International Trade and Knowledge Diffusion: A survey of recent empirical evidence
چکیده
In 1990, Zvi Griliches wrote that “a major unresolved issue in the area of economics of technology is the identification and measurement of R&D spillovers, the benefits that one company or industry receives from the R&D activity of another.” In the years since, much effort has been devoted to understanding and quantifying international R&D spillovers. This paper surveys the accumulated evidence and discusses promising new directions in recent work. Boston University School of Management, 595 Commonwealth Ave., Room 522H, Boston, MA, 02215. email: mmacgarv@bu.edu This paper borrows heavily from my Ph.D. dissertation at the University of California at Berkeley. I am grateful to Bronwyn Hall for her guidance and support. Introduction In 1990, Zvi Griliches wrote that “a major unresolved issue in the area of economics of technology is the identification and measurement of R&D spillovers, the benefits that one company or industry receives from the R&D activity of another.” In the years since, much effort has been devoted to understanding and quantifying these R&D spillovers. Because the “spillovers” arising from technological innovation can cross national boundaries, the effect of innovation on national comparative advantage depends crucially on the geographic scope of its diffusion. This paper surveys the empirical evidence on the factors that affect the rate of diffusion of new technological knowledge across borders. It also highlights my dissertation research on international trade and knowledge diffusion, and suggests some potential new directions for this line of research. The Institute for International Economics suggests that "exports are a conduit for information -competitively valuable information about innovation, rivals, and regulation." The empirical evidence on whether international trade is a conduit for technology diffusion, however, remains contradictory. This paper considers the following questions: Do firms and countries learn about foreign technology through their activities in international markets? How do the different variables that characterize international economic relationships (i.e., trade, investment, communication, proximity) compare as channels for technological diffusion? Which firms and countries learn the most about foreign innovation? What are the specific mechanisms through which inventors learn about foreign technology? 1 Griliches (1990), p. 1688 2 Much of contemporary trade theory is concerned with the extent to which comparative advantage can be determined by productivity improvements ensuing from technological progress. 3 p. 20, Why Exports Really Matter!, Richardson and Rindal (1995) 2. A review of the empirical literature Griliches (1979) introduced a model of inter-industry knowledge spillovers in which the amount of knowledge “borrowed” by a given industry i from all other industries is the sum of knowledge in those industries weighted by their technological closeness, or relevance, to industry i. This basic framework, in which the stock of knowledge in one segment of the economy (be it a firm, industry, or country) spills over and benefits another segment as an increasing function of the interaction or relatedness between the two segments, permeates essentially all research on the international diffusion of technological knowledge. The framework was taken up by Coe and Helpman (1995), who estimated a very large elasticity of domestic total factor productivity (TFP) with respect to foreign research and development (R&D) capital stocks weighted by the share of imports from the foreign country in the domestic country’s total imports. Coe and Helpman estimate a domestic R&D elasticity of 23% among G-7 countries. Interpretations of this finding grew more cautious when Keller (1997) obtained results similar to Coe and Helpman's after replacing actual trade weights with randomly chosen numbers. Keller (2001) has extended the analysis of R&D-productivity spillovers to include the influence of FDI, language barriers, and geographic distance at the industry level. He concludes that while import composition might not matter when the trading partners in question are symmetric, for countries that receive a large share of imports from one country, they do matter. Xu and Wang (1999) find a significant positive relationship between imports and R&D spillovers in a Coe-and-Helpman-style framework when attention is restricted to trade in capital goods. Other research at the aggregate level has focused on patents. Eaton and Kortum (1996) find that, after controlling for geographic distance and other influences, bilateral imports do not help to predict international patent applications. Sjöholm (1996) finds that Swedish citations to foreign patents are correlated with bilateral import shares. Peri (2003) studies patterns of patent citations across regions, and finds that the negative effect of geographic distance on knowledge flows is smaller than its effect on international trade. Peri finds that while sharing a common language significantly increases citations across regions, belonging to a common trade zone is not significantly associated with citations. MacGarvie (2003) finds evidence of a positive relationship between aggregate bilateral trade and European patent citations at the country level, but no significant relationship between USPTO citations and trade when controlling for other correlates of cross-country citations like language similarities, geographic distance, and bilateral FDI. Sharing a common language has a positive and significant effect on the flow of citations across countries, and distance is negatively associated with crosscountry citations. Bilateral FDI has a positive, but insignificant, coefficient. Table 1 summarizes the findings of studies that analyze the relationship between international trade and patent citations at the aggregate level. Empirical research on international trade has recently begun to examine how exporting firms differ from other firms in an industry. Among the central findings of this literature are that exporters tend to be larger, more productive, and supply higher-quality products than non-exporting firms. This finding has led researchers to ask whether firms 4 The use of patent citations as an indicator of knowledge spillovers is discussed in detail in the following section. 5 See James Tybout (2001) "Plantand Firm-level Evidence on " New" Trade Theories", Handbook of International Economics export because they are more productive, or whether they learn something in the export market that confers a productivity advantage. Conducting Granger causality tests on productivity and export data from Mexico, Colombia and Morocco, Clerides, Lach and Tybout (1998) find no evidence that exporting improves productivity, nor do Bernard and Jensen (1998) with respect to American manufacturing plants. In contrast, similar tests on data from Chinese (Kraay (1997)) and African (Bigsten et al. (1999)) firms do find evidence that exporting causes productivity improvements. Van Biesebroeck (2003) finds that productivity in firms in Sub-Saharan Africa increases after they enter export markets. Blalock and Gertler (2002) find that Indonesian firms experience a one-time productivity jump upon entering export markets. Hallward et al. (2002) test the hypothesis that firms self-select to compete in export markets. After looking at firms' export status in the first year of their existence, they conclude that it is in aiming for export markets that firms make decisions about investment, training, technology and other factors that enhance productivity. Shaver and Salomon (2002) find that Spanish exporters are more innovative, and interpret this as evidence of learning by exporting. Haddad and Harrison (1999) assess the contribution of FDI spillovers to the productivity of Moroccan plants. Aitken and Harrison (1999) investigate the relationship between foreign ownership in an industry and the productivity of domestic firms in that industry. They find that while foreign ownership is associated with higher productivity for small Venezuelan firms, this foreign investment is negatively associated with productivity spillovers to domestically-owned plants. Kogut and Chang (1991) document the "asset-seeking" motive for FDI, in which firms establish a presence in a foreign country in order to gain access to local technology. Branstetter (2000) finds evidence that FDI flows increase knowledge spillovers between Japan and the United States. While the literature on spillovers from FDI is well developed, research on learning through trade is less extensive. The FDI literature emphasizes three mechanisms for technology diffusion through FDI: demonstration effects, labor mobility, and linkages with buyers and suppliers. Might these same mechanisms facilitate technology diffusion through exporting and importing? Demonstration effects, which operate when firms observe and imitate the products or practices of foreign firms, seem likely to be affected by trade -for example, when foreign firms copy or modify aspects of exporters' products, or when the exporter's study of the foreign market yields information about foreign innovations. An example of this type of learning is the evolution of Fuji Film Co. during the 1970s and 1980s, when its market share in the United States expanded substantially due to technical improvements in its product that brought it in line with Kodak, the market leader. Another example is the case of Ellis Agricultural Equipment of Australia, which imports American-made harvesters and distributes them to Australian farmers. When Ellis noticed that the American harvesters lacked features that would make them more compatible with the unique Australian soil, they invented a new harvester based on the American version and applied for a patent to protect the invention. Labor mobility seems a less likely candidate, with the exception of exporters with R&D centers in the foreign market that hire researchers away from their domestic 6 See Kumar (1996), Blomström and Kokko (1997), and Moran (2001) for reviews of the literature on learning and FDI. 7 See http://www.ipaustralia.au, the Australian Patent Office website. competitors. Linkages with buyers and suppliers seem likely to be an important source of learning for exporters and importers. As an example of learning through communication with suppliers, consider Usinor, a French manufacturer of steel products and the largest supplier of auto body components in Europe. Usinor emphasizes "close technical partnerships with customers," communicating with buyers' technical departments, tailoring products to buyers' specifications and providing technical assistance. Usinor's USPTO patent number 6,398,286 is a reinforced light-weight automobile engine hood. It cites patents assigned to Mazda, Chrysler, Suzuki Motor, and DaimlerChrysler, as well as other metals producers (Azko and Alcoa). These citations could reflect information obtained through communication with buyers over product specifications. An example of how learning by exporting can improve productivity is the case of Fluid Management, a small Illinois manufacturer of paint-mixing equipment, which after entering international markets discovered a foreign supplier of "revolutionary" machinery now used in its production process. Evidence of how firms learn by exporting is offered by Rhee, Ross-Larsen and Pursell (1984), who found that almost half the Korean exporters they surveyed claimed to have directly benefited from technical information provided by foreign buyers, and that approximately forty percent said that contacts with foreign buyers had improved techniques of quality control and production. These benefits were created through visits both by foreign engineers to Korean plants, and by Korean engineers to the foreign buyers. The exchange of blueprints and specifications, and feedback on design, quality 8 See section 5 of this paper for evidence that exporters obtain access to foreign technology by hiring foreign workers. 9 The information reported here was obtained from Usinor's 2001 Annual Report and the European Patent Office website: http://www.european-patent-office.org. 10 Richardson and Rindal (1995), p. 20. and technical performance were also cited as means by which Korean exporters benefited from their relationships with foreign buyers. In a related study, Pack and Westphal (1986) found a common pattern. Starting from technological capability used or acquired in previous production a firm would reverse engineer a new product. The firm would then agree to supply the product to a foreign buyer, but would be unable to supply the product according to the buyer's expectations without some form of technology transfer from the export buyer. With the collaboration of the firm's R&D staff, these transfers would cumulate over time and the firm would begin to diversify and upgrade its product varieties. Westphal (2001), stressing that "the importance of technology transfers from export buyers can hardly be overstated," explains that export buyers provide technology through blueprints and specifications, technical advice offered during plant visits, and consultations with the buyer's engineers. Of course, not all forms of international trade are ideally suited for information gathering by firms. Imports of agricultural products and raw materials seem unlikely to generate significant knowledge flows to the firm. Firms that export to lower-income countries because their products have been rendered obsolete by innovation on the part of domestic competitors, and firms that seek to reduce costs by investing in production in low-wage countries, are also less likely to benefit from being exposed to foreign markets. 3. Patent Citations One of the obstacles faced by empirical research on knowledge diffusion is the difficulty of measuring knowledge spillovers. Information on patent citations can provide a window onto this process of diffusion. When a patent is granted, its documentation includes a list of other patents whose claims are similar. These references, or "citations", in some sense resemble citations in academic papers. One difference, however, is that patent citations are added by the patent examiner as well as by the inventor, and therefore provide a more objective measure of the technical similarity of inventions. Citations delimit the property rights of the citing patent by listing the “prior art” excluded from its claim. By combining information on the date and place of origin and technological area of a citing patent with corresponding information on the characteristics of the cited patent, one can use patent citations as a way of tracking the influence of past inventions across time and geographic boundaries. These data have been used to investigate patterns of knowledge spillovers (Caballero and Jaffe (1993), Jaffe, Trajtenberg and Henderson (1993), and Jaffe and Trajtenberg (1999)), the “basicness” or importance of particular inventions (Henderson, Jaffe and Trajtenberg (1998)), and the relationship between firms' market values and citation-weighted patents (Hall, Jaffe and Trajtenberg (2000)). Jaffe and Trajtenberg (1999) find that patents whose inventors reside in the same country are typically 30% to 80% more likely to cite each other than inventors from other countries, and that these citations are made sooner than citations in other countries. Almeida (1996) analyzes 144 patents held by semiconductor producers and finds evidence of knowledge flows between multinational firms operating in the US and American inventors. Frost (2001) uses a larger sample of USPTO patents to characterize the extent to which multinationals learn from foreign subsidiaries. Globerman, Kokko and Sjöholm (2000) find that outward FDI increases citations to foreign patents among 11 Griliches (1990) notes that the fact that most citations are added by examiners contributes to their objectivity, but also to their questionable status as indicators of influence rather than importance. “On the other hand, they bring us closer to something that might be interpreted as measuring the social rather than just the private returns to these patents.” p. 1689 Swedish firms, but that inward FDI has no significant effect on citations. Branstetter (2000) finds evidence that FDI flows increase knowledge spillovers between Japan and the United States. MacGarvie (2003) shows that exporters and importers cite more patents from their markets of destination and origin than do non-trading firms. However, while the association between imports and citations is robust to corrections for endogeneity and selection bias, the relationship between exporting and citations is not. Singh (2003) finds that foreign subsidiaries of multinationals cite host-country patents more often than they are cited by host-country inventors, suggesting that foreign subsidiaries gain more in terms of local knowledge than they contribute. These findings are summarized in Table 2. Some have questioned the value of the information contained in patents and patent citations. Thompson and Fox Kean (2002) find that the geographical localization described by Jaffe, Trajtenberg and Henderson (1993) is largely spurious at the regional level, although international localization effects persist. A study on the meaning of patent citations conducted by Jaffe, Trajtenberg and Fogarty, in which a cohort of patentees and the inventors of the patents they cited were interviewed, finds that patent citations can indeed be thought of as a legitimate (though noisy) measure of communication between inventors. Duguet and MacGarvie (2002) provide new evidence on the legitimacy of patent citations as a measure of knowledge flows. France's Ministère de l'industrie et du commerce extérieur conducted a survey in 1993 in which roughly 4000 firms were asked 12 The authors asked patentees about two patents they had cited and a third "placebo" patent that was similar but which they had not cited. One of the inventors of the cited patents was also surveyed. They found that "inventors report significant communication, at least some of which is in forms that suggest spillovers from the cited inventor to the citing inventor." whether they had obtained or disseminated new technologies via a number of specified sources in France, the European Community (EC), Non-EC Europe, the United States, Japan, or another country. Duguet and MacGarvie test whether firms that claimed to have obtained new technology from a foreign country cite more patents from that country. We find that citations are indeed significantly correlated with firms' statements about technological diffusion for several of the channels and regions covered by the survey. In particular, backward citations per patent are positively and significantly correlated with learning through R&D collaboration, licensing foreign technology, mergers and acquisitions, and equipment purchases. Forward citations are positively and significantly correlated with the firms' perceived transfer of technology through equipment sales, mergers, and joint ventures and alliances. Patents are only granted for inventions that are novel, and as a result, patent citations capture the contribution or similarity of other patented technologies to the invention -they do not measure technology transfer that does not cumulate in new inventions. In other words, I will assess whether exporting firms' inventions owe more to patented technologies from abroad than do the inventions of non-exporters, but I cannot assess whether exporters are more likely to, for example, reverse-engineer or exactly replicate a product or process common in an export market. In this paper's analysis of citations, that type of non-inventive activity will only be captured to the extent that it contributes to learning that results in a patent. The results based on the innovation survey can tell us more, however, and I use these to supplement the results based on citations. 4. Firm-level analysis of trade and knowledge diffusion As discussed in section 2, the answer to the question of whether exporting has a causal effect on firms’ productivity remains ambiguous. The failure of studies that attempt to identify the causal effect of exporting on productivity to reach a consensus suggests the importance of taking a different approach to answering the question. By using patent citations to measure learning, MacGarvie (2003) is able to test explicitly whether exporters learn more about foreign technology than do non-exporters. In my dissertation, I use a new firm-level trade dataset assembled by researchers at INSEE and information on French firms' patents and patent citations. The trade dataset contains information on firms' exports and imports by country, year and product between 1986 and 1992. Using this uniquely detailed source, I test whether the firms' inventions are influenced by foreign inventions in a way that is correlated with their activities in foreign markets. I also test whether exporting and importing firms' patents are cited more often by foreign patents. Thus, I obtain estimates of how much French firms learn about foreign technology as well as how much their counterparts in the United States, Japan, and Europe learn about French technology through trade. Table 2 summarizes the