the patent rights have been assigned. Besides used as a simple measurement indicator, patent counts are also one of the main dependent variables used in quantitative studies of technological innovation. One of the main research questions these studies focused on is: what the relationship between R&D investment and patent outputs? Jaffe and Trajtenberg (2002) summarized the main findings of early research as follows: In the cross-section, patents are roughly proportional to R&D, with the ratio varying by industry and being higher for small firms. For particular firms over time (the within panel dimension), patents are correlated with R&D, typically with decreasing returns to R&D, with the strongest relationship being simultaneous and contemporaneous between R&D and patent applications. In multivariate models including R&D, patents and performance measures (e.g., productivity growth, profitability, market value), most of the information is in the correlation between R&D and performance. Patent counts are more weakly explanatory power once R&D is included. Detail information on the technological composition of firms patents can be used to locate firms research programs in technology space. variations across this space in technological opportunity and spillovers of R&D have measurable effects on research performance (6-7). Earlier studies mentioned by Jaffe and Trajtenberg focused primarily on testing the relationship between firm patents counts and firm R&D. Recently, the implementation of patent counts has been expanded to public sectors and studies of regional and national innovation systems. 20

34 Jaffe (1989) showed the empirical evidence of the geographical spillover effects from university research on commercial innovation. Using state-level panel data, he found that corporate patents were determined by not only R&D conducted by corporations but also R&D performed by universities located in the same state. The positive impact from university R&D on corporate patents is roughly one-tenth of the impact from corporate R&D. This spillover effect is larger and more significant in scientific-based sectors such as drugs, chemicals, and electronics and smaller in traditional sectors such as mechanical arts. Acs et al. (2002) compared the patent counts and LBIO innovation counts using the model developed by Jaffe (1989). Regressing patent counts and LBIO innovation counts on a vector of explanatory variables measuring both industry and university R&D, the paper found no significant difference between the results of the regressions using patent counts and those using LBIO innovation counts. Both regressions revealed significant academic research spillover effects from university R&D to corporate patent outputs with a similar size as Jaffe (1989) found. Trajtenberg (2001) evaluated the innovation in Israel from 1968 to 1997 using over 7,000 Israeli patents taken in the United States. He found out the number of Israeli patents in the United States was highly correlated with current and lagging domestic industrial R&D. Furman et al. (2002) presented an empirical examination of the determinants of country-level production of international patents. They found that the performance of a country s innovation output, measured by its international patent counts, was not only determined by innovation inputs, such as R&D manpower and expenditures, but also affected by research performed by the academic sector, openness 21

35 to international trade, the knowledge stock, the degree of technological specialization, and the extent of intellectual property protection. These studies not only proved the use of patent counts in regression analysis is a valid assessment tool but also suggested that a firm s technological innovation capabilities, measured by its patenting activities, are actually determined by many external factors beyond corporate R&D. In addition to including more explanatory variables in their research models, quantitative studies employing patent statistics have also improved by introducing count data fixed effect regression methods. The technical aspects of the patent statistics regression analysis will be discussed in Chapter Determinants of a Firm s Technological Innovation Capability The technological innovation performance of a firm is not only determined by the internal R&D investment of a firm but also the external environment, accessibility of external finance and knowledge sources, and the technology trajectories of the firm. Freeman and Soete (1997) summarized the characteristics of successful innovating firms as follows: Strong in-house professional R&D; performance of basic research or close connections with those conducting such research; the use of patents to gain protection and to bargain with competitors; large enough size to finance fairly heavy R&D expenditure over long periods; shorter lead times than competitors; readiness to take high risks; early and imaginative identification of a potential market; careful attention to the potential market and substantial efforts to involve, educate and assist users; entrepreneurship strong enough effectively to co-ordinate R&D, production and 22

36 marketing; good communication with the outside scientific world as well as with customers (203). Several researchers have suggested a number of driving forces for innovation. Based on the concepts of the innovation system (Lundvall 1992; Nelson 1993; Edquist 1997) and Porter s five forces industry cluster theory (Porter 1990), Pedmore (1998) listed various sources of innovation and channels for the flow of innovation for industrial enterprises: in-house R&D within the enterprise; external learning from commercial competitors, materials and equipment suppliers, consultants, and clients or customers; knowledge spillovers from universities and government laboratories and technical institutes; and knowledge acquired through generally available information through patent disclosures, journals, conferences, fairs, or exhibitions. The flow of innovation could take the form of acquisition of equipment, services, intellectual property, or human capital. Another driving force for innovation is market demand, both domestic and foreign. Schmookler (1966) found strong empirical evidence for the role of market forces in shaping the rate and direction of inventive activity. He pointed out that innovation is a two-sided activity. On the one hand, it involves a technology push: the creation of new scientific and technological knowledge; on the other hand, the pull of market demand involves the recognition of the potential for new products and processes Learning from the Outside: The Bottom-up Model The literature on the development of industrial technology in China offers two main different explanations for its growth in the last several decades. One reasons involves the application of the bottom-up model, suggested by the studies of successful stories of Asian newly industrializing economies (NIEs) (Kim 1997; Hobday 1995). This model 23