convergence and polarization in global income levels: a

Research Policy 32 (2003) 1055–1079

Convergence and polarization in global income levels: a review ofrecent results on the role of international technology diffusion

Guan Gonga, Wolfgang Kellera,b,c,d,∗a University of Texas, Austin, TX, USA

b Economics Department, Brown University, Box B, 64 Waterman Street, Providence, RI 02912, USAc National Bureau of Economic Research (NBER) Summer Institute, Cambridge, MA, UK

d Center for Economic and Policy Research (CEPR), London, UK

Received 25 October 2001; received in revised form 20 August 2002; accepted 11 September 2002

Abstract

We review the recent literature on technological change and diffusion to shed new light on the evolution of the world’scross-country income distribution. Technology is viewed as non-rival knowledge in the sense that firms in more than onecountry can simultaneously use it. R&D investments generate often also a return outside the innovating firm itself; theseknowledge externalities are called technology spillovers. We emphasize that technology is to some extent tacit, and technologydiffusion often involves the face-to-face interaction of people. Our paper reviews the evidence on whether international trade,foreign direct investment, and other cross-border activities are important for technology diffusion.© 2002 Elsevier Science B.V. All rights reserved.

Keywords: Technology spillovers; Tacit knowledge; Foreign direct investment; Trade; Communications

1. Introduction

Technology is important in explaining income lev-els across countries. The accumulation of physical andhuman capital matters as well, but that cannot ex-plain much of today’s cross-country incomes differ-ences (Easterly and Levine, 2001; Prescott, 1998). Ifthe rate of technical change differs across countries,this affects the world’s distribution of income. Newinformation and communication technologies (ICTs)have been developed relatively fast in the United States(US), for example, and this might help explainingwhy the US lead in per-capita income over Japan

∗ Corresponding author. Tel.:+1-401-863-9030;fax: +1-401-863-1970.E-mail address: [email protected] (W. Keller).

has increased from 10% in 1990 to 20% by 1999(Economist, 2000; McKinsey, 2000).

Recent work has shown, however, that the majorsources of technical change leading to productivitygrowth in OECD countries are not domestic; instead,they lie abroad (Eaton and Kortum, 1999; Keller,2002a).1 The international diffusion of technology istherefore a major determinant of per-capita income inthe world. Because most developing countries spendrelatively less on basic science and innovations—formal R&D spending, for instance, is highly con-centrated in a handful of OECD countries—poorercountries rely even more on foreign sources of pro-ductivity growth than OECD countries do.

1 Eaton and Kortum (1999)estimate for example that for-eign research accounts for 87% of productivity growth in France(Table 5). SeeSection 3for further results.

0048-7333/02/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved.PII: S0048-7333(02)00136-1

1056 G. Gong, W. Keller / Research Policy 32 (2003) 1055–1079

This means that convergence in income turns on thedegree of international technology diffusion. Strongdiffusion is a force towards convergence, because itequalizes differences in technology across countries.Conversely, the absence of international technologydiffusion favors divergence. Technical change in thepresence of non-uniform international technologydiffusion is at the heart of the recent “digital divide”discussion, for example, the widespread fear that thedevelopment of the Internet might not lead to con-vergence, but instead to a further polarization of theworld’s income distribution.

While it has been recognized since the classic‘Solow residual’ paper (Solow, 1957) that rates of fac-tor accumulation do not account for the major part ofeconomic growth, the view that technological changehas both domestic and foreign sources is less common.Arguably, the rapidly rising level of economic inte-gration in the late 20th century, fostered by advancesin transportation as well as in information and com-munication technology, makes the exclusive focus ondomestic technological change obsolete. Sometimesthe reason for productivity increases does indeed liein purely domestic activities, such as the learningeffects resulting from cumulative production for do-mestic demand. However, productivity also increasesdue to learning through the interaction between for-eign and domestic firms. The greater importance oftechnology adoption from abroad—versus domes-tic technical change—in less developed comparedwith more developed countries suggests that learn-ing through international economic activity might beparticularly important for less developed countries.

This paper takes a look at the evidence to ex-pand on these ideas. In the next section, we discussthe concept of international technology diffusion.We also provide some references to the underlyingtheories; the sections that follow are first and fore-most a review of recent empirical work. Basic resultson international technology diffusion are discussedin Section 3. Section 4is devoted to internationaltrade, foreign direct investment, and other channelsof diffusion. In Section 5, we review the evidenceon the heterogeneity of diffusion across productsand industries.Section 6discusses findings on thegeographic localization of international technologydiffusion, and Section 7 examines how this haschanged over time. InSection 8, we discuss some

evidence on major country-specific determinants forsuccessful international technology diffusion. Finally,Section 9summarizes the major findings, suggestsdirections for future research, and discusses policyimplications.

2. Conceptual issues

2.1. International technology diffusion in this paper

It is central to much of the recent work to view tech-nology as knowledge, as emphasized in the theoriesof endogenous technical change that emerged about10 years ago (Aghion and Howitt, 1992; Grossmanand Helpman, 1991; Romer, 1990; Segerstrom et al.,1990).2 In this framework, technology has three majorcharacteristics.3

1. Technology is non-rival in the sense that themarginal costs for an additional firm or individualto use the technology are negligible.

2. The return to investments towards new technologyare partly private and partly public.

3. Technological change is the outcome of activi-ties by private agents who intentionally devoteresources towards the invention of new productsand processes.

Out of these, points 1 and 2 are key for ourpurposes. The first characteristic means that tech-nological knowledge can serve users beyond thosewho are currently employing it without raising thecosts to the original set of users. Other authorshave coined the terms ‘perfectly expansible’ (David,1992) and ‘infinitely expansible’ (Quah, 2001a,b) to

2 SeeGrossman and Helpman (1995)and Aghion and Howitt(1998)for broader overviews. I will also discuss some related workon learning-by-doing and human capital accumulation, as empha-sized byLucas (1993, 1988), that falls into the broad category ofmodels of knowledge accumulation.

3 SeeRomer (1990). Many of these ideas have been discussedin the literature before; important contributors include Paul David,Giovanni Dosi, Robert Evenson, Jan Fagerberg, Richard Nel-son, Keith Pavitt, Nathan Rosenberg, Luc Soete, Sidney Winter,Larry Westphal, and others (seeFagerberg, 1994andEvenson andWestphal, 1995for overviews). What distinguishes the recent workis that it includes fully specified general equilibrium models. Thismeans that these technology effects can in principle be simulatedand estimated in a well-defined framework.

G. Gong, W. Keller / Research Policy 32 (2003) 1055–1079 1057

positively define this characteristic. It distinguishesknowledge from rival factor inputs such as humanand physical capital; the latter can only be used byone firm at a time, or put differently, the marginalcosts to use the same factor for a second firm areinfinite.

The partially private, partially public nature of thereturn to technological investments implies that whilethere is a force that might be strong enough to sustainthe private incentive to innovate (the private return,which is often a temporary monopoly secured by apatent), technological investments may also createbenefits to firms and individuals external to the in-ventor by adding to their knowledge base (the publicreturn). These benefits are usually called knowledgespillovers. An example is that the design of a newproduct might speed up the invention of a competingproduct, because the second inventor can learn fromthe first by carefully studying the product, or even theproduction design.

One contribution of these theories of technicalchange is that they have supplied improved microfoundations for thinking about knowledge spillovers.This survey of international technology diffusion islargely an analysis of the empirical literature of in-ternational technology spillovers and their effects onproductivity. We do not attempt to provide a generalreview of the models of technology diffusion here,which is beyond the scope of our paper.4

Two basic mechanisms for international economicactivities to lead to technology diffusion have beenemphasized.5

(a) Direct learning about foreign technological know-ledge.

(b) Employing specialized and advanced intermediateproducts that have been invented abroad.

Technological knowledge in this literature is typi-cally the design, orblueprint, for a new intermediateproduct. Direct international learning about such tech-nology means that a blueprint is known not only toa firm in the country where the blueprint was first

4 The interested reader is referred to the survey byGeroski(2000). For an introduction to the broader literature, see alsoRosenberg (1976), David and Olsen (1992), Jaffe and Trajtenberg(1996), andStoneman and Kwon (1994).

5 SeeRivera-Batiz and Romer (1991)for an exposition.

developed (or firms, if there are domestic spillovers),it also becomes known to firms in other coun-tries. Such learning involves a positive externality—hence: spillover—if the technological knowledgeis obtained at less than the original cost to theinventor.

The productivity of domestic invention is assumedto be increasing in a country’s stock of knowledge,which itself is typically proportional to the number ofdomestically known product designs. This assumptioncaptures the idea that creating a new product becomeseasier as the number of already known product designsis larger. Thus, by adding to the domestic knowledgestock, international spillovers raise the productivityof domestic inventive activity. It might be called anactive spillover, in the sense that the foreign blueprintbecomes part of the domestic R&D laboratories’ stockof knowledge that can be actively used to invent newproducts.

According to point (b), technology diffuses interna-tionally through foreign intermediate goods. The ideais that employing the foreign intermediate good in-volves the implicit usage of the design knowledge thatwas created with the R&D investment of the foreigninventor. In this sense, the technological knowledgeof the blueprint is embodied in the intermediate good.As long as the intermediate good costs less than itsopportunity costs—which include the R&D costs ofproduct development— there is a gain from havingaccess to foreign intermediate goods. This might becalled a passive technology spillover: although animporting country has indirect access to the results offoreign R&D, the technological knowledge embodiedin the imported intermediate as such is not avail-able to domestic inventors—only the manufacturedoutcome of it is.

What kind of international activities could lead tosuch active or passive spillovers? Clearly, the lattermight be related to international trade and foreign di-rect investment (FDI) through intermediate goods im-ports and purchases from foreign-owned multinationalsubsidiaries, respectively. Employing the intermediategood that embodies technological knowledge is nec-essary for passive spillovers to occur. An implicationof this is that the patterns of passive internationaltechnology spillovers might share certain characteris-tics of international trade and FDI. For instance, thevolume of international trade between two countries


is declining with bilateral geographic distance (seeLeamer and Levinsohn, 1995). This suggests thatto the extent that passive spillovers are important,international technology diffusion might have such aspatial structure as well.

There could also be interesting relations betweenactive and passive spillovers. For instance, tradeliberalization and market deregulation might havea significant impact upon technological knowledgegeneration and diffusion.6 Trade liberalization makesimported intermediated goods—say, machinery—relatively cheaper, which induces a reduction of thedemand for possibly less advanced domestic inter-mediate goods and an expansion of imports of suchgoods. The substitution of imported for domesticmachinery might affect other forms of knowledgein the domestic economy. Just as older vintages oflocally produced machinery are rendered obsolete byimporting newer and more efficient machinery goods,so, too, do certain forms of human capital lose theirvalue. For example, the ‘in house’ engineers who werepreviously employed to extend the economic lifetimeof the domestic machinery goods in a world in whichnewer foreign machines were difficult or impossibleto obtain are now becoming redundant because theirskills are embodied in the more advanced importedmachinery equipment goods. This could relax some ofthe technical skill constraints that slow down the eco-nomic development of many less developed countries.At the same time, the adoption of the more advancedforeign technologies requires a new set of operationaland maintenance skills (Dahlman et al., 1987). As aresult, importing advanced foreign technology doesnot imply that domestic skill accumulation becomesunnecessary, only that a different skill profile, witha more flexible labor force, is required. The sub-stitution for domestic skilled labor through importsof intermediate goods might on net also lead to anincrease in the ratio of passive (‘embodied’) technol-ogy spillovers relative to active technology spillovers.This can occur if the direct effect of importing em-bodied foreign technology on the incentives for do-mestic skill accumulation in form of human capital

6 Detailed studies of the impact of trade liberalization policiesin Latin American provide some evidence in support of this point.See e.g.Katz (2000)and Cimoli (2001).

is strongly negative, so that an economy has to relyincreasingly on embodied technology from abroad.

With an increasing degree of ‘commoditization’of goods and services, passive spillovers might be-come the dominant way for technology diffusions.For instance, there is some evidence showing that theimportance of so-called “Internationally IntegratedProduction Systems” has increased significantly as apercentage of GDP during the course of the 1990sthroughout Latin America (Cimoli and Katz, 2001).For example, when Ford launched its Ford Tau-rus model in Argentina in 1974, this required some300,000 h of domestic engineering efforts over one anda half years to adapt the German-designed blue printsto the local production conditions. The domestic con-tent for such car was about to 90% of the total valueof the vehicle, and nearly 400 subcontractors suppliedparts and components. With the commoditization ofautomobiles, this localized technological regime witha large active spillover potential is gone. Today FordArgentina is part of a world wide integrated productionsystem, and its operation is strongly based on importedparts and components. The local content of productiondropped to less than 50%. Although the productivitygap between the Argentine automobile industry andthe world frontier has been somewhat shrinking overtime, the process in which local operations have in-creasingly turned into assembly activities might leadto an increasing isolation of peripheral countries fromthe places in the world where technology is generated.

In contrast to passive spillovers, the purchase ofintermediate goods is neither necessary nor sufficientfor spillovers of the active kind. The technologicalknowledge that allows the inventor to produce a newproduct (or operate a new process) is often summa-rized in a production plan, or blueprint. Typically, thisis now stored in electronic format, say, on a computerdiskette. This blueprint is non-rival, or, infinitely ex-pansible, in the above sense. What governs the accessto this technological knowledge, or, put differently,what are the major determinants of active interna-tional technology spillovers? There might be legalconstraints such as patents that allow the owner to ex-clude others from using the blueprint knowledge.7 Ifthe focus is on technological feasibility, however, the

7 Indeed, in most standard models, it is assumed that the knowl-edge is patent-protected.


knowledge codified in the blueprint can diffuse in anyway and pattern in which a file on a computer diskcan be distributed. And since the late 20th century, itcan be sent at very low cost over computer networksto even remote places in the world.

Should one therefore expect that active technologyspillovers are, in the absence of legal constraints,global? Not necessarily. Consider first that from thepoint of view of the inventor, the leaking out ofknowledge to others isthe opposite of what he or sheis interested in. Unless there is an explicit technologylicensing contract, the inventor has the incentive ofkeeping the knowledge secret. If international eco-nomic relations involve person-to-person contacts ofpeople who have the technological knowledge andothers who do not, this could make it more difficult forthe inventor to prevent knowledge spillovers from oc-curring. From the point of view of the learner, interna-tional contacts make it easier to copy foreign technolo-gies. At a broader level, international activity such asimporting, exporting, or FDI might also help to estab-lish and sustain channels of communication that stim-ulate cross-border learning of production methods,product design, organizational methods, consumerpreferences, and market conditions.8 The process oftechnology transfer that is described by product-cyclemodels is a good example of such learning effects(Vernon, 1966; Grossman and Helpman, 1991).

These considerations are strengthened when oneexplicitly recognizes the fact that the view of techno-logical knowledge that is limited to codified knowl-edge is too narrow. In addition to the information thatis codified in the blueprint, there is other informationthat must be acquired if the technological knowledgeis to be utilized effectively. Many careful studies oftechnology and how it is transferred conclude thatonly the broad outlines of technological knowledgeare codified—the remainder remains ‘tacit’ (Polanyi,1958).9 In Polanyi’s view, knowledge is to some ex-tent tacit because the person who is actively engagedin a problem-solving activity cannot necessarily de-

8 It might also be possible to acquire the technological knowl-edge embodied in an intermediate good by taking it apart andreverse-engineering it. This would require importing one unit ofa particular good, but not a substantial quantity of them.

9 The following draws also onArrow (1969), David (1992),Evenson and Westphal (1995), Teece (1977), von Hippel (1994),and references given there.

fine (and hence prescribe) what exactly he or she isdoing. In this view, technological knowledge is onlypartially codified because it is impossible or at leastvery costly to do so.

Teece (1977), for instance, finds that the non-codi-fied part of the costs of transferring technologybetween plants is substantial. In his sample of 26projects, he estimates that the costs of the transfer wereon average almost 20% of the total project costs.10

It is of course possible to convert tacit to codifiedknowledge (software expert systems do just that), andthe costs of doing this has fallen in some areas overrecent years. However, also more recent analysis hasshown that non-codified knowledge continues to beimportant for understanding patterns in the creationand diffusion of knowledge (von Hippel, 1994).

What are the implications of non-codified knowl-edge for the role of international economic activityin technology diffusion?Polanyi (1958, p. 53) arguesthat tacit knowledge can be passed on only “by exam-ple from master to apprentice”. A broader view is thatnon-codified knowledge is usually transferred throughdemonstrations, through personal instructions, as wellas through the provision of expert services (David,1992, p. 221). In general, if knowledge is partlynon-codified, then person-to-person communicationbecomes relatively more important for the diffusionof knowledge.

No doubt, the quality of communication bet-ween persons at different locations has dramaticallyimproved recently. However, person-to-person com-munication means often still face-to-face commu-nication. Together with the fact that it is costly forpeople to move in geographic space, this suggests thatthe higher is the relative importance of non-codifiedknowledge, the more are technology creation and dif-

10 For transferring machinery equipment technology, this sharewas 36% (Teece, 1977, p. 248). Teece’s estimates are a lowerbound for technology transfer to less developed countries, as hissample includes to two-thirds relatively advanced countries. Helists the following types of costs (245–246): pre-engineering oftechnological exchanges, costs associated with transferring theprocess/product design and the associated engineering, R&D per-sonnel costs during the transfer phase, pre-start-up training andexcess manufacturing costs. Teece also shows suggestive evidenceon when these costs are particularly high or low, for instance inrelation to similar production experience (–) or general level ofdevelopment (–).


fusion geographically centralized.11 Further, engag-ing in international economic activity such as tradeor FDI is invariably accompanied by communication,not infrequently in form of face-to-face contacts.From this perspective of partially-codified knowl-edge, it appears that not only passive spillovers—embodied technology in intermediate goods—butalso active spillovers are linked to the patterns ofinternational economic activity, instead of being uni-formly distributed (or distributable) throughout theworld.12

2.2. Alternative views on technological knowledgeand its diffusion

Alternative views includeMankiw (1995) andParente and Prescott (2000)who think of techno-logical knowledge as a global pool of knowledge,available to firms and individuals in all countries.Their explanations for differences in per-capita in-come across countries differ. In Mankiw’s view, theexplanation lies in differences in complementary fac-

11 Evidence supporting this view is provided invon Hippel(1994) and Feldman and Lichtenberg (1997). The formerfinds that when technological knowledge is non-codified, theproblem-solving activity must often be located on-site—it cannotbe done effectively from a distance. Von Hippel also finds thatwhen knowledge is non-codified in different locations, optimalbusiness strategy typically implies to adopt a sequential and iter-ative, not simultaneous, pattern of problem-solving.Feldman andLichtenberg (1997)construct measures of tacitness of knowledgefor their study of R&D activities in the European Union. Theyfind that the more tacit knowledge is, the more centrally locatedare the R&D activities.12 Note another distinction frequently made in the literature

on spillovers, that between rent and knowledge spillovers (e.g.Griliches, 1995). The former are productivity spillovers only in ameasurement sense, because they occur solely because, for exam-ple in computing the productivity of the intermediate good-usingindustry, one does not properly take account of the high qualityof the intermediate good. This leads to an overestimate of theproductivity of the intermediate-using industry. Major reasons forthat are unavailable, or slowly adjusted price indices. By con-trast, knowledge spillovers are ‘true’ spillovers in the sense thatthey involve a positive externality. In practice, it has been diffi-cult to separate these two types of spillovers empirically. Becausepassive spillovers as characterized above involve the purchase ofgoods, they might involve often an element of rent spillovers.Productivity calculations are also complicated by the need to dothis in a way that accounts for new goods (seeFeenstra et al.,1994).

tor accumulation, especially of physical and humancapital.13 Recent analysis suggests that this hypothesisis difficult to maintain (Klenow and Rodriguez-Clare,1997; Hall and Jones, 1999). Instead, a large part ofcross-country differences in income per-capita haveto do with total factor productivity (TFP).14

In Parente and Prescott (2000)view, even thoughtechnological knowledge is global, productivity dif-ferences are not primarily explained by differences inthe availability of rival factors. Instead, the main causeof cross-country productivity differences is that thereexist differences in theactually employed technolog-ical knowledge. According to Parente and Prescott,this results in turn from cross-country differences inthe countries’ resistance to adopt the world’s frontiertechnological knowledge.

This emphasis on policy differences resulting fromdifferent political–economy equilibria is a priori plau-sible. At the same time, while policy differences playa certain role, it seems at this point ambitious toshow that policy differences are the central reason forcross-country TFP differences. There are plenty of in-stances where linking TFP differences to the activitiesof lobbies, state bureaucracies, or self-interested politi-cians is not obvious. More empirical work is neededthat identifies the resistance-to-new-technology factorand shows that it has major productivity effects at themacro-level.15 Moreover, if technological knowledgeis global and countries differ in their resistance toadopt it, then TFP is country-specific, and bilateralor spatial characteristics should play no role for thedistribution of technological knowledge in the world.

13 It often involves noting that rich countries have higher-qualitycapital than poorer countries. Differences in technological knowl-edge are thereby subsumed into differences in the quality of capitalgoods. For example,Mankiw (1995, p. 281) argues that countriesshare the same production function, but “when an economy dou-bles its capital stock, it does not give each worker twice as manyshovels. Instead, it replaces shovels with bulldozers”. Followingthis line of argument leads to an analysis of growth from an ac-counting perspective. It is not very helpful, though, for explainingthe fundamental causes of growth and income differences. Seealso Romer (1992).14 See alsoPrescott (1998)and Easterly and Levine (2001), as

well as other contributions at the recent World Bank conferenceon cross-country growth regressions, athttp://www.worldbank.org/research/growth/regressions.htm.15 The debate surroundingOlson (1982)thesis might be instruc-

tive as well.

http://www.worldbank.org/research/growth/regressions.htm

http://www.worldbank.org/research/growth/regressions.htm


This implication conflicts with the evidence that arediscussed inSection 6.

In a set of papers, Quah has discussed character-istics that might characterize the knowledge-driven,or ‘weightless’ economy (e.g.Quah, 2001a,b). In hisframework, technological knowledge is disembod-ied, codified, and global, whereas human capital isembodied, tacit, and local knowledge. This might bea good starting point to further study internationaltechnology diffusion. Instead, Quah’s main interestis to analyze how certain infinitely expansible prod-uct innovations such as computer software, videoentertainment, and genetic databases might lead toa process of technical change that is much more in-fluenced by consumers (or, demand) than previously,and less so by entrepreneurs (or, supply).

A more substantial departure from neoclassical ex-planations on technological change and its diffusionis the evolutionary approach especially as laid outby Nelson and Winter (1982). We cannot do justicehere to the large literature in this area.16 Instead,we simply want to emphasize some points that areparticularly important in the present context. Differ-ently from the neoclassical perspectives of technol-ogy as “blueprint” or “designs” that can be tradedon markets, evolutionary theory portrays technology,or know-how to do things, as organizationally em-bedded. The process of innovation is thus based onthe joint activities of many actors within organiza-tions and environments. Country-specific factors arethought to affect the process of technological diffusionthrough various channels. The concept of a “nationalinnovation system” (NIS), developed byLundvall(1992) and Nelson (1993), is used as an analyticdevice to examine countries with a system-theoreticapproach. A basic assumption underlying the NISapproach is that innovation capabilities of firms arein a sense national, and can be enhanced by nationalpolicies.

Before turning to the discussion of empirical resultson international technology diffusion, we highlight inthe following that not all growth effects that have beendiscussed in this context have to do with internationaltechnology diffusion.

16 SeeWitt (1993), Dosi and Nelson (1994), as well asNelsonand Winter (2002)for introductions.

2.3. Growth effects that are not related tointernational technology diffusion

The degree of international economic activity, fromthe one extreme of autarky to the other of full integra-tion, can have important implications for a country’sproductivity through mechanisms other than interna-tional technology spillovers. Among the importantones are give I the following sections.17

2.3.1. Trade and domestic monopolyThe liberalization of international trade might

reduce the monopoly power of domestic firms,thereby affecting pricing behavior and the efficiencywith which domestic resources are utilized (Tybout,2000 discusses the evidence). Clearly, this effect isnot related to technological knowledge diffusion fromabroad.18

2.3.2. Trade and knowledge accumulation throughlearning-by-doing

If trade liberalization triggers changes in the domes-tic resource allocation, this might lead to changes ina country’s growth rate.Young (1991)for instancedevelops a model with two countries, north and south,that each produces a range of products. A country’sgrowth rate is determined by the prevailing levelof learning-by-doing, which is the cost-lowering ef-fect from cumulative production. The potential forlearning-by-doing is high for more recently inventedproducts while for older products, learning-by-doingeffects are exhausted. Young shows that trade liber-alization might slow down the rate of growth in thesouth relative to autarky. This happens if the north hasa comparative advantage in newly invented productsso that trade liberalization leads to a specializationin the south on goods in which learning-by-doing isexhausted.

2.3.3. Trade and endogenous technical changeGrossman and Helpman (1991, Chapter 6) show

that such results can be obtained also in models of en-

17 For simplicity, the focus here is on international trade, butanalogous arguments apply as well to FDI and other forms ofinternational economic activity.18 Instead, this mechanism is similar to the “non-technological”

and “policy” determinants of TFP that are emphasized inSolow(2001)andKlenow (2001), respectively; it is also related toParenteand Prescott (2000)main thesis.


dogenous technical change. Assume that there are twofinal goods, X and Y, produced with varying factorintensity, plus an R&D activity that produces inputsfor the X good. Productivity growth is related to Xproduction, because analogous to the static model ofEthier (1982), an increasing set of specialized inter-mediate inputs allows the production of more outputwith the same amount of inputs. In this situation,the growth effects of trade liberalization depend onwhether it raises or lowers the relative price of goodY (that is, it again depends on comparative advan-tage). If trade liberalization raises it, resources moveout of X production (and the R&D laboratories) andinto the Y sector. As a consequence, growth slowsdown relative to before. The opposite occurs if tradeliberalization lowers the relative price of good Y.

Trade liberalization might have important growtheffects through changes in the domestic resource allo-cation, although we are not aware of much evidencethat shows this to be the case (see also the overviewin Tybout, 2000, pp. 37–38). It is worth emphasizing,though, that these particular modelsdo not involvethe international diffusion of technology. This is clearfrom the fact that it is possible to find a tax and subsidypolicy that affects the domestic resource allocationin just the same way as trade liberalization does, andwhich therefore has the same growth effects.19 Thesemodels highlight the fact that there might be growtheffects from trade liberalization that are not relatedto international technology diffusion. Arguably, onemight expect the diffusion effects associated withinternational trade to be of dominant importance, atleast for less developed countries. We will discussthe available evidence later. In any case, insofar aswhat diffuses is additional knowledge, it is difficultto see how this could lower a country’s growth rateand welfare relative to autarky: access to more, ratherthan less, knowledge, just like more choices, shouldleave a country at least weakly better off.

19 The standard mechanisms of international technology diffusionare not present in either Grossman and Helpman’s or Young’smodel. In the former model, neither are the specialized intermediateinputs internationally tradable, nor are there domestic learningeffects from foreign R&D. In the latter model, each country’s stockof knowledge depends only on its range of production; there isno direct link between domestic and foreign stocks of knowledge.Moreover, the learning-by-doing stages that the north has passedthrough are not transferable: they have to be repeated by the south.

We now turn to discussing basic results on theinternational diffusion of technology.

3. International technology diffusion: basicmagnitudes

A number of different approaches have been em-ployed to study empirically the importance of inter-national technology diffusion. The first and largestset of papers consists of so-called international R&Dspillover regressions.20 This literature studies theproductivity effects of foreign R&D on domesticproductivity. It is analogous to the literature that hasexamined the effects of other firms’ R&D on a givenfirm’s productivity in a closed economy (Griliches,1979, 1995; Scherer, 1984). Typically, a productionfunction approach is used to relate TFP to measuresof domestic (R) and foreign R&D (S) activities21

ln TFPct = αc + αt + βr ln Rct + βs ln Sct + εct,

wherec indexes country andt subscripts time;αc andαt make for a generalized and time-varying intercept,andεct is an error term. The definition ofS, the termcapturing the impact of foreign R&D, is typicallygiven by a weighted sum of other countries’ R&D:

Sct =∑h�=c

ωchtSht,

whereωcht is a bilateral weight that captures the rela-tive importance of R&D in countryh for productivityin countryc. In the earlier literature on inter-industrytechnology diffusion in a given country, theω areoften input–output shares. More recent studies of in-ternational R&D spillovers, for example, byMohnen(1992)andCoe and Helpman (1995), have employedimport shares as weights. Depending on the particularchannel of technology diffusion that authors analyze,also FDI or other weights have been employed (seeSection 4).

These regressions are partial-equilibrium in nature,and the R&D expenditures, as well as the weights ù,are often endogenous. In addition, omitted variablesmight be causing biases. The work in this literature

20 SeeMohnen (2001)for a recent survey.21 For a discussion of TFP as an indicator of technical change,

seeHulten (2001).


varies in the extent to which the authors address thesepotential problems. At its best, this approach providesuseful information on the long-run average relation-ship between R&D and productivity in a reduced-formframework.

The importance of international technology dif-fusion can be calculated in a number of ways. Oneapproach is to compare the TFP elasticities of do-mestic and foreign R&D.Coe and Helpman (1995)estimate for their sample of 22 industrialized coun-tries a domestic R&D elasticity of about 8% for the 15smaller countries, and 23% for the G-7 countries.22

The corresponding elasticities with respect to foreignR&D are about 12 and 6%, respectively. Thus, for the15 smaller countries, the effect from foreign R&D islarger than that from domestic R&D, with a factor ofabout 1.5. The share of about 20% for foreign R&D inthe total elasticity effect for the G-7 countries is simi-lar toKeller (2002b)estimate in his analysis of the G-7countries plus Sweden using data at the industry-level.By contrast,Park (1995)in his analysis of aggregatedata for ten OECD countries (including the G-7) esti-mates that foreign R&D accounts for about two thirdsof the total effect of R&D on productivity (domes-tic R&D elasticity of 7%, foreign R&D elasticity of17%).

In a series of papers,Eaton and Kortum (1997,1999) and Eaton et al. (1998)have developed gen-eral equilibrium models in which productivity growthis related to increases in the quality of intermediategoods.23 As is often the case in empirical work basedon a structural model, in order to arrive at a frameworkthat can be estimated, Eaton and Kortum have to makesome strong assumptions. For instance, the quality oftechnological knowledge that is discovered in a coun-try in Eaton and Kortum (1999)is a random variablewith a Pareto distribution, while the distribution of thetime until it has diffused to other countries is expo-nential. Assumptions such as these are difficult to testin the context of a given model. Eaton and Kortum’sempirical results are thus better viewed as estimatingor simulating a particular model, rather than selectingone model among several, or testing it.

22 These are Canada, France, Germany, Italy, Japan, the UnitedKingdom, and the United States.23 This is as in the ‘quality-ladder’ model ofAghion and Howitt

(1992); see alsoHowitt (2000).

Eaton and Kortum’s work shows that the pay-offto this can be high. Their framework allows not onlystudying the reduced-form relationship between R&Dand productivity, but also the models’ predictionsfor the transitional adjustment path to the long-runequilibrium (Eaton and Kortum, 1997). They alsoexamine the effects of changes in economic policy(Eaton et al., 1998), and Eaton and Kortum (1999)model R&D and the diffusion of knowledge togetherwith the related activity of international patenting.With regard to the relative importance of technologydiffusion from abroad, the latter paper presents resultsbased on data for the G-5 countries around the year1988. Eaton and Kortum estimate that the part of pro-ductivity growth that is due to domestic as opposedto foreign R&D is between 11 and 16% in Germany,France, and the UK; it is around 35% for Japan, andabout 60% for the United States (1999, Table 5).

The recent work byKeller (2001c, 2002a)repre-sents a third approach. Unlike Eaton and Kortum’swork based on general equilibrium models, Kellerstudies the relationship between productivity and for-eign R&D in a single-equation, partial-equilibriumframework. And unlike the R&D spillovers literaturediscussed above,Keller (2001c, 2002a)estimates theTFP effect of foreign R&D jointly with the impor-tance of one or more channels of diffusion for foreignR&D. One can view this as estimating the weightsω of the foreign R&D variable together with the pa-rameter̂a that measures the TFP elasticity. Estimatingthe weights instead of assuming particular weightsthat are taken from data tables means to impose lessstructure ex-ante. Given how little is known on in-ternational technology diffusion to date, this seemsto be reasonable. Using this method,Keller (2002a)estimates that between 1983 and 1995, the contribu-tion of technology diffusion from G-5 countries is onaverage almost 90% of the total R&D effect on pro-ductivity in nine other OECD countries (Table 6).24

This shows that irrespective of which particularapproach is taken, the results invariably point toa substantial contribution of foreign technologicalactivity to domestic productivity. It also seems thatthe relative contribution of international technologydiffusion to domestic productivity growth isinversely

24 These are Australia, Canada, Denmark, Finland, Italy, TheNetherlands, Norway, Spain, and Sweden.


correlated with economicsize and with thelevel ofdevelopment.25

4. Channels of technology diffusion

Many authors have studied through which mecha-nisms, or channels, international technology diffusionprimarily occurs. We first look at work that has con-sidered international trade.

4.1. Trade

Coe and Helpman (1995)are the first to provideevidence on the importance of trade for internationaltechnology diffusion along the lines of recent theory.Using the spillover regressions framework shown onp. 17 with bilateral import shares as weights, theseauthors examine two predictions. First,Coe and Help-man (1995)study whether a country’s productivity isincreasing in the extent to which it imports from high-as opposed to low-knowledge countries (an importcomposition effect). Second, for a given compositionof imports, these authors analyze whether a country’sproductivity is higher, the higher is its overall importshare. Coe and Helpman’s regression results suggestthat there is support for both predictions. The authorsconclude that international R&D spillovers are relatedto the composition of imports, and that the overall levelof imports is important for international technologydiffusion as well.Coe et al. (1997)find similar effectsin their analysis of foreign technology diffusion fromhighly industrialized to 77 less developed countries.26

By contrast,Eaton and Kortum (1996)find thatonce distance and other influences are controlledfor, bilateral imports do not help to predict bilateralpatenting activity, the indicator of international tech-nology diffusion in their model. Moreover,Keller(1998) repeats theCoe and Helpman (1995)regres-

25 This would be very important for less developed countries.Primarily due to the much greater difficulty in obtaining com-parable and high-quality data for less developed countries, thereis a relative paucity of results in this regard. See, however,Coeet al. (1997), Connolly (1998), andMayer (2001), as well as thediscussion later.26 Nadiri and Kim (1996)show results from estimating cost

functions in OECD countries; their approach also uses thebilateral-imports weighted spillover variable.

sions with counterfactual ‘import’ shares.27 For thereto be strong evidence for trade-related internationalR&D spillovers, he argues, one should estimate apositive effect from foreign R&D when bilateral im-port shares are employed, butno strong effect whencounterfactual ‘import’ shares are used. Keller findssimilarly high coefficients and levels of explainedvariation when counterfactual instead of actual importshares are used. He concludes that, on the basis oftheir analysis, Coe and Helpman’s claim cannot beupheld: the import composition of a country has nota strong influence on the regression results.28 Keller(1998)results have led some to doubt the importanceof trade for international technology diffusion.

More recent work has strengthened the evidencefor import-related international technology diffusion.Keller (1997b, 2000)has extended his analysis of esti-mating international R&D spillovers with counterfac-tual data by moving to industry-level data for eight in-dustrialized countries. His results suggest that importcomposition might not matter for technology diffusionif countries’ import patterns are more or less symmet-ric, but they do matter if countries receive a relativelyhigh share of its total imports from one particularcountry—such as is the case for Canada, for example,which imports about 80% from the United States.29

27 That is, instead of bilateral import shares,Keller (1998)usescounterfactual (or, made-up) shares in creating the foreign R&DvariableS (see p. 16).28 Keller (1997a)presents analogous results in the context of

inter-industry technology flows.29 As Coe and Hoffmaister (1999)emphasize,Keller (1998)cre-

ates his counterfactual ‘import’ shares in a way that makes theirvariance across different simulations rather small, and values higherthan 0.3 are essentially never obtained. Thus, his import sharesare not fully random in that sense. This was first pointed out in1996 by an anonymous referee at the European Economic Reviewto both Keller and the editor, Elhanan Helpman.Keller (1998)paper includes therefore results with the unweighted sum of for-eign R&D as well. These do not support theCoe and Helpman(1995) claim that international R&D spillovers are related to thecomposition of imports either, which isKeller (1998)point (seehis Table 2). In a different context,Keller (1997b, 2000)showsthat certain randomizations of import shares lead to equally strongempirical results as those based on observed imports, while othersdo not. He also discusses the relationship between random importshares and unweighted sum-of-foreign-R&D regressions.Coe andHoffmaister (1999)show that for three types of randomizations ofimport shares, theCoe and Helpman (1995)framework doesnotlead to a finding of international R&D spillovers. This means thatthe bilateral import shares have some power.


Moreover, Xu and Wang (1999)show that theimport composition effect is robust when one consid-ers capital goods trade instead of all-manufacturinggoods trade. Xu and Wang show that if the weightsin the construction of the foreign R&D variableSare capital goods import shares, one obtains anR2 of77.1% versus 74.9% withKeller (1998)counterfac-tual shares, and 70.9% inCoe and Helpman (1995).Also Lumenga-Neso et al. (2001)have revisitedthe results ofCoe and Helpman (1995)and Keller(1998). Instead of computing the foreign knowledgevariable as a bilateral-import share weighted sum offoreign R&D, Lumenga-Neso, Olarreaga, and Schiffconstruct an alternative variable that aims at takingaccount of previous rounds of imports as well. Thiscaptures the case that even if some countryc importsonly from some other countryh, for instance, theformer might still gain access to technology fromcountries other thanh—if country h has in turn im-ported from those other countries before. Using thisvariable, Lumenga-Neso, Olarreaga, and Schiff findslightly stronger results thanKeller (1998).

Other evidence on the importance of trade for in-ternational technology diffusion includesSjöholm(1996), who examines citations in patent applicationsof Swedish firms to patents owned by inventors inother countries. Patent citations as indicator for knowl-edge flows have proven to be useful recently in a num-ber of other studies (seeSections 6 and 9). Controllingfor a number of other correlates and also conductingan extreme-bounds analysis, Sjöholm finds a positivecorrelation between Swedish patent citations and bi-lateral imports. This result is consistent with importscontributing to international knowledge spillovers.

Two recent papers byEaton and Kortum (2001a,b)also focus on the importance of trade for the interna-tional diffusion of technology. The authors combinethe structure of technology diffusion and growth inEaton and Kortum (1999)with that of the Ricar-dian trade model due toDornbusch et al. (1977). Acountry’s productivity level is related to its implicitaccess to foreign technology through equipment goodimports. Thus, their notion of technology diffusionis one of passive (embodied) technology diffusion.Transport costs that increase in geographic distanceimply that productivity in remote countries is rela-tively low, ceteris paribus, or equivalently, the priceof equipment goods is relatively high.Eaton and

Kortum (2001a)use this framework to predict, forexample, that 25% of the cross-country productivitydifferences among 34 more-, as well as less-developedcountries can be attributed to differences in the rel-ative price of equipment. It would be interesting tosee how international trade might at the same time in-crease the likelihood of active knowledge spillovers.This would also allow the estimation of the rela-tive importance of active and passive spillovers forinternational technology diffusion.

The work discussed so far focuses on technologydiffusion related to imports. Learning through export-ing, however, might be important as well (e.g.Rheeet al., 1984). A number of recent studies, includ-ing Bernard and Jensen (1999)—using US data—andClerides et al. (1998)—using data from Colombia,Mexico, and Morocco—have used micro data to seewhether there is evidence for learning-through-expor-ting. While exporting firms tend to be more produc-tive than non-exporters in the cross-section, neitherpaper finds robust evidence that past exporting experi-ence (Granger-) causes improvements in performanceonce other differences across firms have been takeninto account. Thus, the available evidence does notpoint to strong learning-through-exporting effects.30

4.2. Other mechanisms

4.2.1. Foreign direct investmentIt is often argued that foreign direct investment

(FDI) involves the transfer of knowledge from onecountry to another (see, e.g. the discussion inCarret al., 2001), making it a potentially important vehiclefor international technology diffusion.31

Lichtenberg and van Pottelsberghe de la Potterie(1996) have analyzed the importance of FDI for in-ternational technology diffusion in thirteen OECDcountries with the same R&D weighting approachthat Coe and Helpman (1995)andKeller (1998)usefor imports.Lichtenberg and van Pottelsberghe de laPotterie (1996)do not find significant effects frominward FDI. Baldwin et al. (1999)find some positiveinward FDI spillover effects in their industry-levelstudy, but overall, the results are mixed. To some

30 For more details, seeTybout (2001).31 Blomström and Kokko (1998)have recently surveyed a number

of ways how FDI could lead to spillovers.


extent, this might have to do with the fact that gooddata on bilateral FDI is available to a much lesserextent than for trade. Therefore, the two above stud-ies employ data on FDI stocks that is derived frombalance-of-payments accounts. However, good mea-sures of multinational activity, such as subsidiaryGDP, sales, or its number of employees, are relatedto FDI stocks derived from balance-of-payment dataonly with a considerable amount of error.32

Xu (2000)uses the Bureau of Economic Analysis’sdata on US multinational enterprise (MNE) activity tostudy the relationship between US outward FDI andproductivity growth in the host country. His analysiscovers total manufacturing between 1966 and 1994for 40 countries, of which about half are more andhalf are less developed countries. Xu finds generally apositive correlation between productivity growth andthe ratio of subsidiary value added to host countryGDP. This effect is stronger and more robust in themore developed countries than in the less developedcountries.

As the availability of micro data has become greater,the study of FDI spillovers has increasingly turnedto it. Aitken and Harrison (1999)examine data onVenezuelan plants between 1976 and 1989. They esti-mate anegative relationship of increased FDI presenceand TFP of domestic plants. This can be rationalized ifthe estimates pick up a combination of the knowledgespillover effect (which is non-negative) and competi-tiveness or other effects. It could also arise from multi-nationals being attracted to industries where domesticcompetitors are structurally weak, if the estimationfails to fully control for this simultaneity effect. Inany case, the effects could be short-term responses;as Aitken and Harrison note, their analysis may failto capture the long-run effects of FDI (p. 617).33

32 One important reason for this is that FDI does not necessarilyinvolve net flows of capital, because the latter can be raised inlocal capital markets.33 FDI presence inAitken and Harrison (1999)is measured as the

share of employment of foreign-owned firms in total employment.This variable can change without a change in employment byforeign-owned firms. Aitken and Harrison report, however, that in-cluding foreign employment and domestic employment separatelyin the regression leads to similar results (p. 610). This point ap-plies also to the studies byGirma and Wakelin (2001)andHaskelet al. (2001), discussed later. We have not seen a discussion ofthis point in the latter papers, so we assume that it does not alterthe results there either.

Girma and Wakelin (2001)use micro-data from theUK Census of Production to study the effects of in-ward FDI in the United Kingdom. The authors focuson the electronics industry between 1980 and 1992.Girma and Wakelin (2001)test whether plant produc-tivity growth is systematically correlated to the ratio offoreign-owned plants to all plants in a given four-digitindustry (and also by foreign investor country). Othervariables are FDI in a given geographic region withinthe UK, and FDI in a given two-digit industry. The firsttests for intra-industry FDI spillovers, the second andthird for FDI spillovers from geographic and techno-logical proximity, respectively. The authors find a pos-itive productivity effect from Japanese and Other for-eign FDI, but not from US FDI. There is also evidencein support of FDI spillovers from technological prox-imity, but no evidence that within-region spillovers arestronger than those across regions.34

Haskel et al. (2001)use the same database, but addboth more years and plants from all of manufacturingto the sample. They confirm theGirma and Wakelin(2001)findings that technological proximity seems tomatter for FDI spillovers, whereas geographic prox-imity within the UK does not.Haskel et al. (2001)estimate positive spillovers from US and French FDI,whereas Japanese FDI is here negatively correlatedwith productivity growth.35 Even though they find

34 Jaffe et al. (1993)as well asAdams and Jaffe (1996)findevidence that distance matters for domestic technology diffusion.To be consistent with that, Girma and Wakelin’s result might haveto be interpreted as saying thatconditional on technology comingfrom abroad, geographic distance within a country does not matter.See alsoSection 6.35 The differences in results relative toGirma and Wakelin (2001)

could be in part related to differences in the estimation technique.Girma and Wakelin favor a variant of theOlley and Pakes (1996)technique, whereasHaskel et al. (2001)employ a time differencingapproach. The former method is conceptually more appealing,but it also requires making more specific assumptions that aredifficult to test. Based on supplemental results for the Girma andWakelin study using the time differencing method—provided to usby Sourafel Girma— one can compare their FDI spillover resultsusing the modified Olley and Pakes technique with those based ontime differencing. At a 5% significance level, there are qualitativedifferences in 3 out of 10 of the key estimated coefficients intheir Table 7. This does not settle the question of which estimatesare closer to the true parameters. However, it suggests that toexamine whether some of the assumptions in the Olley and Pakesframework can be tested or generalized is an interesting researchproject.


partial evidence for statistically significant positivespillovers, Haskel, Pereira, and Slaughter’s analysis ofthe economic benefit of these FDI spillovers for UKfirms suggests that it was quite small compared to thesubsidies that the UK government paid to attract theseforeign-owned multinational firms.

Kinoshita (2000)presents firm-level evidence on theeffects of inward FDI in the Czech Republic between1995 and 1998.36 Unlike the other studies, Kinoshitahas information on R&D expenditures of the domesticfirms, which might be an important omitted variable.At the same time, she has only a very short panel ofdata. In any case, controlling for R&D, Kinoshita failsto find positive spillovers from inward FDI into theCzech Republic; however, there is a robust effect if theFDI variable is interacted with the firm’s own R&Dspending. A plausible interpretation of this is that theeffects of international technology diffusion throughFDI are conditional on a relatively high ‘absorptivecapacity’ (Cohen and Levinthal, 1989), as measuredby own R&D investments.37

Globerman et al. (2000)analyze 220 patent applica-tions by Swedish MNEs and non-MNEs in the year of1986. These applications contain patent citations thatthe authors relate to both inward FDIfrom the citedcountries as well as to outward FDIto the cited coun-tries. Using a conditional logit estimation framework,Globerman et al. (2000)estimate a robust correla-tion between outward FDI and patent citations—not only for MNEs, but also for non-MNEs—whereas there is none on the inward side.Branstetter(2001a) analyzes a larger number of patent appli-cations and patent citations between the US andJapan.38 While Branstetter (2001a)can measure FDIonly by establishment counts, he can control for thefirms’ R&D spending, which is a plus (see above).

36 The data comes from two surveys of the Czech StatisticalOffice. Because there is relatively little documentation on these sofar, some caution is required in interpreting what follows.37 Haskel et al. (2001)find the FDI effects to be strongest for

plants at the low- to medium-range of the distribution of skillintensities (skilled to unskilled workers); this seems to be not theplants with the highest absorptive capacity.Kinoshita (2000)alsodistinguishes FDI spillover effects for domestically-owned fromthose for foreign-owned firms, finding that only the former benefit.It would be interesting to see whether this also holds in the UKsample.38 This is based on data for Japanese-owned FDI in the United

States.

The data also allows him to distinguish product de-velopment subsidiaries—which presumably are rela-tively knowledge-intensive—from other subsidiaries.Branstetter finds that more FDI is associated withmore patent citations, both from US firms to Japanesefirms and vice versa.

4.3. Communication patterns

Some evidence has begun to emerge on the impor-tance of communication for international technologydiffusion. Recent innovations have improved thecommunication and monitoring abilities between geo-graphically distant plants, which has made it easier tooutsource certain stages of production. This suggeststhat communication between geographically distantpersons might also play a more important role todaythan it used to. At the same time, this does not nec-essarily mean that person-to-person communicationpatterns have become aspatial, for at least two rea-sons. First, remote communication might only be animperfect substitute for face-to-face communication,and this point is reinforced if knowledge is to someextent tacit. Second, communication is to some ex-tent spatial as long as international trade is, becausetrade requires some communication to develop in thefirst place. Moreover, more communication is also theconsequence of international trade. Analogous argu-ments apply to FDI and other international economicactivities.

Keller (2001c, 2002a)uses the imperfect proxyof bilateral language skills in the population at largeto examine the importance of communication pat-terns for international technology diffusion amongOECD countries. His results suggest that differencesin communication patterns might have quite strongeffects. For instance, if the share of English speakersin Spain (at 17%) would rise to the level of the shareof English speakers in The Netherlands (at 77%),Keller (2002a)estimates that for Spain, this wouldbe equivalent to a 15% boost of technology diffusionfrom English-speaking countries.39

39 A number of other channels of technology diffusion have beenemphasized, but we are not aware of much recent empirical workalong these lines. According to one argument, a country’s out-ward FDI might facilitate technology learning in the FDI sendingcountry. For instance, the FDI by Asian companies in the Silicon


4.4. Simultaneous analysis of multiple-channels oftechnology diffusion

Although a few recent studies have considered sev-eral channels of technology diffusion simultaneously,the majority of them still examines only one channelin isolation.40 As documented byKraay et al. (2001),firms do not typically engage in one type of interna-tional activity by itself; rather, they are engaged inseveral of them. Kraay, Isoalaga, and Tybout use infor-mation on whether a firm is (i) importing intermediategoods, (ii) exporting, or (iii) foreign-owned, as wellas combinations of these. As inClerides et al. (1998),the data comes from three panels of firms in Morocco,Mexico, and Colombia. Kraay, Isoalaga, and Tyboutfind no significant effects of international economicinteractions on the firm’s marginal cost or the qual-ity of the good that it produces. By contrast,Keller(2001c)’s industry-level analysis of spillovers amongthe G-7 countries finds significant effects for imports,inward FDI, as well as communication links. In abreakdown of the total effect, he attributes more than50% of the total effect to imports, and the remainderto equal parts to FDI and communication links.

5. Heterogeneity and inter-industry effects

R&D spending is highly concentrated by indus-try. Among OECD countries, for instance,Keller(2002a)reports that about 80% of all manufacturingR&D is conducted in four three-digit ISIC indus-tries: chemical products (including drugs), electricaland non-electrical machinery (including computersand telecommunication equipment), and transporta-

Valley area is sometimes characterized as being motivated by thedesire to learn early about new technological developments, andthis learning is apparently facilitated by geographical proximity.The result byLichtenberg and van Pottelsberghe de la Potterie(1996)who find that a country’soutward FDI gives access to for-eign technology is consistent with that. The importance of returnmigration of foreign-trained students and engineers has also beenemphasized in explaining the relatively fast acquisition of techno-logical capability of countries like South Korea (Westphal et al.,1985), but we are not aware of more recent empirical work alongthese lines.40 See, however,Globerman et al. (2000)and Lichtenberg and

van Pottelsberghe de la Potterie (1996), who both analyze tradeand FDI.

tion equipment. Because it is a priori plausible thatinternational technology diffusion is most importantin industries that account for a substantial part ofall R&D, some authors have restricted their analysisright from the start to such ‘high-technology’ indus-tries (e.g.Bernstein and Mohnen, 1998). Other au-thors have tried to compare the magnitude of effectsacross industries. These studies fall primarily into thefollowing two broad categories.

A first set of papers, includingCoe et al. (1997)and Mayer (2001), is based on studies that explorethe importance of international trade as a mecha-nism of international technology diffusion (see alsoXu and Wang, 1999, discussed inSection 4). Us-ing foreign knowledge stocks that are constructedas import-weighted sums of R&D in industrializedcountries,Coe et al. (1997)find stronger and morerobust evidence for north–south spillovers using ma-chinery and equipment import data (SITC class 7)than employing either all-manufacturing or total im-port data.41 Mayer (2001)argues that this includesstill many consumption and equipment goods that areunlikely to lead to much technology diffusion. Hisfocus on machinery imports is confirmed in the sensethat the machinery-imports based variable enters witha coefficient that is twice as large as in the corre-sponding regression using the variable based on allSITC 7 imports (Mayer, 2001, Tables 6 and 8).42

Another way to examine heterogeneity is to seewhether the estimated spillover effects vary in magni-tude across industries.Keller (2001c, 2002a)analyzespanels of twelve manufacturing industries. Eight ofthese twelve industries account for only about 20%of all manufacturing R&D, whereas—as mentionedearlier—the remaining four industries make up forthe remaining 80%. Keller shows separate regressionresults for the sample of low-R&D industries, estimat-ing that the elasticity of TFP with respect to R&D isonly about 70% of the average elasticity for all twelveindustries. In his study of trade and technology diffu-sion in Latin America using the spillover regressionsframework from above,Blyde (2001)finds stronger

41 Connolly (1998)in her cross-country growth analysis of inno-vation and imitation uses the SITC class 7 data as well.42 For similar reasons,Keller (1997b, 2000)uses disaggregated

data as well. He relates data on specialized machinery imports—atthe three- and four-digit SITC level—to productivity at the two-and three-digit ISIC class level.


effects from OECD imports than from Latin Amer-ican imports. He attributes this result primarily tothe fact that OECD imports have a higher embodiedtechnology content than Latin American imports.

These results are consistent with international tech-nology spillovers varying greatly in strength acrossdifferent set of goods. Another literature has startedto assess the importance of international technologydiffusion from one set of products to another. In par-ticular, some authors have suggested that advances inICTs might have strong productivity effects in otherindustries, perhaps even abroad.Gera et al. (1999)estimates of technology spillovers from the US toCanada suggest that technology spillovers embodiedin ICT imports from the US have about four times theeffect on labor productivity in Canadian industries asspillovers embodied in non-ICT imports.

Keller (2002b) analysis provides an estimate ofproductivity effects from four different sources ofR&D: domestic R&D as well as foreign R&D, bothin the same industry as well as in other industries.Using an input–output matrix to model inter-industrytechnology flows, his findings suggest that the impor-tance of foreign R&D outside the industry itself is asimportant for productivity as the foreign R&D withinthe industry.43

6. Geographic localization of internationaltechnology diffusion

It was noted earlier that income convergencedepends on whether technology spillovers are localor global, with strong international diffusion favoringconvergence and geographically localized diffusionmaking divergence more likely. Thus, a number of au-thors, includingJaffe et al. (1993), Irwin and Klenow(1994), Jaffe and Trajtenberg (1998), Eaton andKortum (1999), and Branstetter (2001b), provideresults on whether technology diffusion within acountry is stronger than across countries.44

43 See also the evidence on spillovers in technologically proximatefields in the section on FDI above.44 In addition to the geographic dimension of technology diffu-

sion, one might also analyze its temporal and sectoral dimension.In particular, one might ask how long it takes until technology hasdiffused from a given country to another country. Such diffusiontimes might systematically vary with level of development and

Using patent citations as their measure of knowl-edge flows,Jaffe et al. (1993)compare the geographiclocation of patent citations with that of the citedpatents in the United States. Their key finding isthat US patents are significantly more often citedby other US patents than they are cited by foreignpatents.Branstetter (2001b)uses R&D and patentingdata on US and Japanese firms to compute weightedR&D spillover stocks analogous toCoe and Helpman(1995) bilateral import share weights.45 Branstetterfinds that within-country spillovers are much strongerthan between-country spillovers. The work on patent-ing and international technology diffusion byEatonand Kortum (1999)andJaffe and Trajtenberg (1998)supports this result as well.

However, Irwin and Klenow (1994) study oflearning-by-doing spillovers in the semiconductor in-dustry finds the opposite.46 These authors estimatethat for eight vintages of semiconductors introducedbetween 1974 and 1992, the learning spillovers fromone US firm to another US firm are not signifi-cantly stronger than those between an US firm anda foreign firm. The different results might be ob-tained because Irwin and Klenow’s learning-by-doingspillovers, which are identified from the effects ofcumulative production on market shares, are differentfrom knowledge spillovers as measured in the otherstudies.47 It could also be due to the market structureof the semiconductor industry, which had only rela-tively few firms that were located primarily in the USand in Japan during this period.

In order to make further progress, it might thereforebe necessary to study geographic effects in technol-

other covariates. Such an analysis will likely be complementaryto studying the geographic dimension of technology diffusion ata given point in time, on which we focus here.45 Patenting data by technological field allows Branstetter to com-

pute weights that are increasing in the similarity of two firms’patenting activities; this captures the idea that R&D expendituresof another firm is more likely to generate spillovers, the closerthe two firms are in technology space (seeJaffe, 1986).46 Learning-by-doing spillovers are most closely related to hu-

man capital models (e.g.Lucas, 1988, 1993), not to models ofendogenous technological change. However, it is not clear thatempirical analysis has been able to separate one from the otherso far. We thus include the following discussion.47 In particular, the effects captured by Irwin and Klenow are

much broader than the notion of knowledge flows as evidencedby patent citations.


ogy diffusion beyond the basic “border effect”. Themost direct approach is to analyze international tech-nology spillovers conditional on geographic distanceand the location of countriesrelative to each other.By estimating whether there is geographic decay inthe strength of technology spillovers,Keller (2001c,2002a)has done this for OECD countries. For ex-ample, if geographic proximity matters for benefitingfrom international technology diffusion, then technol-ogy created in the US should have a stronger influenceon productivity in Canada than in Germany; and thelatter effect should in turn be stronger than the effectin Australia, ceteris paribus.

Keller (2002a)uses industry-level data from 14OECD countries between 1970 and 1995. He relatesproductivity in nine mostly smaller OECD countriesto R&D conducted in the G-5 countries, and allowsfor the possibility that the strength of the R&D effectsare conditional on a country’s bilateral distance tothe G-5 countries. He estimates a simple exponentialdecay function:

ln TFPcit = β ln

Rcit + γ

∑h�=c

Rhit e−δDch

+ α′X + εcit.

Here,Rh is R&D in the G-5 countryh, andDch is thegeographic distance between countriesc and h. ThevectorX is a set of other variables that help to con-trol for the effects of simultaneity. Of key interest inthis framework is the parameterδ: for δ > 0, variationin productivity levels is best accounted for by givinga lower weight to R&D conducted in countries thatare located relatively far away, whereas ifδ = 0, thengeographic distance and relative location do not mat-ter.Keller (2002a)estimatesδ to be positive, which isconsistent with the geographic localization of interna-tional technology diffusion.48

What is the magnitude of this effect? The expo-nential specification allows to compute the ‘half-lifeof technology’ in terms of distance—the distanceafter which half of the technological knowledge thatoriginates from a technology sending country has dis-appeared.Keller (2002a)estimates it to be only about

48 See alsoBottazzi and Peri (1999)analysis of geographic effectsin the diffusion of technology among European regions.

1200 km. This corresponds to a very rapid geographicdecay. If literally true, it implies that for instanceAustralia, with its remote geographic location relativeto the G-5 countries, benefits extremely little from in-ternational technology diffusion. Instead of analyzingtechnology diffusion from the G-5 to mostly smallerOECD countries,Keller (2001c)studies technologydiffusion among the G-7 countries, which accountfor about 90% of the world’s R&D spending. Usingsimilar methods, he confirms the finding of localiza-tion of technology diffusion, estimating a half-life oftechnology between 800 and 1900 km.

From an economic viewpoint, the finding that geo-graphic distance matters is only the first step, ofcourse. The question is, why does distance matter?At a prescriptive level, it is clearly impossible todesign economic policies that “move Australia geo-graphically closer to the G-5 countries”, at least in aliteral sense. Thus, it is important to see whether thegeography effect can be related to the economic in-teractions across countries, including those discussedin Section 4.

Keller (2001c)considers the technology diffusionchannels of imports, FDI, and communication. Heintroduces variables that measure the strength of eachof the respective bilateral channels in analogy to geo-graphic distance in the equation on p. 38. According tohis analysis, each channel individually has a positiveeffect on international technology diffusion. Whenthe channels of imports, FDI, and communication areconsidered together with distance, Keller doesnot es-timate a geographic localization effect anymore. Thissuggests that a substantial portion of the distance ef-fect in technology diffusion, and may be all of it, canbe accounted for by differences in imports, FDI, andcommunication links across countries.49

7. Changes in the scope and strength ofdiffusion over time

Today’s level of economic integration in the worldis high by historical standards (e.g.Feenstra, 1995).

49 See alsoSjöholm (1996)who finds that bilateral geographicdistance between Sweden and other countries has no robust effecton the number of citations in Swedish patent applications, oncedifferences in bilateral international trade have been controlled for.


International trade has been growing faster than in-come, and transport costs for goods have been falling.Multinational activity has been growing faster thantrade, spurred in part by the development of newcommunication technologies. And the Internet isconsidered by many to be the primary symbol of aglobally integrated world.

What does this mean for technological diffusion?Is there now a common pool of global technology,and distance is, as some have suggested, “dead”?The recent papers byKeller (2001c, 2002a)providesome initial evidence on this. He estimates whetherthe parameter that captures localization in his spec-ification has changed over time. Keller finds that theabsolute value of the distance parameter has fallensubstantially between the mid-1970s and the 1990s.This is consistent with a strong decline over timein the degree to which technology is geographicallylocalized. It constitutes some initial quantitative evi-dence that technological knowledge has become lesscountry-specific recently.

8. Major determinants of successfulinternational technology diffusion

We have noted earlier that technology diffu-sion in form of spillovers means that it involvesexternalities—economic relations that are not coveredby market transactions. As such, spillovers are relatedto the much larger topic of market and institutionalfailures in economic development in general. Thisliterature is vast, and we will not attempt to provide acomprehensive survey.50 In the specific case of tech-nology transfer through spillovers though, there arebig differences in the degree of success that countrieshave in benefiting from foreign technology. Anotherstrand of the literature has therefore asked what themajor determinants of successful technology diffusionfrom abroad are. Differences in the degree of successof technology adoption could mean that income vari-ation in the world is either increasing (divergence)or decreasing (convergence) over time. There aresome preliminary findings suggesting that betweenthe years 1983 and 1995, technology diffusion fromthe G-7 countries had stronger effects on growth in

50 For this, seeHandbook of Development Economics (1995).

the relatively rich than in the poorer countries (Keller,2001b). This might in part explain the trend towardsincome divergence for the world as a whole duringthis period.

Two determinants of successful technology diffu-sion that have been emphasized are human capital(Nelson and Phelps, 1966) and R&D (Cohen andLevinthal, 1989). Both are associated with the notionof absorptive capacity, the idea that a firm or coun-try needs to have a certain type of skill in order tobe able to successfully adopt foreign technologicalknowledge.51 This can be, first of all, in form of sci-ence and engineering human capital. It can also be inform of another factor, first emphasized byCohen andLevinthal (1989): a firm might need to invest itselfinto R&D. These authors argue that this is criticalto enabling the firm to understand and evaluate newtechnological trends and innovations, which is a nec-essary condition for successful technology adoption.

Along these lines,Eaton and Kortum (1996)findthat inward technology diffusion as measured by in-ternational patenting is increasing in the level of acountry’s human capital.Xu (2000)provides evidencesuggesting that the reason why relatively rich countriesbenefit from hosting US multinational subsidiaries,while poorer countries do not as much has to do witha threshold level of human capital in the host country.Caselli and John Coleman (2001)use data on importsof office, computing, and accounting machinery asa measure of inward technology diffusion, arguingthat many countries do not have a domestic computerindustry, so that computer technology comes neces-sarily from abroad. They find that computer importsare positively correlated with measures of human cap-ital, which is consistent with the results byEaton andKortum (1996)andXu (2000). In these three papers,the measures of human capital have only a quantitydimension, the average number of years of schoolingor school enrollment in the population.Hanushek andKimko (2000) have recently argued that the qualitydimension of human capital is at least as important.Their results using standardized science and engineer-ing test scores for around thirty countries seems tobear this out. I expect that as test score data becomesavailable for a larger set of less developed countries,

51 Keller (1996) presents a model of absorptive capacity in agrowth context.


it will be able to show that science and engineer-ing skills are important in facilitating technologydiffusion.

Griffith et al. (2000a,b)have recently emphasizedthe importance of indigenous R&D to facilitate tech-nology diffusion from abroad (see alsoKinoshita,2000, discussed inSection 4). Griffith et al. (2000a)use industry-level data from 12 OECD countries forthe years 1974 to 1990 to study what are the maindeterminants of productivity dynamics in this sample.The authors construct a productivity gap measure,defined as TFP relative to TFP in the leader country;this is a measure of the potential for productivitycatch-up.Griffith et al. (2000a)find that subsequentTFP growth is negatively related to the productivitygap, which is consistent with productivity conver-gence. If the productivity gap variable is interactedwith R&D, the authors estimate a strong negativecoefficient. This is consistent with absorptive ca-pacity being empirically important, because it saysthat catch-up is particularly rapid if there are sub-stantial R&D investments in low-productivity indus-tries.52

We now turn to a summary and some concludingremarks.

9. Summary and implications

It is evident from our discussion that the literaturethat empirically quantifies the effects of internationaltechnology diffusion is a fairly recent one. While weknow today much more than we did merely a coupleof years ago, there are only few results that are solidlyestablished at this point. To date, different strands ofresearch on a particular topic give seemingly or actu-ally conflicting answers; some of the results that havebeen reviewed above might still be overturned, andmany important topics have not been considered yetat all. Nevertheless, we will try to draw some generallessons from what we know right now.

In general, there is a substantial amount of evidencethat underlines the importance of international tech-nology diffusion for growth and development. Onceauthors allow for technology to be either of domes-

52 These authors also find that interacting human capital with theproductivity gap leads to a smaller effect.

tic or international origin, studies do find internationalsources to be important.53 The evidence so far sug-gests that the relative importance of foreign sourcesfor productivity growth is decreasing in size and levelof development. It is probably no exaggeration to be-lieve that the relative importance of foreign technologyin most less developed countries is at least 90%, andprobably higher. At the same time, there is no indi-cation that international learning is inevitable or auto-matic, merely a function of whatGerschenkron (1962)has called economic backwardness. Instead, differen-tial learning effects seem to be in part explained by theextent and the way in which firms of different coun-tries engage in international economic activities. Whatis the evidence on this so far?

As we previously argued, foreign technologyspillovers can be of two different types: activespillovers and passive spillovers. Active spilloversinvolve the direct process of learning about foreigntechnological knowledge, which can be either in cod-ified or in non-codified form. Passive spillovers areresulting from employing intermediate goods thatembody technological knowledge invented abroad.Active spillovers involve domestic agents learningabout foreign knowledge to the extent that they canuse, modify, and improve upon the technology. Pas-sive spillovers do not lead to those abilities, implyingthat an effective technology gap remains. Becausepassive spillovers are associated with intermediategoods, the patterns of passive spillovers are likelyfollowing international transactions that involve suchgoods. Active spillovers are not associated with tech-nology embodied in goods, but this does not meanthat the patterns of active spillovers are arbitrary, orthat they are universal. In particular, the importance ofnon-codified and tacit elements of technology means

53 Does this matter relative to analyses that emphasize differencesin the endowment with quality-adjusted capital (Mankiw, 1995)or in policies towards technology adoption (Parente and Prescott,2000)? It is true that at a given point in time, the process of inter-national technology diffusion must manifest itself in cross-countrytechnology differences, which translate into income differences.That all approaches will in the end lead to some statistics in thenational income accounts is not surprising—after all, each of themtries to explain GDP. The important questions are rather: whichframework is helpful for thinking about productivity growth, andwhich is useful to guide economic policy towards raising produc-tivity growth.


in practice that the acquisition of active spilloversoften involves face-to-face contacts.

There is evidence showing that the degree of com-moditization of goods and services is increasing,which is in part related to the development of so-calledInternationally Integrated Production Systems. Iftechnology can be increasingly commoditized aswell, to codified and embodied knowledge, this couldmean that face-to-face contacts become less impor-tant. Similarly, it would mean that passive spilloversmight gain in relative importance at the expense ofactive spillovers. This could also mean an increasingconcentration of fundamental technological expertise(of the active type) in a small number of countries ofthe world.

International trade has been a suggested as a ma-jor channel for technology diffusion early on. Therefinements of R&D spillover regressions in responseto the papers byCoe and Helpman (1995)andKeller(1998)now point to a robust link between imports andtechnology diffusion. This evidence is strengthenedby results from relating patent citations to importpatterns as well as by estimating general equilibriummodels of embodied technology diffusion. While theevidence using micro-data is much weaker, especiallyon learning effects from exporting, my sense is thatat least qualitatively, trade facilitates internationaltechnology diffusion. Some results, for instance byLumenga-Neso et al. (2001), indicate that there mightbe much to be learned from future work that studiesthe full dynamics of trade and technology diffusioninstead of only the long-run stationary relationships.

On FDI as a mechanism of technology diffusion,the results so far are mixed. The result that FDIdoes not necessarily lead to strong positive spilloversis a sensible one. After all, a cornerstone of thetheory of FDI says that firms choose to operatethrough a fully-owned subsidiary instead of throughjoint ventures or technology licensing because FDIhelps to keep the private return to technology invest-mentsinternal to the firm—that is, no leaking out ofknowledge.

In general, the evidence for positive effects from in-ward FDI is stronger for more developed than for lessdeveloped countries. This might have to do with thefact that outsourcing of relatively low-skill activities ismore likely for north–south FDI than for north–northFDI (e.g.Hanson et al., 2001). The former could have

a lower learning potential than the latter, not nec-essarily because the activities are different as such,but because they are integrated with the host countryeconomy to different degrees (in terms of backwardand forward linkages).54 At any rate, the fact that theonly study of those discussed above finding positiveeffects from inward FDI in a non-OECD country isone where firm-level data on R&D is available—Kinoshita (2000)—suggests that local effort in adopt-ing advanced foreign technology is very important.

Other mechanisms such as person-to-person com-munication have in our view not been analyzed to asufficient degree yet. In part, this seems to have todo with a paucity of relevant data. This channel isalso difficult to separate from trade and FDI, becausethe ease of communication is affecting the likelihoodof trade relations, for example. To some extent, therelatively large contribution of trade and the smallercontribution of communication estimated inKeller(2001c)might simply reflect that trade is a more prox-imate correlate of international technology diffusionthan communication is.

The literature on technology spillovers has movedincreasingly away from single-channel analyses tomultiple-channel analyses, or at least to work thatsimultaneously controls for overall (not-further-specified-) spillovers. To analyze multiple-channelsespecially in the context of a fully-specified generalequilibrium model is not easy, but the returns fromdoing this seem to be high.

There are a number of results suggesting that thestrength of technology diffusion across different setsof products varies: for certain types of high-tech prod-ucts, international technology diffusion could easilybe two or three times as strong as for the average man-ufacturing good, ceteris paribus. However, it is alsotrue that price deflators for R&D-intensive productsare, due to fast-rising product quality, particularly hardto calculate. While it is plausible that active and pas-sive spillovers associated with high-tech products arelarger than those associated with lower-tech products,a part of the estimated difference might be spurious.

The question to what extent foreign technologydiffuses throughout the domestic economy, or affects

54 The processing plants along the US–Mexican border(maquiladoras) might be a good example of FDI with relativelyfew backward and forward linkages.


only a small segment of it is a subject that deservesfurther analysis. As the recent debate on whetheradvances in ICTs have raised the structural rate ofgrowth in the US and elsewhere has begun to illustrate,the optimistic view about future productivity growthvery much depends on economy-wide diffusion. Theresult that there are FDI spillovers in technologicallyproximate areas—outside of the narrowly-definedFDI industry itself—byGirma and Wakelin (2001)andHaskel et al. (2001)is interesting in this context.

Work on the level of geographic localization of tech-nology diffusion has begun to shed light on whethertechnology diffusion is a force towards convergence ordivergence. It would be useful to extend this analysis toa broad set of countries, which would allow to forecastwhat will happen to the world’s distribution of income.The same is true for the result byKeller (2001c, 2002a)that technology has become less localized. It is impor-tant to find out whether that has happened only amongOECD countries, or also in a broader set of countries:has technology become increasingly global, or merelymore common among the relatively rich countries?

We also know very little right now about what arethe primary causes that have led to this lower degreeof knowledge localization among OECD countries.Learning more about these causes is important forthinking about what kind of policies might be effectivein fostering the diffusion of technology from abroad.

Other important issues include the following. First,empirical studies of international technology diffu-sion provide to some extent different results becauseof conceptual and measurement differences. One im-portant issue is how narrow or broad the concept ofknowledge diffusion is. Patent citations are the nar-rowest measure that has been used, which is followedby patents or patent applications per se. A broadernotion of technology diffusion is implicitly utilizedby studies that infer international technology diffu-sion from the relationship between foreign R&D andproductivity. The least narrow notion of technologyspillovers is implied when some measure of inter-national economic activity is related to productivity(e.g. the share of foreign-owned employment in totalemployment in much of the work on FDI spillovers).

The best measure of spillovers in the sense of anexternality is probably a patent citation. All othermeasures might pick up also market-based technologytransactions (such as royalty payments and licensing

fees) and, probably more importantly, the outcome ofchanges in the competitive position of firms. Theseeffects are not externalities, and they might have dif-ferent welfare implications. However, a well-knownproblem of using patent statistics—both applicationsand citations—is that technological knowledge is notall patented (Griliches, 1990provides a broader dis-cussion). This could be because applying for a patent,or citing another patent, is a strategic decision of thefirm. Even if this is not an issue, as long as a largepart of knowledge is tacit, patent statistics will neces-sarily miss that part, because codification is necessaryfor patenting to occur.55

Thus, while some approaches to identify techno-logy spillovers capture more than the pure externalityeffect, patent statistics might capture less than whatwe are interested in. Because a clearly dominatingmeasure to identify technology diffusion is not avail-able, there are good reasons for applying differentapproaches, as long as the relative advantages anddisadvantages are recognized.

Second, a promising development for future empir-ical work on international technology diffusion is theincreasing availability of large micro (firm or plant)level data sets. At times, there are sometimes exag-gerated expectations. Micro-level data allow avoidingcomposition and aggregation biases that are oftenpresent in industry- and aggregate-level analysis.Other issues seem to be becoming even more pressingat the firm-level, such as the unavailability of price de-flators. And to interpret a cross-sectional correlationof foreign ownership and productivity as evidencefor FDI spillovers would be just inappropriate at thefirm-level as it is at a more aggregate-level. As thestudies discussed above indicate, however, micro-leveldata can help to take a major step forward. The biggestcontribution of micro-level data sets comes probablyfrom a better estimation of micro behavior, as thedata records the outcome of economic decisions right

55 The inclusion of patent citations is at times imposed by thepatent examiner, which probably reduces the usefulness of patentcitations as a measure of knowledge flows. It is also worth keepingin mind that the number of patent applications differs acrosscountries because the propensity to patent seems to differ acrosscountries. For example,Eaton and Kortum (1996)scale down thenumber of Japanese patents by a factor of 4.9 for this reason, anapproach that is widely used in the literature. The value of 4.9 isbased on a 1992 working paper by Okada.


at the decision-taking level. Empirical micro studiesof technology diffusion that reach their full poten-tial involve probably modeling the micro behavioras well.

Third, the evidence reviewed above indicates thatmicro-level studies paint often a more pessimistic pic-ture on the strength of learning externalities throughinternational economic activity than industry-levelor aggregate studies do. This could be because theformer avoid certain biases (see above); it might alsohave to do with the fact that data on the learning effortof host country firms has typically not been available.A third reason could be that micro-level studies sofar are not based on a framework that explicitly al-lows for learning and spillover externalities betweenfirms. That is, so far the micro evidence is basedon ‘no-externality’ models. Future empirical workwill hopefully employ models where the presumedlearning effects from international economic activityexist as a matter of theory. At the same time, theempirical analysis should avoid picking up spuriouseffects.

What are the policy implications of this literature?The evidence is not strong enough yet to provide sup-port for specific policy measures, such as a particularsubsidy to a multinational enterprise for locating in acountry. This would require more agreement on quan-titative effects than there is right now. At this point,the results suggest that the international dimension oftechnological change is of key importance for mostcountries. In this situation, a closed-off internationaleconomic regime must have detrimental welfare con-sequences for a country.

There is also evidence that the relative importanceof the international technology diffusion has been in-creasing with the level of economic integration in theworld. This suggests that the performance advantageof outward-oriented economies over inward-orientedeconomies will be higher in the future than it is to-day. What if future research were to establish thatinternational technology diffusion raises productivitymore in advanced than in less developed countries?The implication for less developed countries wouldnot be that a turn towards autarky is beneficial—onthe contrary, given the evidence on positive produc-tivity effects from international technology diffusion.Rather, if the benefits from operating in an interna-tional economic environment differ across countries,

this suggests to investigate further the major reasonsfor this—what works, and why?

While there is no consensus yet on the exactmagnitude of spillover benefits, it is clear thatwell-functioning markets and an undistorted trade-and foreign-investment regime are conducive to theselearning effects. However, the evidence suggests thatthe latter are quite difficult to pin down. For one,it does not seem to be primarily a matter of simplyspecializing on high-tech goods. Rather, the goodscharacteristics are only part of what is important.India, for example, aimed at producing relativelyadvanced products during its era of import substitu-tion policy, but their quality was low compared tointernational standards. In contrast, south–east Asiancountries such as Hong Kong and South Korea wereinitially specializing in relatively low-tech products,and moved successfully into the range of higher-techproducts only gradually.

Technology diffusion is by no means automatic.Technological knowledge spillovers appear to be re-sulting from a deliberate commitment to learning andmatching international performance standards throughongoing interaction with foreigners. Local efforts areclearly necessary for successful technology adoptionas well. For that, technological capability to acquire,assimilate, or create technology, is needed in a coun-try to successfully develop economically.56 This isparticularly important for developing countries, giventhat they purchase or otherwise acquire most of theirtechnology from abroad. Developing technologicalcapability requires appropriate policies that help build-ing the necessary institutions and provide the correctincentives. This framework will also contribute toefficient domestic technology transfers, for instancebetween different industries. At the same time, theongoing interaction with foreign firms and consumersseems to be a process of knowledge discovery forfirms that cannot be had from interacting only withother domestic firms. To model as well as empiricallycapture these in a convincing way continues to be achallenge.

56 This is consistent with the emphasis of evolutionary theories onthe importance of networks and links between diverse institutionsas a major factor behind the innovative capacity of a nation (cf.Nelson, 1993).


Acknowledgements

We would like to thank Sourafel Girma for provid-ing some regression results. We have also benefitedfrom conversations with Bob Baldwin and ElhananHelpman, as well as from comments by Peter Klenow,Sam Kortum, Jörg Mayer, and Maurice Schiff. Thispaper draws heavily onKeller (2001a).

References

Adams, J., Jaffe, A., 1996. Bounding the effects of R&D: aninvestigation using matched establishment-firm data. RANDJournal of Economics 27, 700–721.

Aghion, P., Howitt, P., 1992. A model of growth through creativedestruction. Econometrica 60, 323–351.

Aghion, P., Howitt, P., 1998. Endogenous Growth Theory. MITPress, Cambridge, MA.

Aitken, B., Harrison, A., 1999. Do domestic firms benefit fromforeign direct investment? Evidence from Venezuela. AmericanEconomic Review 89, 605–618.

Arrow, K., 1969. Classificatory notes on the production andtransmission of technological knowledge. American EconomicReview Paper and Proceedings 59, 29–35.

Baldwin, R., Braconier, H., Forslid, R., 1999. Multinationals,endogenous growth and technological spillovers: theory andevidence. CEPR Discussion Paper No. 2155.

Behrman, J., Srinivasan, T.N. (Eds.), Handbook of DevelopmentEconomics, vols. 3A and 3B. Elsevier, Amsterdam.

Bernard, A., Jensen, B., 1999. Exceptional exporter performance:cause, effect or both? Journal of International Economics 47,1–25.

Bernstein, J., Mohnen, P., 1998. International R&D spilloversbetween US and Japanese R&D intensive sectors. Journal ofInternational Economics 44, 315–338.

Blyde, J., 2001. Trade and technology diffusion in Latin America.Inter-American Development Bank Working Paper, Washington,DC.

Blomström, M., Kokko, A., 1998. Multinational corporations andspillovers. Journal of Economic Surveys 12, 247–277.

Bottazzi, L., Peri, G., 1999. Innovation, demand, and knowledgespillovers: theory and evidence from european regions. CEPRWorking Paper No. 2279, London, UK, October.

Branstetter, L., 2001a. Is foreign direct investment a channel ofknowledge spillovers? Evidence from Japan’s FDI in the UnitedStates, NBER Working Paper No. 8015.

Branstetter, L., 2001b. Are knowledge spillovers international orintranational in scope? Microeconometric evidence from theUS and Japan. Journal of International Economics 53, 53–79.

Carr, D., Markusen, J., Maskus, K., 2001. Estimating theknowledge-driven model of the multinational enterprise.American Economic Review 91, 693–708.

Caselli, F., John Coleman II, W., 2001. Cross-country technologydiffusion: the case of computers. American Economic Review91, 328–335.

Cimoli, M., 2001. Networks, market structures and economicshocks, the structural changes of innovation systems in LatinAmerica. In: Proceedings of the Paper Presented at Seminaron Redes Peroductivas e Institucionales en America Latina,Buenos Aires, 9–12 April 2001.

Cimoli, M., Katz, J., 2001. Structure reforms, technology gapsand economics development: a Latin American perspective.In: Proceedings of the Paper Presented at DRUID-Nelson andWinter Conference, Aalborg, 12–15 June 2001.

Clerides, S., Lach, S., Tybout, J., 1998. Is learning by exportingimportant? Micro-dynamic evidence from Colombia, Mexico,and Morocco. Quarterly Journal of Economics 113, 903–948.

Coe, D.T., Helpman, E., 1995. International R&D spillovers.European Economic Review 39, 859–887.

Coe, D., Hoffmaister A., 1999. Are there international R&Dspillovers among randomly matched trade partners? A responseto Keller, IMF Working Paper No. 99/18, February.

Coe, D., Helpman, E., Hoffmaister, A., 1997. North–southspillovers. Economic Journal 107, 134–149.

Cohen, W., Levinthal, D., 1989. Innovation and learning: the twofaces of R&D. Economic Journal 99, 569–596.

Connolly, M., 1998. The dual nature of trade: measuring its impacton imitation and growth. Working Paper, Duke University,December.

Dahlman, C.J., Ross-Larson, B., Westphal, L.E., 1987. Managingtechnological development: lessons from the newly indus-trializing countries. World Development 15, 759–775.

David, P., 1992. Knowledge, property, and the systems dynamicsof technological change. In: Summers, L., Shah, A. (Eds.),Proceedings of the World Bank Annual Conference onDevelopment Economics 1992, pp. 215–248.

David, P., Olsen, T., 1992. Technology adoption, learningspillovers and the optimal duration of patent based monopolies.International Journal of Industrial Organization 10, 517–544.

Dornbusch, R., Fischer, S., Samuelson, P., 1977. Comparativeadvantage, trade, and payments in a Ricardian model with acontinuum of goods. American Economic Review 67, 823–839.

Dosi, G., Nelson, R., 1994. An introduction to evolutionary theoriesin economics. Journal of Evolutionary Economics 4 (3), 15–72.

Easterly, W., Levine, R., 2001. It’s not factor accumulation:stylized facts and growth models. In: Paper Presented at theConference on What Have We Learned from a Decade ofEmpirical Research on Growth? 26 February 2001, The WorldBank. World Bank Economic Review 15, 177–219.

Eaton, J., Kortum, S., 1996. Trade in ideas: patenting andproductivity in the OECD. Journal of International Economics40, 251–278.

Eaton, J., Kortum, S., 1997. Engines of growth: domestic andforeign sources of innovation. Japan and the World Economy9, 235–259.

Eaton, J., Kortum, S., 1999. International patenting and technologydiffusion: theory and measurement. International EconomicReview 40, 537–570.

Eaton, J., Kortum, S., 2001a. Trade in capital goods. EuropeanEconomic Review 45 (7), 1195–1235.

Eaton, J., Kortum, S., 2001b. Technology, geography, and trade.Econometrica (in press).


Eaton, J., Gutierrez, E., Kortum, S., 1998. European technologypolicy. Economic Policy 27, 405–438.

Economist, 2000. Untangling E-conomics. A survey of the neweconomy. The Economist, 23 September.

Ethier, W., 1982. National and international returns to scale.American Economic Review.

Evenson, R., Westphal, L., 1995. Technology change andtechnology strategy. In: Behrman, J., Srinivasan, T.N. (Eds.),Handbook of Development Economics, vol. 3A. North-Holland,Amsterdam.

Fagerberg, J., 1994. Technology and international differences ingrowth rates. Journal of Economic Literature 32 (September),1147–1175.

Feenstra, R., 1995. Integration of trade and disintegration ofproduction in the global economy. Journal of EconomicPerspectives 1998 (Fall), 31–50.

Feenstra, R., Markusen, J., Zeile, W., 1994. Accounting for growthwith new inputs. International Economic Review 35, 429–447.

Feldman, M., Lichtenberg, F., 1997. The impact and organizationof publicly-funded research and development in the EuropeanCommunity. NBER Working Paper No. 6040.

Gera, S., Gu, W., Lee, F., 1999. Information technology andproductivity growth: an empirical analysis for Canada and theUnited States. Canadian Journal of Economics 32, 384–407.

Geroski, P.A., 2000. Models of technology diffusion. ResearchPolicy 29, 603–625.

Gerschenkron, A., 1962. Economic Backwardness in HistoricalPerspective. Harvard University Press, Cambridge, MA.

Girma, S., Wakelin, K., 2001. Regional underdevelopment: is FDIthe solution? A semi-parametric analysis. GEP Research PaperNo. 2001/11, University of Nottingham, UK.

Globerman, S., Kokko, A., Sjöholm, F., 2000. Internationaltechnology diffusion: evidence from Swedish patent data.Kyklos 53, 17–38.

Griffith, R., Redding, S., Van Reenen, J., 2000a. Mapping two facesof R&D: productivity growth in a panel of OECD industries.Mimeo, London School of Economics, November.

Griffith, R., Redding, S., Van Reenen, J., 2000b. R&D andabsorptive capacity: from theory to data. Mimeo, LondonSchool of Economics, November.

Griliches, Z., 1979. Issues in assessing the contribution of researchand development to productivity growth. Bell Journal ofEconomics 10, 92–116.

Griliches, Z., 1990. Patent statistics as economic indicators: asurvey. Journal of Economic Literature 28, 1661–1707.

Griliches, Z., 1995. R&D and productivity: econometric resultsand measurement issues. In: Stoneman, P. (Ed.), Handbookof the Economics of Innovation and Technological Change.Blackwell, Oxford, pp. 52–89.

Grossman, G., Helpman, E., 1991. Innovation and Growth in theWorld Economy. MIT Press, Cambridge, MA.

Grossman, G., Helpman, E., 1995. Technology and trade. In:Grossman, G., Rogoff, K. (Eds.), Handbook of InternationalEconomics, vol. 3. North-Holland, Amsterdam.

Hall, R., Jones, C., 1999. Why do some countries produce somuch more output per worker than others? Quarterly Journalof Economics 114, 83–116.

Hanson, G., Mataloni, R., Slaughter, M., 2001. Expansionstrategies of US multinational firms. NBER Working Paper No.8433.

Hanushek, E., Kimko, D., 2000. Schooling, labor-force quality, andthe growth of nations. American Economic Review, December.

Haskel, J., Pereira, S., Slaughter, M., 2001. Does inward foreigndirect investment boost the productivity of domestic firms? In:Paper Presented at the NBER Summer Institute, August.

Howitt, P., 2000. Endogenous growth and cross-country incomedifferences. American Economic Review.

Hulten, C.R., 2001. Total factor productivity. In: Hulten, C.R.,Dean, E.R., Harper, M.J. (Eds.), New Developments inProductivity Analysis. The University of Chicago Press,Chicago and London, pp. 1–53.

Griliches, Z., 1995. R&D and productivity: econometric resultsand measurement issues. In: Stoneman, P. (Ed.), Handbookof the Economics of Innovation and Technological Change.Blackwell, Oxford, pp. 52–89.

Irwin, D., Klenow, P., 1994. Learning spillovers in the semi-conductor industry. Journal of Political Economy.

Jaffe, A., 1986. Technological opportunity and spillover ofR&D: evidence from firm’s patents, profits, and market value.American Economic Review 76, 984–1001.

Jaffe, A., Trajtenberg, M., 1996. Modelling the flows of knowledgespillovers. NBER Working Paper, Cambridge, MA.

Jaffe, A., Trajtenberg, M., 1998. International knowledge flows:evidence from patent citations. NBER Working Paper No. 6507,April.

Jaffe, A., Trajtenberg, M., Henderson, R., 1993. Geographiclocalization of knowledge spillovers as evidenced by patentcitations. Quarterly Journal of Economics 108, 577–598.

Katz, J., 2000. Reformas estructurales, productividad y ConductaTecnologica, Fondo de Cultura YCEPAL, Santiago (Englishedition: Structural Reforms, Productivity and TechnologicalChange in Latin America, ECLAC, Santiago, Chile).

Keller, W., 1996. Absorptive capacity: on the creation andacquisition of technology in development. Journal of Develop-ment Economics 49, 199–227.

Keller, W., 1997a. Technology flows between industries: identi-fication and productivity effects. Economic Systems Research9, 213–220.

Keller, W., 1997b. How trade patterns and technology flows affectproductivity growth. World Bank Policy Research WorkingPaper No. 1831, September.

Keller, W., 1998. Are international R&D spillovers trade related?Analyzing spillovers among randomly matched trade partners.European Economic Review 42, 1469–1481.

Keller, W., 2000. Do trade patterns and technology flows affectproductivity growth? World Bank Economic Review 14, 17–47.

Keller, W., 2001a. International technology diffusion. NBERWorking Paper No. 8573, October.

Keller, W., 2001b. Do the new information and trade links of the1990s lead to convergence or divergence? In: Paper Presentedat the NBER Summer Institute, Cambridge, MA, August.

Keller, W., 2001c. Knowledge spillovers at the world’s technologyfrontier. CEPR Working Paper No. 2815, May.

Keller, W., 2002a. Geographic localization of internationaltechnology diffusion. American Economic Review 92, 120–142.


Keller, W., 2002b. Trade and the transmission of technology.Journal of Economic Growth 7, 5–24.

Kinoshita, Y., 2000. R&D and technology spillover via FDI:innovation and absorptive capacity. Mimeo, William DavidsonInstitute at the University of Michigan Business School,October.

Klenow, P., 2001. Comments on the Paper by Easterly and Levineat the Conference on What Have We Learned from a Decade ofEmpirical Research on Growth? 26 February 2001, The WorldBank. World Bank Economic Review.

Klenow, P., Rodriguez-Clare, A., 1997. The neoclassical revival ingrowth economics: has it gone too far? NBER MacroeconomicsAnnual 12, 73–102.

Kraay, A., I. Soloaga, J. Tybout, 2001. Product quality, productiveefficiency, and international technology diffusion: evidence fromplant-level panel data. In: Paper Presented at the NBER SummerInstitute, July.

Leamer, E., Levinsohn, J., 1995. International trade theory: theevidence. In: Grossman, G. Rogoff, K. (Eds.), Handbook ofInternational Economics, vol. 3. Elsevier, Amsterdam.

Lichtenberg, F., van Pottelsberghe de la Potterie, B., 1996.International R&D spillovers: a re-examination. NBER WorkingPaper No. 5688.

Lucas, R., 1988. On the mechanics of economic development.Journal of Monetary Economics 22, 3–42.

Lucas, R., 1993. Making a miracle. Econometrica 61, 251–272.Lumenga-Neso, O., Olarreaga, M. Schiff, M., 2001. On ‘indirect’

trade-related R&D spillovers. Mimeo, The World Bank,February.

Lundvall, B.-Å. (Ed.), 1992. National Systems of Innovation:Towards a Theory of Innovation and Interactive Learning.Pinter, London.

Mankiw, N.G., 1995. The growth of nations. Brookings Papers onEconomic Activity 1, 275–310.

McKinsey, 2000. Why the Japanese economy is not growing:micro-barriers to productivity growth. McKinsey GlobalInstitute, Washington, DC, July.

Mayer, J., 2001. Technology diffusion, human capital andeconomic growth in developing countries. UNCTAD DiscussionPaper No. 154, Geneva, June.

Mohnen, P., 1992. The Relationship between R&D andProductivity Growth in Canada and other IndustrializedCountries. Minister of Supply and Services Canada, Ottawa.

Mohnen, P., 2001. International R&D spillovers and economicgrowth. In: Pohjola, M. (Ed.), Information Technology,Productivity, and Economic Growth: International Evidence,UNU/WIDER and Sitra. Oxford University Press, Oxford.

Nadiri, M.I., Kim, S., 1996. International R&D spillovers, tradeand productivity in major OECD countries. NBER WorkingPaper No. 5801.

Nelson, R. (Ed.), 1993. National Innovation Systems: AComparative Analysis. Oxford University Press, Oxford, 1993.

Nelson, R., Phelps, E., 1966. Investment in humans, technologicaldiffusion and economic growth. American Economic Review56, 69–75.

Nelson, R., Winter, S., 1982. An Evolutionary Theory of EconomicChange. Belknap Press, Cambridge.

Nelson, R., Winter, S., 2002. Evolutionary theorizing in economics.The Journal of Economic Perspectives 16, 23–46.

Olley, S., Pakes, A., 1996. The dynamics of productivity inthe telecommunications equipment industry. Econometrica 64,1263–1297.

Olson, M., 1982. The Rise and Fall of Nations. Yale UniversityPress, New Haven, CT.

Parente, S., Prescott, E., 2000. Barriers to Riches. MIT Press,Cambridge, MA.

Park, W., 1995. International R&D spillovers and OECD economicgrowth. Economic Inquiry 23, 571–591.

Polanyi, M., 1958. Personal Knowledge Towards a Post-CriticalPhilosophy. University of Chicago Press, Chicago, IL.

Prescott, E.C., 1998. Needed: a theory of total factor productivity.International Economic Review 39, 525–551.

Quah, D., 2001a. Technology Dissemination and economic growth.Working Paper, LSE, April.

Quah, D., 2001b. Demand-driven knowledge clusters in aweightless economy, LSE, April.

Rhee, Y., Ross-Larson, B., Pursell, G., 1984. Korea’s CompetitiveEdge: Managing the Entry into World Markets. Johns HopkinsUniversity Press for the World Bank, Baltimore, MD.

Rivera-Batiz, L., Romer, P., 1991. Economic integration andendogenous growth. Quarterly Journal of Economics 106, 531–555.

Romer, P., 1990. Endogenous technological change. Journal ofPolitical Economy 98, S71–S102.

Romer, P., 1992. Two strategies for economic development:using ideas and producing ideas. In: Summers, L., Shah, A.(Eds.), Proceedings of the World Bank Annual Conference onDevelopment Economics 1992, pp. 63–91.

Rosenberg, N., 1976. On technological expectations. EconomicJournal 86, 223–235.

Scherer, F., 1984. Using linked patent and R&D data to measureinterindustry technology flows. In: Griliches, Z. (Ed.), R&D,Patents, and Productivity. The University of Chicago for NBER,pp. 417–461.

Segerstrom, P., Anant, T., Dinopoulos, E., 1990. A Schumpeterianmodel of the product life cycle. American Economic Review80, 1077–1092.

Sjöholm, F., 1996. International transfer of knowledge: therole of international trade and geographic proximity.Weltwirtschaftliches Archives 132, 97–115.

Solow, R., 1957. Technical change and the aggregate productionfunction. Review of Economics and Statistics.

Solow, R., 2001. Applying growth theory across countries. In:Comments Presented at the Conference on What Have WeLearned from a Decade of Empirical Research on Growth?26 February 2001, The World Bank. World Bank EconomicReview.

Stoneman, P., Kwon, M.-J., 1994. The diffusion of multiple processtechnologies. Economic Journal 104, 420–431.

Teece, D., 1977. Technology transfer by multinational firms: theresource cost of transferring technological know-how. EconomicJournal 87, 242–261.

Tybout, J., 2000. Manufacturing firms in developing countries:how well do they do, and why? Journal of Economic Literature38 (March), 11–44.


Tybout, J., 2001. Plant- and firm-level evidence on “new” tradetheories. NBER Working Paper No. 8418, August.

Vernon, R., 1966. International investment and international tradein the product cycle. Quarterly Journal of Economics 80, 190–227.

von Hippel, E., 1994. Sticky information and the locus of problemsolving: implications for innovation. Management Science 40,429–439.

Westphal, L., Kim, L., Dahlman, C., 1985. Reflections on theRepublic of Korea’s acquisition of technological capability. In:

Rosenberg, N., Frischtak, C. (Eds.), International TechnologyTransfer. Praeger, New York, pp. 167–227.

Witt, U., 1993. Introduction. In: Witt, U. (Ed.), EvolutionaryEconomics. Edward Elgar, Cheltenham, pp. xii–xxvii.

Xu, B., 2000. Journal of Development Economics 62, 477–493.Xu, B., Wang, J., 1999. Capital goods trade and R&D

spillovers in the OECD. Canadian Journal of Economics 32,1258–1274.

Young, A., 1991. Learning-by-doing and the dynamic effects ofinternational trade. Quarterly Journal of Economics.

convergence and polarization in global income levels: a

Documents