Geographic Localization of International Technology Diffusion∗

Income convergence across countries turns on whether technological knowledge spillovers are global or local. I estimate the amount of spillovers from R&D expenditures on a geographic basis, using a new data set which encompasses most of the world’s innovative activity between 1970 and 1995. I find that technology is to a substantial degree local, not global, as the benefits from spillovers are declining with distance. The distance at which the amount of spillovers is halved is about 1,200 kilometers. I also find that over time, technological knowledge has become considerably more global. Moreover, language skills are important for spillover diffusion. ∗I thank Ulrich Ammon, Don Andrews, Steve Bronars, Peter Debaere, Rob Feenstra, Don Fullerton, Dan Hamermesh, Sam Kortum, Preston McAfee, Paul Schreyer, Carol Shiue, Tony Venables, and seminar participants at IGIER/Bocconi, the NBER, Rutgers, Texas A&M, Tilburg, the University of Texas, and the World Bank for helpful comments. Criticism and suggestions by two referees have been particularly useful. I would like to thank Kim Figueira as well as Colin Webb for help with data. This work has been supported by the National Science Foundation under grant number SES-9818902. Anne Golla and Paula Hernandez-Verme have provided excellent research assistance. †Department of Economics, University of Texas, Austin, TX 78712; email: [email protected] ‡Also affliliated with the NBER, Cambridge, MA, and the CEPR, London. 1 Income convergence in a large class of models turns on whether the scope of technological knowledge spillovers is global or local. Global spillovers favor convergence at the level of cities and regions, as well as at the level of countries, whereas spillovers that are geographically limited in scope can lead to economic clusters with persistently different levels of output per capita. Whether or not the industrialized countries will remain the rich permanently, and what are the prospects for less developed countries to catch up, are both questions whose answers hinge on the scope of knowledge spillovers. It is widely held that technological knowledge is truly global because increasing economic integration through trade as well as new means of telecommunications and the internet ensure that people in all countries have access to the same pool of knowledge. Even differences in the technology that is actually employed (as documented in James Harrigan 1997, e.g.) are consistent with a global pool of technology if the rate of complementary human and physical capital investments or the incentive to adopt new technology varies across countries.1 Alternatively, technological knowledge may be to some extent local. Helsinki, for instance, is located about 1,500 kilometers away from Bonn, around 6,900 kilometers from Washington, D.C., and 7,800 kilometers from Tokyo, while the distance from Canberra to Bonn, Washington, and Tokyo is 16,500, 16,000, and 8,000 kilometers, respectively. If knowledge spillovers are local, then productivity in Finland should, ceteris paribus, be higher than in Australia, because the former is closer than the latter to Germany, the U.S., and Japan, the three countries that account for more than 75 percent of the world’s spending on research and development (R&D). I will investigate whether knowledge spillovers are global or local by examining whether the distance between countries affects the magnitude of productivity gains from each others’ R&D spending. Geographic distance should not matter for international technology diffusion if there is a global pool of technological knowledge or a country’s technology level depends only on idiosyncratic non-spatial These points are emphasized by Gregory Mankiw (1995) and Edward Prescott (1998), respectively.