Bayoumia Coe Helpman (1999) R&D Spillovers and Global Growth

Journal of International Economics 47 (1999) 399–428

R&D spillovers and global growtha a , b*Tamim Bayoumi , David T. Coe , Elhanan Helpman

aInternational Monetary Fund, 700 19th Street, NW, Washington, DC 20431, USAbTel-Aviv University, The Canadian Institute for Advanced Research, Harvard University,

Cambridge, MA 02138, USA

Received 14 August 1996; received in revised form 12 August 1997; accepted 12 August 1997

Abstract

We examine the growth promoting roles of R&D, international R&D spillovers, and tradein a world econometric model. A country can raise its total factor productivity by investingin R&D. But countries can also boost their productivity by trading with other countries thathave large ‘stocks of knowledge’ from their cumulative R&D activities. We use a specialversion of MULTIMOD that incorporates R&D spillovers among industrial countries andfrom industrial countries to developing countries. Our simulations suggest that R&D, R&Dspillovers, and trade play important roles in boosting growth in industrial and developingcountries. 1999 Elsevier Science B.V. All rights reserved.

Keywords: Productivity; R&D; Spillovers; North-South

JEL classification: 031; 040

1. Introduction

National economies are embedded in a global system that generates mutualinterdependence across countries. In this system each country depends on thesupply of consumer goods, intermediate products, and capital goods from its tradepartners, and it relies on the trade partners to supply markets for its own products.But — as is becoming more and more apparent — countries also rely on eachother for technology transfer, and they learn from each other manufacturing

*Corresponding author. Tel.: 11 202 6238984; fax: 11 202 6234132; e-mail: [email protected]

0022-1996/99/$ – see front matter 1999 Elsevier Science B.V. All rights reserved.PI I : S0022-1996( 98 )00018-X

400 T. Bayoumi et al. / Journal of International Economics 47 (1999) 399 –428

methods, modes of organization, marketing, and product design. These featuresaffect their well-being and link their growth rates.

Much research has been done in recent years to clarify such links. Some of ithas been theoretical, some has been empirical. In this paper we contribute to theempirical literature by providing a quantitative evaluation of the importance ofR&D and trade in influencing total factor productivity and output growth. For thispurpose we incorporate previous estimates of international R&D spillovers —among industrial countries and from industrial to developing countries — into amulticountry macroeconometric model in order to simulate the influence ofchanges in R&D and trade on the evolution of the world economy, with particularemphasis on the interactions between individual economies and regions.

Estimates of international R&D spillovers, which underline trade relations as themajor transmission mechanism, are taken from Coe and Helpman (1995) and Coeet al. (1997). They have been embedded in the IMF’s MULTIMOD econometricmodel for this study. The augmented model was then used to simulate changes inR&D in the industrial countries and in the exposure to trade of the developingcountries in order to obtain estimates of induced changes in total factor productivi-ty, capital, output, and consumption in each of 12 ‘countries’. The countriesconsist of the G-7 countries plus five industrial and developing country regions.

Our simulations suggest that the interplay between R&D and capital investmentis important. While R&D has a direct effect on productivity and thereby on output,one fourth to one third of the total increase in output results from investment incapital that is induced by the higher levels of productivity. And we find thatinternational R&D spillovers, leveraged by investment, are very important. Were

1]the United States to increase its R&D investment by of 1% of GDP and2

maintain the new R&D/GDP ratio thereafter, its output would rise by over 9%after 80 years, the output of the other industrial countries would increase by almost3%, and the output of the developing countries would increase by about 3.5%. If

1]all industrial countries were to raise their R&D investment by of 1%, their2

output would rise after 80 years by more than 17% and the output of developingcountries would rise by over 10%. Clearly, not only industrial countries benefitfrom R&D investment; developing countries are also major beneficiaries of R&Dinvestment in the industrial countries. We also find that further expansion of tradeby the developing countries by an amount equivalent to 5% of their GDP wouldraise their output by about 6.5% after 80 years. This indicates that trade expansioncan increase access to foreign technology, and thereby contribute to growth indeveloping countries.

We outline in Section 2 the theoretical framework of MULTIMOD and thetheoretical considerations that have guided the specification of the R&D spilloverequations incorporated into the model. In Section 3 we describe key features of theempirical model that are important to understand the simulations reported inSection 4. Conclusions are drawn in Section 5.

T. Bayoumi et al. / Journal of International Economics 47 (1999) 399 –428 401

2. Theoretical framework

The theoretical structure that drives MULTIMOD’s long-run supply behavior is1neoclassical. Each country has a Cobb-Douglas production function of the form

a 12aY 5 FK L , 0 , a , 1 , (1)

where Y is output, K is capital, L is labor, and F is total factor productivity.Although the coefficients and variables differ across countries, and the variablesdiffer across time, we omit country and time subscripts for expositional con-venience.

The world capital stock is ultimately determined by the level of world saving,which is derived from an aggregate consumption function. The allocation ofconsumption over time is derived from the maximization of an intertemporalutility function subject to a budget constraint. An individual’s flow of utility attime t is given by

12sct]]u 5 , (2)t 1 2 s

where c is the individual’s aggregate consumption at time t. The parameter st

determines the intertemporal elasticity of substitution in consumption. For anindividual who is alive at time t and who will live until T .t, the discounted flowof utility at time t equals

T

2d (r2t )U 5Ee u dt, (3)t t

t

where d represents his subjective discount rate and u is given in (2).t

Following Blanchard (1985), it is assumed that every individual faces a timeand age invariant probability of death, l, and has access to perfect annuitymarkets. As a result, an individual who is alive at time t maximizes the expectedvalue of U (given in (3)). The consumer faces an intertemporal budget constraintt

such that, at each point in time, the expected present value of aggregateconsumption equals the expected present value of labor income plus the value ofcapital owned at time t. The solution to this problem yields a consumption

1For more details about MULTIMOD see Masson et al. (1990); recent papers based on MULTIMODsimulation results are Bartolini et al. (1995) and Gagnon et al. (1996). When we refer to a feature of acountry, we mean a feature of a country or a country block. Our exposition focuses on the structure ofindustrial countries and the newly industrialized countries. Developing countries are treated somewhatdifferently, as explained in the next section.


function, where consumption is proportional to wealth (human and financial). Thefactor of proportionality depends on the subjective rate of time preference, on theprobability of death, and on the intertemporal elasticity of substitution inconsumption. Aggregating across individuals yields an aggregate consumptionfunction for the country, with consumption proportional to the country’s aggregatehuman and non-human wealth. For the country as a whole, the factor ofproportionality depends on the same parameters as the individual’s factor ofproportionality and also on the rate of population growth. This consumptionfunction is used to derive aggregate savings.

Saving and investment are jointly determined, and for the world as a whole,aggregate investment equals aggregate savings. Investment is allocated across

2countries to equalize risk-premia-adjusted rates of return. The output of eachcountry is treated as a distinct product. Given aggregate consumption andinvestment, the allocation of spending across countries depends on relative prices.These patterns of spending determine bilateral imports and exports.

In the standard version of MULTIMOD, total factor productivity and the laborforce are exogenous. Although in each country investment need not equal savings,because the gap can be financed by international capital flows, the intertemporalbudget constraints imply that the long-run growth of the capital stock isdetermined by the growth of labor and the growth of total factor productivity. Inthe long run, the growth of output is also determined by the same factors, and thecapital output ratio is constant. An implication of these relationships is that thelong-run growth rate of per capita output is entirely determined by the growth rateof total factor productivity. These features are familiar from the neoclassicalgrowth models of Solow (1956) and Cass (1965).

We augment the standard version of MULTIMOD with equations that relatetotal factor productivity to R&D investment and trade. In doing so, total factorproductivity becomes endogenous, as suggested by the ‘new’ growth theory (seeRomer (1990), Grossman and Helpman (1991), and Aghion and Howitt (1992)).But we do not follow the new growth theory all the way, since we do notendogenize R&D investment as a function of economic factors. Rather, we holdconstant the ratio of R&D investment to GDP. In tracing out the effects of anincrease in R&D investment we take account of the fact that, by temporarilyraising the marginal product of capital, improvements in total factor productivityinduce capital accumulation, which continues until the marginal product of capitalfalls to the level of the real long-run rate of interest. R&D investment thus affectsoutput directly through total factor productivity and indirectly through inducedcapital accumulation. The model enables us to evaluate each of these components.

It is important to note that our model incorporates diminishing returns to the3reproducible factors of production (physical and R&D capital) in aggregate. This

2In the short run, investment deviates from this rule, as discussed in the next section.3That is to say, it is not an ‘AK’ model; see, for example, Romer (1990).


implies that a permanent increase in R&D investment will have a level effect onoutput, but will not permanently raise the rate of growth of output. As is apparentfrom our simulation results, however, it takes more than 80 years to approach thenew steady state, and hence the impact on growth is very long lived.

The theoretical basis for our modeling of total factor productivity, which uses aconstant returns to scale aggregate Cobb-Douglas production function such as (1),is provided by Grossman and Helpman (1991, chapter 5). For example, let theproduction function of final output be

a g 12a 2gY 5 AK L D , 0 , a, g, a 1 g , 1 , (4)Y

where L is the amount of labor used directly in the manufacturing of final outputY

and D is a symmetric CES index of intermediate inputs. The parameter A is1 / (´21)constant. We know that in this case D5n L in equilibrium, where nD

represents the number of available intermediates, L the labor force employed inD

the manufacturing of intermediates (we assume for simplicity that intermediatesare manufactured only with labor), and ´.1 is the elasticity of substitutionbetween intermediate inputs. Using the demand functions for inputs that areimplied by (4) and the pricing of intermediates (i.e. a constant markup overmarginal costs, with the price /marginal-cost ratio equal to 1 /(121/´)), it followsthat the aggregate production function for final output can be represented by (1).In this reduced form, L equals direct plus indirect labor (L5L 1L ) while F canY D

be represented by

(12a 2g ) / (´21)F 5 Bn . (5)

In (5) the constant B depends on the parameters of the production function (4).It is clear from (5) that in this model total factor productivity depends on the

available assortment of intermediate inputs (n): the more intermediates are used inproduction, the higher is total factor productivity. On the other hand, intermediateinputs have to be developed. As a result, the number of available intermediates is afunction of past R&D investment. We therefore have a link between currentproductivity and cumulative R&D investment. This type of link is central to ourspecification presented in the next section.

But we do not wish to restrict our empirical specification to a narrowly definedstructural link between R&D and total factor productivity as described above.Rather, we use this theory to guide our empirical specification. As pointed out byGrossman and Helpman (1991), there are a number of channels through whichtotal factor productivity of a country is affected by the R&D investment of itstrade partners in addition to its own R&D investment level. Foreign trade plays animportant role in these transmission mechanisms. For example, foreign tradeenables a country to employ a larger variety of intermediate inputs, includingcapital goods, and it stimulates learning from trade partners. For these reasons wespecify a functional relationship between total factor productivity and cumulative


R&D levels that is broader than (5), and which builds on previous empirical work.The precise specification of these links is described in the next section.

3. Empirical model

In the version of MULTIMOD used here, total factor productivity is en-dogenously determined by the stock of R&D capital, international R&D spillovers,and trade. Total factor productivity together with capital and labor inputs thendetermine potential output. This supply side is augmented by short-run dynamicslargely emanating from changes in aggregate demand caused by the interactionbetween sticky prices and forward-looking expectations. While changes inaggregate demand move actual output temporarily away from its potential level,monetary policy is neutral in the long run. There is, however, a long-run impactfrom fiscal policy, reflecting the wedge between the discount rate of individualsand of the government caused by the probability of death. MULTIMOD alsoincorporates rational expectations in goods, financial, and labor markets. Theforward-looking aspect of the model means that changes in expectations of futureincreases in productivity or wealth can have immediate effects on, for example,current consumption and investment.

Our version of MULTIMOD consists of 12 linked econometric models: a modelfor each of the G-7 countries (the United States, Japan, Germany, France, Italy, theUnited Kingdom, and Canada), an aggregate model for the other industrialcountries, and four regional models for non-oil-exporting developing countries.The developing countries are disaggregated into regional models for Africa, theWestern Hemisphere, the newly industrializing economies of Asia (the NIEsconsisting of Hong Kong, Korea, Singapore, and Taiwan), and other non-oil-exporting developing countries; the latter group is mainly Asian countries otherthan the NIEs. Most parameters have been estimated with pooled annual data. Themost important features of these models are summarized below to help understand

4the simulation results presented in the next section.Output is determined by aggregate demand in the short run and by the

underlying level of aggregate supply — ‘potential output’ — in the long run. APOTCobb-Douglas production function such as (1) determines potential output (Y ).

In logarithmic form, and omitting country subscripts and time subscripts fromcurrent period variables,

POTlogY 5 k 1 a logK 1 (1 2 a)logL 1 logF,

4Complete equation specification and parameter values for the industrial countries are presented inMasson et al. (1990); and for the developing countries in Bayoumi et al. (1995). There is also a verysimple model for the oil-exporting developing countries, but there is no model for the economies intransition of central and east Europe and the former Soviet Union.


where a is capital’s share of national income and k is a country-specific constant.The real stock of capital is endogenous, as discussed below, and labor supply isdetermined by the natural rate of unemployment and demographic factors, both ofwhich are exogenous.

We endogenize total factor productivity based on the estimation results in Coeand Helpman (1995) for the industrial countries and in Coe et al. (1997) for the

5developing countries. Both of these studies present estimation results based on anumber of alternative specifications. In the preferred specification in both studies,total factor productivity is determined by the stock of R&D capital and the share

6of imports of manufactures in GDP. For the industrial countries, which dovirtually all of the R&D in the world economy, total factor productivity is

D Fdetermined by both domestic R&D capital (S ) and foreign R&D capital (S ).Trade is assumed to be the vehicle for R&D spillovers and thus foreign R&Dcapital, which is defined below, affects total factor productivity through itsinteraction with the import share (m). The equation determining total factorproductivity (F ) for each of the G-7 countries is,

D FlogF 5 f 1 0.24logS 1 0.26m ? logS 2 3.18m ,1

where f is a country-specific constant. For the small industrial countries in1

aggregate, total factor productivity is determined in the same manner except thatdomestic R&D capital has a smaller impact, which may reflect that they performR&D across a narrower range of R&D activities,

D FlogF 5 f 1 0.08logS 1 0.26m ? logS 2 3.18m .2

The developing countries generally do little, if any, R&D. Their domestic R&Dcapital is assumed to be constant. In each of the non-oil-developing country

5The equations reported in Coe and Helpman (1995) (CH) have been modified in light ofLichtenberg and van Pottelsberghe de la Potterie (1997) who point out that the CH estimates of theelasticity of TFP with respect to the foreign R&D capital stock are sensitive to the normalization of theforeign R&D capital stock series, which was indexed to equal 1.0 in 1985. For this reason we havere-estimated the CH equations, using the unindexed data on foreign R&D capital stocks and adding theimport share as an independent variable (the import share was omitted in the CH specification becausethe estimated coefficient was insignificant). This modification has only a small effect on the estimatedelasticity of TFP with respect to the foreign R&D capital stock (the new elasticity is 0.26 comparedwith 0.29 in CH). We report the corrected estimates in the text below and use them in the simulations.

6All of the coefficients in the equations used here were significantly different from zero inestimation. Coe and Helpman (1995) use imports from industrial countries of goods and servicesinstead of imports of manufactures. Coe et al. (1997, Table 3) report results using imports fromindustrial countries of goods and services, of machinery and equipment, and of manufactures; we reportthe results for manufactures below and use them in the simulations. The simulations are based onimports of manufactures from all countries since MULTIMOD does not distinguish between importsfrom specific countries or regions. The relevant coefficients have been adjusted to reflect different meanvalues for the import shares.


regions and in the newly industrializing economies, total factor productivity is7determined as,

FlogF 5 f 1 0.43m ? logS 2 5.09m .3

In Coe et al. (1997) total factor productivity in the developing countries alsodepends on human capital, proxied by secondary school enrolment ratios, as wellas on country-specific time trends; these are ignored here since they play no role inthe simulations discussed below.

The simulation results reported in the next section will, of course, dependimportantly on the size of the estimated parameters in these equations. Theelasticity of total factor productivity with respect to domestic R&D capital (log

DS ) is within the range of results from the studies surveyed by Nadiri (1993) and,at least for the small industrial countries, by Griliches (1988). The foreign R&D

Fcapital stock (log S ) and the import share (m) are interacted in the equations, sotheir elasticities will depend on the level of the other variable. Given an averagevalue of the import share (m) of about 0.3 for the industrial countries and 0.14 forthe developing countries, the elasticity of total factor productivity with respect toforeign R&D capital is 0.08 for the industrial and 0.12 for the developingcountries. These elasticities are, if anything, at the low end of the range of resultsfrom the studies of R&D spillovers or externalities surveyed by Nadiri (1993) and

8Mohnen (1994). Although each of the coefficients on the import share by itself isnegative, given an average value of the logarithm of foreign R&D capital of about12, the ‘total’ elasticity of total factor productivity with respect to the importshares is about 0.1 for the developing countries and roughly zero for the industrial

9countries.The domestic R&D capital stocks of the G-7 industrial countries and the small

industrial countries in aggregate consist of their cumulative real investment inR&D (R), allowing for depreciation,

D DS 5 0.95S 1 R ,t21

Dwhere S is beginning of period. As noted above, real R&D expenditures are aconstant share of the simulated level of potential GDP. The foreign R&D capitalstock, which is defined in the same manner for all countries and groups of

7We specify this equation in log levels even though it was estimated in Coe et al. (1997) as the firstdifference of logs; the way that the equation is specified in the model makes no difference to thesimulation results.

8See also Eaton and Kortum (1996), who find large and significant international technologyspillovers based on patent data.

9 sm m sm FThe elasticities are defined as a m for foreign R&D capital and (a 1a log S ) for the importm sm Fshare, where a is the coefficient on m, and a is the coefficient on m?log S . The elasticities reported

above are the same as in Coe and Helpman (1995) and Coe et al. (1997); given the adjustments to thecoefficients referred to in footnote 6, these are the same elasticities used in the simulations.


countries, is a weighted average of industrial countries’ domestic R&D capitalstocks using bilateral import shares as weights. For a specific country or country

Fgrouping j, the foreign R&D capital stock (S ) is,j

8 8F DS 5Oa S , with Oa 5 1 and a 5 0 ,j ji i ji ii

I51 I51

where a are the elements of a 1238 matrix of the manufactures imports ofji

country j from industrial country i as a proportion of total manufactures imports ofcountry j from all industrial countries (see Appendix A). We also presentsimulation results in which the foreign R&D capital is a simple average of

1]industrial countries’ domestic R&D capital stocks (i.e. where a 5a5 for allji 8

j,i). This specification is suitable when the stock of knowledge that results fromR&D investment grows by one dollar for each dollar of R&D, independently ofthe investing country. In this case foreign R&D capital stocks are not trade related.

Investment in MULTIMOD is modeled as a gradual adjustment of the capitalstock towards its optimal level, which is determined by the gap between themarket value of the existing stock and its replacement cost, following Tobin

MAR(1969). The market value of the capital stock (K ), defined as the discountedvalue of future after-tax income accruing to owners of capital, is calculated usingan iterative process in which today’s market value reflects the present value ofafter-tax income for owners of capital (PROFIT),

`

MAR 2i(t 2t )K 5E e PROFIT dt ,t

t

where i is the real interest rate (the model uses, however, a discrete timeformulation). Future increases in profitability or total factor productivity aretranslated into the current market value of the capital stock and hence intoincreases in current investment. The response of the real capital stock (K) tochanges in the market value of capital, however, is estimated to be very slow,

MARDlogK 5 0.08(K /K ) .t21

Investment is derived from the change in the real capital stock plus depreciation.Private consumption is also dependent on future income through a forward-

looking term in wealth, as discussed in the previous section. Some individuals areassumed to be liquidity constrained in the short run, so that real consumption (C)

Ddepends partly on changes in current real disposable income (Y ) and on realLlong-term interest rates (i ) in addition to wealth (W ),

L DDlogC 5 0.095log(W /C ) 2 0.59i 1 0.35DlogY .t21 t21

In the long run, consumption moves proportionately with wealth. The structure of


the regional developing country models is similar except that investment andconsumption depend on imports as well as the factors discussed above. Thesecountries are also assumed to face external finance constraints and greater

10domestic liquidity constraints.Long-term interest rates are a moving average of current and expected future

short-term rates. Financial assets of the industrial countries are assumed to beperfect substitutes, and nominal exchange rates for the industrial and newlyindustrializing economies are determined by open interest parity. Each regionaldeveloping ‘country’ has a freely floating exchange rate, with the market ratedetermined by the external financing constraint rather than by international assetarbitrage. Devaluations always improve the current account, and appreciationsalways worsen it, so the system is stable, i.e. the Marshall-Lerner conditions aresatisfied.

Exports and imports are mainly determined by relative prices and activity in allof the models. Export prices are assumed to move with the domestic output pricein the long run, but respond to price movements in export markets in the short run.Import prices are a weighted average of the export prices of trade partners. Theindustrial countries and the NIEs produce manufactured goods, which areimperfect substitutes. Each country’s or region’s imports of manufactured goodsare allocated as exports across the other manufactures-producing countries andregions through a trade matrix, with the initial pattern based on historic tradingpatterns. Trade shares adjust to changes in relative prices. Non-oil primarycommodities are produced by the developing countries, who also producemanufactured goods. The average price of non-oil primary commodities adjusts inthe short run to clear the market, with production and supply eventuallyresponding to changes in relative prices.

In the next section, we use this augmented model to trace out the long-runeffects of exogenous increases in research and development expenditures and tradeon total factor productivity and potential output. While these simulations highlightthe new features of the model that are the focus of this paper, endogenizing totalfactor productivity also has implications for more standard policy simulations.This is particularly true for fiscal policy simulations that, as previously noted, havelong-run impacts on the level of the real capital stock — through crowding out andcrowding in effects — and output in the standard version of MULTIMOD. In ouraugmented model, the assumption that R&D expenditures remain constant relativeto GDP results in an endogenous response of total factor productivity that tends tomagnify the long-run effects of fiscal policy on potential output. The long-run risein potential output of a sustained reduction in real government spending in all

10See Bayoumi et al. (1995).


industrial countries, for example, is more than twice as large in our augmented11version of MULTIMOD compared with the standard version.

4. Simulation results

We focus on three types of simulations to illustrate the empirical significance ofinternational R&D spillovers: an increase in R&D expenditures in individual G-7countries, a simultaneous increase in R&D expenditures in all industrial countries,and increased openness in the developing countries. In each case, we mainly focuson the long-run effects. The simulation results are largely independent of thebaseline, which is taken from the IMF’s (International Monetary Fund, 1995)World Economic Outlook projections to the year 2000 extended such that each

12country slowly moves to a steady state by the year 2075. In each simulation, realgovernment expenditure is assumed to remain constant relative to potential GDP,with tax rates varying to achieve a pre-specified path for government debt. Otherassumptions, such as keeping real government expenditures constant, impliedimplausible secular trends in the size of government that made it difficult to solvethe model. Finally, the money supply is also kept proportional to potential GDP,which leaves the price level broadly unchanged. As the model is neutral in thelong-run with respect to nominal shocks, this assumption has no impact on theresults.

Before discussing the R&D simulations, we need to address the accountingissue of where R&D expenditures fit into the model. In the early 1990s, about 50%of business sector R&D expenditures were labor costs, 40% were other current

13expenditures, and 10% were capital expenditures. In the simulations discussedbelow, the increases in R&D expenditures are assumed to raise businessconsumption, an element of aggregate demand introduced into the model for these

11If government consumption is changed in only one country, the differences between the twoversions of the model is much less pronounced because there are offsetting effects on total factorproductivity from the endogenous change in R&D and in imports.

12The simulated shocks are assumed to be expected, and variables representing expectations areconsistent with the model’s predictions. The model, which is considerably more difficult to solvenumerically than the standard version of MULTIMOD, was solved using the NEW STACK option inportable TROLL; see Juillard and Laxton (1996).

13These estimates are from OECD (1995a) and refer to the average of the G-7 countries other thanthe United States (for which a breakdown is not available). Only R&D capital expenditures would beincluded directly as an element of aggregate demand, although these represented less than 1% ofbusiness fixed investment in the early 1990s in the G-7 countries other than the United States. OtherR&D expenditures would affect aggregate demand indirectly through their effects on incomes andproduction.


simulations. The simulated increases in R&D expenditures, which are sustainedthroughout the simulation period, are assumed to be financed out of current andfuture business profits. The reduction in the discounted value of future profitslowers the market value of the physical capital stock and hence reduces physicalinvestment. Thus, enterprises forego fixed investment in order to increase R&Dexpenditures.

The impact of an exogenous sustained increase in US R&D expenditures is1]shown in Fig. 1 and Table 1. The increase in R&D expenditures is equivalent to 2

of 1% of GDP, which represents an increase in the level of real US R&Dexpenditures of about 25% relative to baseline. While an increase of this size is

14large, it is not unprecedented over a span of a few years. Higher R&Dexpenditures boost the future US R&D capital stock above its baseline level. Thebulk of the rise in the R&D capital stock takes place early in the simulation periodas a progressively larger proportion of the higher R&D expenditures are needed toreplace a growing amount of obsolete R&D capital. After 15 years, the R&Dcapital stock has increased by about half its long-run value and by 2075 it hasrisen by almost the full amount of its steady-state increase of about 40%.

The higher R&D capital stock implies an increase in the future level of totalfactor productivity and potential output. The long-run partial equilibrium effectfrom the increase in the domestic R&D capital stock — i.e. the impact without anyinduced effects from changes in investment, foreign R&D capital, trade, or othervariables endogenous to the model — is an increase in US total factor productivity

15and potential output of 7.3% (cf. Table 1, line 3). The general equilibrium effectsmagnify, however, the partial equilibrium effects so that the rise in potential outputis larger than the rise in total factor productivity. The increase in total factorproductivity, for example, increases expected future profits, although the positiveeffect of this increase is offset by the negative effect of the extra costs to firms tofinance the higher level of R&D spending. In the first few years, the negativeeffect of the increased expenditures on R&D dominates, and both the market valueof the capital stock and business fixed investment fall. The boost to aggregatedemand from higher R&D spending and consumption also increases real interestrates, which further reduce investment in the short run. From 2003 onward,however, the discounted benefits from future profits cause both the market value ofthe capital stock and investment to rise sharply. Physical investment increasesrelatively fast for the next 15–20 years and then begins to taper off as the actual

14For example, real business sector R&D expenditures increased 27% in the 3 years to 1984 in theUnited States, and single-year increases of 10% or higher are not uncommon in other industrialcountries (OECD, 1991, 1995b). The model is broadly linear, so the simulated effects of a differentsized shock would be roughly proportional.

15This is simply the coefficient on domestic R&D capital in the equation for total factor productivitymultiplied by the (logarithmic) percentage change in the domestic R&D capital stock (the latter is32.1% rather than the arithmetic 37.8% reported in Table 1).


Fig. 1. Increased R&D in the United States (deviations from baseline, in percent).

capital stock slowly adjusts to the higher level of its market value. In contrast toinvestment, real consumption rises steadily throughout the simulation as forward-looking consumers react to the expected increase in future wealth.


Table 1Increased R&D expenditures in the United States (deviations from baseline, in percent)

1996 2000 2010 2030 2050 2075

United StatesPotential output — 1.6 4.4 7.5 8.9 9.5Total factor productivity — 1.7 4.3 6.4 7.1 7.3

from domestic R&D — 1.7 4.3 6.5 7.3 7.6from foreign R&D — — — 20.1 20.2 20.3

Capital — 20.4 0.4 3.6 5.9 7.2Investment 20.4 20.9 1.9 5.3 6.8 7.5Consumption 0.1 1.4 3.4 5.4 6.5 7.1

aR&D spending/GDP 0.5 0.5 0.5 0.5 0.5 0.5Domestic R&D stock — 7.4 19.8 31.3 35.7 37.8Foreign R&D stock — — 0.2 0.9 1.6 2.4

aManufactures imports /GDP — 20.1 20.6 21.1 21.2 21.2

Other industrial countriesPotential output — 0.2 0.6 1.4 2.1 2.8Total factor productivity — 0.2 0.6 1.1 1.5 1.7

from domestic R&D — — 0.1 0.2 0.4 0.5from foreign R&D — 0.2 0.6 0.9 1.1 1.2

Capital — 20.1 20.1 0.9 2.0 3.0Investment 20.1 20.4 0.3 1.7 2.7 3.6Consumption — 0.4 1.2 2.2 2.9 3.4

aR&D spending/GDP — — — — — —Domestic R&D stock — — 0.2 0.8 1.6 2.3Foreign R&D stock — 4.7 12.5 19.7 22.7 24.2

aManufactures imports /GDP — — 0.3 0.5 0.5 0.5

Developing countriesPotential output 0.1 0.3 1.1 2.3 2.9 3.4Total factor productivity — 0.5 1.2 1.9 2.1 2.3

from foreign R&D — 0.3 0.8 1.3 1.5 1.6from trade — — — — — 0.1

Capital — 20.2 0.3 2.2 3.3 4.2Investment 0.1 20.6 1.3 3.1 3.8 4.5Consumption — 0.3 1.0 2.2 2.8 3.3Foreign R&D stock — 4.0 10.4 16.2 18.7 20.1

aManufactures imports /GDP 0.2 — 0.2 0.3 0.3 0.4

1]R&D expenditures are exogenously increased by an amount equivalent to of 1% of the baseline level2

of GDP in 1996, with the R&D/GDP ratio maintained constant thereafter.aIn percentage points.

Taking account of all the general equilibrium effects embodied in the model, thelevel of potential output in the United States is about 4.5% higher in 2010compared with the baseline, and 9.5% higher in 2075. The long-run impact of allthe induced, general equilibrium effects is to increase potential output by 2


percentage points above the partial equilibrium effects stemming from just the risein the domestic R&D capital stock. The time pattern of potential output reflectsmainly the simulated paths of the increases in the R&D and physical capitalstocks. During the first 15 years, almost all of the increase in potential output isdue to higher total factor productivity, but by 2075 the rise in the physical capitalstock accounts for about one fourth of the total increase in output. The annualgrowth of real output is more than 0.3 percentage points higher during the first 10years of the simulation compared with the baseline. Growth remains stronger thanin the baseline, although by progressively smaller amounts, throughout the 80years of the simulation. In the last 25 years, for example, potential output growthis only 0.025 percentage points higher than in the baseline. The relatively longtransition period reflects the slow build up of the R&D capital stock and the slowresponse of actual investment to the increase in the market value of the capitalstock. In the long run, the rate of growth returns to the same level as in the

16baseline.The rise in output in the United States relative to the rest of the world requires a

real devaluation of the US dollar to create the needed demand for higher US17exports. This is a standard result from multicountry models, and represents one

channel through which other countries are affected by the higher output in theUnited States. In our model, R&D spillovers represent an additional channel ofinfluence through which other countries benefit from the increase in US R&Dexpenditures. The foreign R&D capital stocks of US industrial and developingcountry trade partners increase 24 and 20%, respectively, by 2075 compared withthe baseline. The increases in the foreign R&D capital stock in specific countriesand regions depend on the relative weight of US imports compared with importsfrom other industrial countries.

Manufactures imports are the vehicle for the R&D spillovers. The higherimports of US industrial country trade partners stemming from the depreciation ofthe dollar magnify the impact on growth from the rise in their foreign R&D capitalstocks. In the United States, on the other hand, manufactures imports as a share ofGDP decline somewhat with the depreciation of the dollar, which offsets thebeneficial spillovers from the higher level of foreign R&D capital; the rise in R&Dinvestment by US trade partners reflects the assumption that R&D expendituresremain constant relative to GDP. The assumption that developing countries otherthan the NIEs are finance constrained implies that their manufactures imports

16In simulations assuming a 15% depreciation rate for R&D capital, growth stabilizes at the baselinelevel by about 2050.

17See, for example, Bryant et al. (1988). This result, which stems from the absence of a distinctionbetween traded and non-traded goods in the model, takes no account of the Balassa-Samuelson effect inwhich differences in productivity growth between traded and non-traded goods cause the exchange rateto appreciate as countries become relatively more wealthy.


relative to GDP remain broadly unchanged from baseline levels. A simulationillustrating how increased openness boosts R&D spillovers to the developingcountries is discussed below.

The rise in foreign R&D capital interacted with the import share boosts totalfactor productivity, investment, and potential output in US trade partners in muchthe same way that the rise in domestic R&D did in the United States. Potentialoutput increases gradually, again slowing after 15–20 years. By 2075, potentialoutput is 2.8% above its baseline level in the other industrial countries, and it is3.4% higher in the developing countries. The long-run impacts of higher R&Dexpenditures in the United States on potential output in individual countries andgroups of countries are shown in the first column of the top panel of Table 2.Canada, the newly industrializing economies of Asia, and the developing countriesof the Western Hemisphere benefit most from higher R&D expenditures in theUnited States, reflecting strong trade linkages.

The increase in output is an important summary measure of the overalleconomic impact of R&D expenditures. Economic welfare, however, largelydepends on real private consumption. In the United States, private consumption is7% above baseline by 2075, considerably less than the 9.5% increase in output.The opposite occurs for the other countries and regions, as shown in the firstcolumn of the lower panel of Table 2. By 2075, the average percentage increase ofconsumption in other industrial countries and in developing country regions otherthan the NIEs is 1.2 times that of output (the result for the NIEs is discussedbelow). This compression of the variability of consumption responses comparedwith those for output reflects the reduction in the US terms of trade caused by theneed to find markets for new goods, and constitutes an important channel thoughwhich the benefits of R&D in one country are disseminated to its trading partners.

The impact of higher R&D expenditures in the United States on consumption inCanada and the developing countries of the Western Hemisphere, both of whichare close trading partners with the United States, is particularly large. Indeed,

18Canadian consumption increases by as much as in the United States. The newlyindustrializing economies is the only region in which the long-run increase inconsumption is smaller than the increase in output. This reflects, at least in part,their trilateral trading arrangements as net importers from Japan and net exportersto the United States. Consumption is lowered by the negative terms of trade shockin the NIEs caused by the depreciation of the dollar against the yen.

The domestic marginal product of R&D capital — or, equivalently, the rate ofreturn on R&D capital — can be calculated in a similar manner to the marginalproduct of physical capital, namely by dividing the increase in the potential output,the variable directly affected by both R&D spending and investment, by the

18The results for Canada in this simulation and the simulations reported below also reflect therelatively low Canadian trade elasticities in MULTIMOD, which enlarge the exchange-rate response,and hence the output and consumption spillovers.

T.B

ayoumi

etal.

/Journal

ofInternational

Econom

ics47

(1999)399

–428415

Table 2Long-run international spillovers from increased R&D in industrial countries (deviations from baseline in 2075, in percent)

Increased R&D in:

United States Japan Germany France Italy United Kingdom Canada All Countries

Potential outputAll industrial countries 5.1 3.7 1.2 1.2 1.7 0.9 1.0 17.5

United States 9.5 2.2 0.4 0.3 0.8 0.3 0.9 14.8Japan 2.7 10.8 0.2 0.2 0.1 0.1 0.3 16.4Germany 2.6 1.7 7.2 1.1 1.6 0.6 0.4 19.2France 2.2 1.1 1.2 10.6 1.5 0.5 0.4 20.9Italy 2.3 1.0 1.4 1.2 15.7 0.4 0.4 25.7United Kingdom 2.9 1.4 1.1 0.8 1.2 10.4 0.5 22.2Canada 7.0 1.4 0.2 0.3 0.9 0.3 15.2 27.9Smaller industrial countries 2.5 1.4 1.4 0.8 0.9 0.6 0.4 15.3

All developing countries 3.5 2.9 0.7 0.7 1.1 0.5 0.4 10.6Africa 2.4 1.6 1.1 1.8 1.4 0.9 0.4 11.2NIEs 5.7 7.0 0.7 0.6 1.5 0.6 0.4 16.6Western Hemisphere 3.7 1.1 0.5 0.4 0.7 0.3 0.4 7.6Other developing countries 3.0 3.0 0.8 0.6 1.1 0.6 0.3 10.2

416T.

Bayoum

iet

al./

Journalof

InternationalE

conomics

47(1999)

399–428

Table 2 (continued)

Increased R&D in:


ConsumptionAll industrial countries 4.8 3.4 1.1 1.1 1.5 0.8 1.0 16.1



1]This table reports the results of eight independent simulations where R&D expenditures are exogenously increased by an amount equivalent to of 1% of2

GDP in 1996 in each G-7 country individually and in all industrial countries simultaneously, with the R&D/GDP ratios maintained constant thereafter.


increase in the stock of R&D capital. Calculated in this manner, the marginalproduct of US R&D is about 50 cents on the dollar in the short run — in otherwords, each dollar increase in the US R&D capital stock raises real US output byaround 50 cents in the following year. By comparison, the short-run marginalreturn on physical capital in the simulations is around 10 cents on the dollar (bothcapital stocks have an assumed annual depreciation rate of 5%). The highermarginal return on R&D capital compared with physical capital reflects thatindividual enterprises do not capture all of the benefits from R&D spending, which

19produces a large gap between the private and social rates of return. The marginalproduct of US R&D spending for US output rises over time to almost one dollarfor each extra dollar of R&D capital stock by the end of the simulation period,reflecting the positive feedback through induced higher investment. Rates of returnwill be higher for other countries, because of their lower ratios of R&D capital

20stock to output. For example, Coe and Helpman (1995), using a similarspecification, estimate the long-run average marginal rate of return of a dollar ofR&D capital across all G-7 countries to be $1.23.

The same approach can be used to quantify the spillovers from R&Dinvestment. Each extra dollar of the US R&D capital stock generates about 10cents of extra output in other industrial and developing countries in the short run.Over time, these benefits rise to about 60 cents for other industrial countries and25 cents for developing nations. Thus the global rates of return from the increasein US R&D expenditures, i.e. the domestic rate of return plus all of the spillovereffects, are about 70 cents in the short run and about $1.85 in the long run. Theseestimated rates of return to R&D are high, but not outside of the range of resultsfrom the empirical literature surveyed by Nadiri (1993).

Higher R&D expenditures in any of the other major industrial countries havebroadly similar effects as higher expenditures in the United States. Table 2 showsthe long-run effects on potential output and consumption in simulations in which

1]R&D expenditures are exogenously increased by an amount equivalent to of 1%2

of GDP in each G-7 country. Compared with the US simulation, the maindifferences are that the domestic effects are often larger while the internationalspillovers are smaller. The larger domestic effects reflect the smaller R&D capitalstocks relative to GDP in these countries, and hence the larger percentage increase

1]from raising R&D by a uniform of 1% of baseline GDP — the long-run increase2

in R&D capital in Canada, for example, is about 100% compared with 40% in the

19An alternative method of calculating the rate of return is to divide the net present value of theincrease in US GDP (using the US real interest rate) by the net present value of the increase in USR&D spending. Using a real rate of 4%, this calculation implies a ratio of 10 to 1 for the present valueof the increase in US GDP to the present value of the increase in US R&D spending. The ratio rises to14 when the benefits to the other industrial countries are accounted for, and to 16 when the benefits tothe developing countries are included.

20The rate of return is defined as the estimated elasticity of total factor productivity with respect toR&D capital multiplied by the ratio of output to R&D capital.


21United States. The spillover effects from R&D in countries other than the UnitedStates are smaller since the size of the simulated increase in R&D expenditures aresmaller (reflecting the lower level of GDP) and since the United States typicallyaccounts for the largest share of other countries’ foreign R&D capital stocks. Theregional distribution of the spillovers also differs, reflecting different bilateral tradepatterns. Higher R&D expenditures in Japan, for example, have a relatively largerimpact on the NIEs and other countries in Asia (included in the other developingcountries region), while increased R&D expenditures in France have a relativelylarger impact in other European countries and in Africa.

Similar spillover patterns are apparent for consumption, as shown in the lowerpanel of Table 2. Unlike the output responses, the domestic gains to consumptionfrom a rise in R&D are smaller for the more open European countries than for theUnited States and Japan, reflecting the greater potency of the terms of trade effect.The importance of trade linkages in determining the long-run rise in consumptioncan also be seen in the large positive consumption spillovers that increases inR&D in European industrial countries have on other countries in the region. Thesespillovers also depend on the magnitude of the trade elasticities for individualcountries, which partly determines the size of the required change in the terms oftrade. This helps explain, for example, the larger consumption spillovers for Italyand Canada than for Germany.

The impact of a simultaneous, exogenous increase in R&D expenditures in all1]industrial countries equivalent to of 1% of GDP is shown in Fig. 2, the last2

column of Table 2, and Table 3. Domestic and foreign R&D capital stocksincrease almost 50% in all countries and groups of countries by 2075. Potentialoutput is 17.5% above baseline by 2075 in the industrial countries as a group and10.6% higher in developing countries. In both cases, higher total factor productivi-ty accounts for roughly three quarters of the increase in output. Private consump-tion rises by an average of 16% above baseline in the industrial countries, with theincrease in European countries being somewhat higher and in North Americasomewhat lower. Consumption in the developing country regions increases by13% on average, with the NIEs gaining the most and the Western Hemisphere theleast.

Keller (1997) examined estimates of the Coe and Helpman (1995) specificationusing random bilateral import shares rather than true bilateral import shares.Treating this exercise as a Monte Carlo study, he finds that in the majority of casesequations with random bilateral import shares perform at least as well as the

21The relatively large domestic effects for Italy and Canada also reflect that they are G-7 countries,where the elasticity of total factor productivity with respect to domestic R&D capital is estimated to beconsiderably larger than in the smaller industrial countries; cf. Coe and Helpman (1995). In Canada, forexample, if the smaller industrial country elasticity were used instead of the G-7 elasticity, the long-runpartial equilibrium effect on potential output from the increase in the domestic R&D capital stockwould be reduced from 16% to about 5%.


Fig. 2. Increased R&D in industrial countries (deviations from baseline, in percent).

equations with true bilateral import shares when judged by the fraction of theexplained variance. In all cases, however, the share of imports in GDP has asignificant effect on TFP. He concludes that while imports as a fraction of GDP


Table 3Increased R&D in all industrial countries (deviations from baseline, in percent)

1996 2000 2010 2030 2050 2075

All industrial countriesPotential output — 2.5 6.8 12.3 15.5 17.5Total factor productivity 0.2 3.0 7.2 10.8 12.3 13.2

from domestic R&D — 2.2 5.6 8.6 10.0 10.8from foreign R&D 0.2 0.8 1.6 2.1 2.3 2.4

Capital — 20.4 0.7 6.6 11.5 14.7Investment 20.3 21.5 3.6 10.5 14.1 15.4Consumption 0.2 2.6 6.5 11.1 14.2 16.1

aR&D spending/GDP 0.5 0.5 0.5 0.5 0.5 0.5Domestic R&D stock — 9.3 24.0 37.7 43.9 47.7Foreign R&D stock — 9.3 24.1 37.8 43.8 47.5

aManufactures imports /GDP 0.2 0.3 0.3 0.3 0.2 0.2

All developing countriesPotential output 0.1 1.1 3.2 6.8 8.9 10.6Total factor productivity 0.1 1.3 3.3 5.5 6.6 7.6

from foreign R&D — 1.2 3.1 5.0 6.1 7.0from trade 0.1 0.1 0.2 0.4 0.5 0.6

Capital — 20.1 1.2 6.6 10.3 12.6Investment 0.5 20.5 3.7 9.3 11.9 13.0Consumption 0.5 1.6 4.2 8.4 11.0 13.0Foreign R&D stock — 8.4 21.7 34.0 39.8 43.4


1]R&D expenditures are exogenously and simultaneously increased by an amount equivalent to of 1%2

of the baseline level of GDP in 1996 in all industrial countries, with the R&D/GDP ratios maintainedconstant thereafter.aIn percentage points.

have an important effect on TFP, bilateral trade shares do not. As it happens, theway his random samples of bilateral import shares are constructed causes thefrequency of such vectors to be concentrated around equal shares for all tradepartners, which is very similar to an equation in which the foreign R&D capitalstock is defined as equally weighted R&D capital stocks of the trading partners.

To examine the sensitivity of our results to the role of bilateral trade shares, wereport in Table 4 a set of simulations using estimates from a model in which theforeign R&D capital stock is an unweighted average of R&D capital stocks oftrade partners. Comparing Table 4 to the upper part of Table 2 indicates that thelong-run effects on the industrial or the developing countries in aggregate of a risein R&D investment in any single country are not particularly sensitive to whetherforeign R&D capital stocks are trade weighted or not, although the equal weightsmodel does alter the distribution of spillovers among the different countriessomewhat. The simulation results from increased R&D in all industrial countriesare virtually identical in either version of the model.

T.B

ayoumi

etal.

/Journal

ofInternational

Econom

ics47

(1999)399

–428421

Table 4Long-run international spillovers from increased R&D in industrial countries: equal weights model (deviations from baseline in 2075, in percent)

Increased R&D in:


Potential outputAll industrial countries 5.1 3.7 1.2 1.3 1.7 1.0 1.1 17.4



This table reports simulation results using a model in which the foreign R&D capital stocks are an unweighted average of the domestic R&D capital stocks in1]the industrial countries. The results are from eight independent simulations where R&D expenditures are exogenously increased by an amount equivalent to 2

of 1% of GDP in 1996 in each G-7 country individually and in all industrial countries simultaneously, with the R&D/GDP ratios maintained constantthereafter.


Fig. 3. Increased trade in developing countries (deviations from baseline, in percent).

Trade has played a relatively minor role in the simulations discussed thus far.This is partly because the developing countries are generally assumed to be financeconstrained, implying that their current accounts cannot change very much from


Table 5Increased trade in developing countries (deviations from baseline, in percent)

1996 2000 2010 2030 2050 2075

Developing countries except NIEsPotential output 1.6 2.0 3.0 4.6 5.6 6.5Total factor productivity 1.3 1.5 2.0 2.9 3.6 4.5

from foreign R&D 0.1 0.4 1.0 2.1 3.0 4.0from trade 1.2 1.1 1.0 0.8 0.7 0.5

Capital 0.4 1.2 2.9 5.5 6.5 6.9Investment 4.6 3.2 4.8 6.3 6.8 6.8Consumption 3.1 2.4 2.1 3.3 4.2 5.2


AfricaPotential output 0.3 0.5 1.6 3.6 4.8 5.9Total factor productivity 0.2 0.5 1.2 2.1 2.9 3.9

from foreign R&D 0.1 0.4 1.1 2.1 3.1 4.2from trade 0.1 0.1 0.1 — 20.2 20.3

Capital 0.2 — 1.4 5.1 6.4 7.0Investment 3.5 20.5 3.9 6.3 6.9 7.1Consumption 4.1 2.0 1.0 2.6 3.6 4.7


Western HemispherePotential output 2.8 3.4 4.5 5.9 7.0 8.2Total factor productivity 2.6 2.6 2.9 3.8 4.6 5.5




Other developing countriesPotential output 1.3 1.7 2.6 4.1 5.0 5.8Total factor productivity 1.2 1.5 2.0 2.7 3.4 4.2




Imports and exports of manufactures are exogenously increased by an amount equivalent to 5% of thebaseline level of GDP in each developing country region except for the NIEs.aIn percentage points.

the baseline levels. The importance of the financing assumption can be gaugedfrom the previous simulations by comparing the impact on the NIEs, which are notassumed to be finance constrained, with the other developing country regions since


the assumed relationship between R&D spending, imports, and growth is the samefor all developing country groups. In the simulation where R&D spending in all

1]industrial countries were increased by of 1% of GDP, for example, the long-run2

rise in the level of output in the NIEs is 16.5% compared with 7.5–11% in theother developing country regions (see Table 2). Although not all of this differenceis due to the financing assumption, since the weights used to calculate the foreignR&D stock and the import shares vary by region, the comparison indicates thatfinancing constraints may reduce significantly the beneficial effects from R&Dspillovers.

The simulation reported in Fig. 3 and Table 5 highlights the role of trade as avehicle for R&D spillovers. In this simulation, the African, Western Hemisphere,and other developing countries region are assumed to adopt more outward orienteddevelopment strategies that have proved so successful for the NIEs. This isimplemented by exogenously increasing imports and exports of manufactures by 5percentage points of baseline GDP. Higher imports of manufactures raisesproductivity in developing countries both directly and through the interactionbetween trade and the unchanged stock of foreign R&D capital. The direct effectfalls slightly over time: as output rises, the external finance constraint results in areal exchange rate depreciation that causes the ratio of real imports to GDP to fallover time. The beneficial effects from increased exposure to foreign R&D capital,however, outweighs this, and total factor productivity for the developing countriesas a whole, which jumps by 1.3% at the start of the simulation, increases steadilyto 4.5% above baseline by 2075. It is important to note that these beneficial effectsstem completely from increased exposure to foreign technology, and not from ahigher level of foreign R&D capital, which is in fact unchanged from the baselinelevel. As in the earlier simulations, higher investment further boosts potentialoutput, which is 6.5% higher by 2075. Consumption, however, only rises by 5.2%because of the adverse impact of the deterioration in the terms of trade. Among thedeveloping countries, the Western Hemisphere benefits the most from the higherlevel of trade, reflecting its close ties with the United States where the stock ofR&D capital is relatively large.

5. Conclusions

This paper has explored the quantitative implications of R&D spending,technological advance, and trade in a world with endogenous growth. We do thisthrough simulations of a special version of MULTIMOD in which total factorproductivity is endogenously determined by R&D spending, R&D spillovers, andtrade.

The simulation results illustrate several features about the gains from R&D.Increases in R&D spending can significantly raise the level of domestic output in

1]an economy. An increase in US R&D investment equivalent to of 1% of GDP,2


for example, raises US real output by over 9% in the long run, with about threequarters of this gain coming though increases in productivity and the remainderfrom higher investment. Half of these output gains occurs during the first 15 years.Over a period of a decade or two, therefore, sustained increases in R&D of thisorder of magnitude can generate a significant boost to the rate of growth of theeconomy. This boost to economic growth, however, eventually peters out — in ourmodel this takes about 80 years — leaving long-run potential output growthunchanged.

Domestic R&D spending can also generate significant spillovers to output inother countries. When all industrial countries raise R&D spending by an amount

1]equivalent to of 1% of GDP, the long-run US output gain is 50% higher than in2

the case when only US R&D spending rises. In our preferred model, the size ofoutput spillovers depends on trade linkages between countries, but we have seenthat the long-run results are not much different when foreign R&D capital stocksare defined with equal weights for all trade partners.

Real consumption rises by less than output in the country carrying out the R&D,while it rises by more than output in other countries. This is because the countrywith higher R&D experiences a deterioration in its terms of trade, which representsan important mechanism through which the benefits of higher domestic R&Dspending are disseminated abroad. As a result, the long-run gain to US consump-

1]tion from an increase in R&D equivalent to of 1% of GDP in all industrial2

countries is more than double that when only US R&D is increased (15% vs. 7%).The size of these consumption spillovers increases with the openness of theeconomy, and particularly benefits close trading partners.

The rate of return calculations provide an additional way to gauge theimportance of R&D spending. An important result is that the short-run rate ofreturn on R&D is about five times as high as the rate of return on physical capital,based on the US calculations discussed above. This is within the range of availableestimates. But our dynamic model also indicates that the long-run benefits of R&Dinvestment are even larger, and that substantial benefits are also conferred on othercountries.

Finally, open trading policies of the type followed by the NIEs can benefitdeveloping nations through facilitating technology transfer from industrial coun-tries. Expanding imports of manufactures in developing countries other than theNIEs by 5 percentage points of GDP — roughly equivalent to the increase that hasoccurred in these regions between 1992 and 1995 — raises output by about 6.5%in the long run, and consumption by 5.2%. These results indicate that part of thesuccess of the NIEs over the last 20 years can be attributed to productivity growthstemming from foreign R&D spillovers through trade. Other factors that haveboosted growth in these countries include rapid increases in labor and capital input(Young, 1995).

These results, of course, reflect the specification and parameters of the totalfactor productivity equations as well as the other simulation properties of


MULTIMOD. Our specification focuses on technology and trade as the keydeterminants of total factor productivity, which is not to deny the importance ofother factors such as investment in human capital. As noted above, our elasticitiesof total factor productivity with respect to domestic and foreign R&D capital arebroadly consistent with those found in the empirical literature. This literature is,however, relatively recent, and further research will undoubtably improve ourunderstanding of the sizes of the parameter estimates used here and the mecha-nisms through which R&D and trade affect total factor productivity. The resultsalso reflect the size of the simulated increases in R&D expenditures and trade,which are relatively large, although not inconsistent with recent experience.Conveniently, the model is broadly linear, so alternative values of the keyelasticities or alternative sizes of the simulated shocks would, to a first approxi-mation, change the simulation results proportionately.

This paper is a first attempt to assess empirically the contribution of R&Dspending, international technology transfer, and trade in the context of a fullyspecified world econometric model. It is also, to the best of our knowledge, thefirst to incorporate aspects of endogenous growth models into a multicountryeconometric model (Helliwell, 1995). Both of these are fertile areas for futureresearch. Notwithstanding their preliminary nature, our results suggest that R&Dlinkages and trade are important factors promoting world economic growth.

Acknowledgements

Elhanan Helpman thanks the NSF and US-Israel BSF for financial support. Wethank Charles Bean, John Helliwell, Alexander Hoffmaister, Per Krusell, DouglasLaxton, Paul Masson, and Steven Symansky for comments on an earlier version ofthe paper; and Toh Kuan and Susanna Mursula for research assistance. The viewsexpressed in this paper are those of the authors and do not represent the views ofthe International Monetary Fund or any other institution.


Appendix A

Bilateral import shares for manufactures (average, 1970–1990)

Imports from:

US JA GR FR IT UK CA SI

Imports of:United States — 0.32 0.09 0.05 0.04 0.06 0.30 0.14Japan 0.46 — 0.08 0.05 0.04 0.04 0.08 0.25Germany 0.09 0.08 — 0.15 0.11 0.09 0.01 0.47France 0.10 0.05 0.24 — 0.14 0.09 0.01 0.37Italy 0.07 0.03 0.28 0.20 — 0.06 0.01 0.35United Kingdom 0.13 0.07 0.19 0.11 0.07 — 0.02 0.41Canada 0.76 0.08 0.03 0.02 0.02 0.04 — 0.05Smaller industrial 0.12 0.09 0.23 0.10 0.08 0.09 0.01 0.28

countriesAfrica 0.09 0.11 0.15 0.29 0.09 0.14 0.01 0.14NIEs 0.28 0.51 0.07 0.03 0.02 0.04 0.01 0.06Western Hemisphere 0.48 0.13 0.12 0.06 0.05 0.04 0.02 0.10Other developing 0.18 0.35 0.16 0.07 0.05 0.09 0.01 0.10

countries

Imports of manufactures of each row country from each of the seven column countries and the smallindustrial countries as a group as a share of total imports of manufactures from these countries. Eachrow sums 1.0.

References

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

Bartolini, L., Razin, A., Symansky, S., 1995. G-7 fiscal restructuring in the 1990s: macroeconomiceffects. Economic Policy 20, 109–146.

Bayoumi, T., Hewitt, D., Symansky, S., 1995. MULTIMOD simulations of the effect on developingcountries of decreasing military spending. In: Currie, D.,Vines, D. (Eds.), North-South Linkages andInternational Macroeconomic Policy. Cambridge University Press, Cambridge.

Blanchard, O.J., 1985. Debt, deficits and finite horizons. Journal of Political Economy 93, 223–247.Bryant, R.C., Henderson, D.W., Holtham, G., Hooper, P., Symansky, S. (Eds.), 1988. Empirical

Macroeconomics for Interdependent Economies. Brookings Institution, Washington, DC.Cass, D., 1965. Optimum growth in an aggregative model of capital accumulation. Review of

Economic Studies 32, 233–240.Coe, D.T., Helpman, E., 1995. International R&D spillovers. European Economic Review 39, 859–887.Coe, D.T., Helpman, E., Hoffmaister, A.W., 1997. North-south R&D spillovers. The Economic Journal

107, 134–149.Eaton, J., Kortum, S., 1996. Trade in ideas: patenting and productivity in the OECD. Journal of

International Economics 40, 251–278.Gagnon, J.E., Masson, P.R., McKibbin, W.J., 1996. German unification: what have we learned from

multi-country models? Economic Modeling 13, 467–498.


Griliches, Z., 1988. Productivity puzzles and R&D: another nonexplanation. Journal of EconomicPerspectives 2, 9–21.

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

Helliwell, J.F., 1995. Modeling the supply side: what are the lessons from recent research on growthand globalization? Paper presented to the meeting of Project Link.

Juillard, M., Laxton, D., 1996. A robust and efficient method for solving nonlinear multicountryrational expectations models. IMF Working Paper no. 96/106.

International Monetary Fund, 1995. World Economic Outlook. IMF, Washington.Keller, W., 1997. Are international R&D spillovers trade-related? Analyzing spillovers among randomly

matched trade partners. European Economic Review (in press).Lichtenberg, F.R., van Pottelsberghe de la Potterie, B., 1997. International R&D spillovers: a comment.

European Economic Review (in press).Masson, P., Symansky, S., Meredith, G., 1990. MULTIMOD Mark II: A revised and extended model.

IMF Occasional Paper 71.Mohnen, P., 1994. The econometric approach to R&D externalities. Cahiers de recherche du

´ ´ ´ `departement des sciences economiques de l’UQAM No. 9408, Universite du Quebec a Montreal.Nadiri, M.I., 1993. Innovations and technological spillovers. NBER Working Paper No. 4423.Organization for Economic Cooperation and Development, 1991. Basic Science and Technology

Statistics. OECD, Paris.Organization for Economic Cooperation and Development, 1995a. Basic Science and Technology

Statistics. OECD, Paris.Organization for Economic Cooperation and Development, 1995b. Main Science and Technology

Indicators. OECD, Paris.Romer, P.M., 1990. Endogenous technological change. Journal of Political Economy 98, S71–S102.Solow, R.M., 1956. A contribution to the theory of economic growth. The Quarterly Journal of

Economics 70, 65–94.Tobin, J., 1969. A general equilibrium approach to monetary theory. Journal of Money, Credit and

Banking 1, 15–29.Young, A., 1995. The tyranny of numbers: confronting the statistical realities of the east Asian growth

experience. The Quarterly Journal of Economics 110, 641–680.

Documents

Bayoumia Coe Helpman (1999) R&D Spillovers and Global Growth