International and intersectoral R&D spillovers in the OECD and East Asian economies.

Park, Jungsoo

This study empirically explores international and intersectoral R&D spillover effects on the total factor productivity growth of manufacturing and nonmanufacturing sectors based on a pooled time-series data set of 14 OECD economies and 3 East Asian economies--Korea, Singapore, and Taiwan. The study finds that foreign manufacturing R&D has strong influence on domestic productivity growths of both sectors and that domestic manufacturing R&D has a substantial intersectoral R&D spillover effect on domestic nonmanufacturing productivity growth. The social rates of return to manufacturing R&D are estimated to be two to six times greater than the private rates of return. (JEL D24, 033, F10)

I. INTRODUCTION

Research and development (R&D) activity, which provides a key to success in productivity competition, has been performed disproportionately across sectors and across economies. This observation may reflect the nature of the production technology of each sector and the needs of each country at its respective developmental stage. As discussed in the R&D spillover literature, if there are linkages through the use of technology-embodied intermediate goods or through other transmission mechanism across sectors and across economies, R&D investments in one sector of a country could bring about productivity rise not only in its performing sector but also in other sectors and in other countries.

The R&D activities in the Organisation for Economic Co-operation and Development (OECD) and East Asian economies have been heavily concentrated in the manufacturing sectors in the past two decades, even though the share of the manufacturing sector is relatively small in comparison with the rest of the aggregate economy in these countries. (1) One intriguing issue is to identify the magnitude of influence of manufacturing R&D on the productivity of the remaining nonmanufacturing sector as well as its influence on the productivity of other countries. This study addresses this and other related issues in a two-sector empirical model of international and intersectoral R&D spillovers on the total factor productivity (TFP) of manufacturing and non-manufacturing sectors based on the pooled time-series data set of the 14 OECD economies and three East Asian newly industrialized economies (NIEs) over the period 1980 to 1995. (2) The two different groups of countries in the sample--the OECD and the East Asian economies--increase the variations in the cross-section. Given the recent success and substantial increase in the indigenous R&D investments in the manufacturing sectors of the East Asian economies, this study investigates further whether the international R&D spillovers arise from these countries as well.

Most empirical studies on international R&D spillovers have focused on R&D effects across business sectors of OECD economies, including the pioneering work of Coe and Helpman (1995). (3) Few empirical studies exist that allow for simultaneous presence of international and intersectoral R&D spillovers, and they all focus on spillovers across disaggregate manufacturing sectors within and among OECD countries. Keller (2002) estimates international and intersectoral R&D spillovers using industry-level data of 13 manufacturing industries in 8 major countries, where his findings show significant presence of R&D spillovers, both domestically and internationally. (4) Using a panel cointegration estimation procedure and using transaction intensity and patent space to construct R&D spillover variables, Frantzen (2002) confirms the findings of Keller (2002). This study complements the literature on international and intersectoral R&D spillover effects by considering the impact of international and intersectoral R&D spillovers between manufacturing and nonmanufacturing sector productivity, an issue that was never addressed in previous studies. The study examines empirical evidence on the existence of productivity link, which may lead economy to a more balanced growth in productivity across the two sectors.

Studies on international R&D spillovers to East Asian economies are very limited. Coe et al. (1997) find substantial R&D spillovers from industrial countries to developing economies using a data set of 77 developing countries where foreign R&D capital stocks were constructed based on OECD R&D investments. Madden et al. (2001) corrects for the potential upward bias of spillover effects in the analysis of Coe et al. (1997) by including domestic R&D for selected East Asian economies. They find that the elasticity of productivity with respect to domestic R&D is higher in the East Asian economies and that the signs and magnitudes of country-specific international spillover effects are mixed across countries without any group patterns. This study adds to the literature on international R&D spillover effect in East Asian economies by simultaneously considering international and intersectoral spillover effects.

Empirical results in this study indicate a stark asymmetry in the intersectoral R&D spillovers between manufacturing and nonmanufacturing sectors. Manufacturing R&D is shown to have a strong intersectoral R&D spillover effect on nonmanufacturing TFP, whereas nonmanufacturing R&D investments do not have any significant impact on the cross-sector productivity. The results on international R&D spillovers in manufacturing sectors are consistent with the earlier findings that there exists a statistically significant international R&D spillover effect when trade is considered as a transmission mechanism. However, the study finds that there exists an asymmetry in the international R&D spillovers, as the R&D capital stocks of the three East Asian NIEs do not influence the productivity of other countries, whereas the R&D capital stocks of the OECD economies do.

The article is organized as follows. In section II, a two-sector production model is presented, where sectoral TFP is linked with domestic and foreign R&D efforts as well as intersectoral R&D activities through the use of intermediate goods. Section III provides the data description. Section IV explains the econometric method in estimating TFP equations to measure the R&D spillover effects in the two sectors. Based on the empirical estimates, section V discusses international and intersectoral R&D spillover effects, TFP elasticities, and private and social rates of returns to R&D. Section VI concludes and draws further implications.

II. MODEL

The model closely follows the theoretical considerations in Grossman and Helpman (1991) and Coe and Helpman (1993), which build on models of innovation-driven growth. The model is modified to include consideration of international and intersectoral spillover effects. An open economy is assumed where there are two sectors: manufacturing (m) and nonmanufacturing (o) sectors. The final output of sector i (i = m or o), [Y.sub.i], is produced by three inputs--physical capital, intermediate inputs, and labor. A Cobb-Douglas production function with constant returns to scale is assumed in both sectors.

(1) [Y.sub.i] [A.sub.i] [K.sup.[[beta].sub.i].sub.i] [L.sup.1- [[beta].sub.i][[eta].sub.i], for i = m or o, [[beta].sub.i] [[eta].sub.i] and [[beta].sub.i] + [[eta].sub.i] < 1,

where [Y.sub.i], [K.sub.i], [D.sub.i], and [L.sub.i] denote the value added, the physical capital stock, an index of intermediate goods, and the total labor hours used in sector i. The technological level of sector i, At, is influenced by the factors contributing to the enhancement of the economy's efficiency and technological environment.

Reorganizing equation (1) and taking the logs gives

(2) [Y.sub.i] [A.sub.i] [K.sup.[[beta].sub.i].sub.i] [([D.sup.i]/[L.sub.i]).sup.[[eta].sup.1-[[beta].sub.i].sub.i];

(3) log [Y.sub.i] = [[beta].sub.i] log [K.sub.i] + (1 - [[beta].sub.i]) * log [L.sub.i] + log [A.sub.i] + [[eta].sub.i] log ([D.sub.i]/[L.sub.i]).

The log of TFP is measured by the following:

(4) log TF[P.sub.i] = log [Y.sub.i] - [[beta].sub.i] * log [K.sub.i]i - (1 - [[beta].sub.i]) * log [L.sub.i] + log [A.sub.i] + [[eta].sub.i] log ([D.sub.i]/[L.sub.i]).

Based on the horizontally differentiated intermediate goods market assumption, the following equations are derived for [D.sub.i]/[L.sub.i], which are from Coe and Helpman (1993) but extended to include intersectoral link consideration.

(5) [D.sub.m]/[L.sub.m] = [F.sub.m]([n.sub.m], [n.sub.o], [X.sub.mm]/[L.sub.m], [X.sub.m]/[L.sub.m] [D.sub.o]/[L.sub.o] = [F.sub.o]([n.sub.m], [n.sub.o], [X.sub.om]/[L.sub.o], [X.sub.oo]/ [L.sub.o]),

where [X.sub.iq] the total units of intermediate goods from the sector q employed in the sector i and the parameters [n.sub.m] and [n.sub.o] are the measures of available manufacturing and nonmanufacturing intermediate inputs. As discussed in Coe and Helpman (1993), [X.sub.iq] can be viewed as the total hours of labor needed to produce the intermediate goods from sector q used in sector i. Given that the input--output linkages are relatively stable, [X.sub.iq]/[L.sub.i]'s are assumed to be constant.

R&D investments contribute in developing new technology of production or in producing intermediate goods that embody the new technology. Through the adoption of newly developed and more efficient intermediate inputs in the production of other sectors, there could be intersectoral R&D spillovers that may not be captured by the innovation rents. In this consideration, the parameters [n.sub.m] and [n.sub.o] in equation (5) are assumed to be functions of cumulative R&D efforts of the respective sector.

Through various channels of diffusion, development of new technology and new technology-embodied intermediate goods could also have positive growth implications in other countries as well. Given a world with international trade, a country may benefit from taking advantage of foreign technical advances through methods ranging from importing technologically advanced goods to adoption of new production processes or technology. As intermediate inputs are traded internationally, [n.sub.m] and [n.sub.o] can be interpreted as the measures of available intermediate inputs that are produced in the world economy. Because not all intermediate inputs in the world are available to all economies due to nontraded goods and barriers to trade across economies, there will be differences in productivity impacts of domestic and foreign R&D as well as sectoral R&D. The following equations are formulated for the two sectors of an economy.

(6) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

where [R.sup.d.sub.m], [R.sup.d.sub.o], [R.sup.f.sub.m], and [R.sup.f.sub.m], enote the domestic manufacturing, domestic nonmanufacturing, foreign manufacturing, and foreign nonmanufacturing R&D stocks, respectively.

Based on the base model and its modified versions, the following questions are asked: (1) Can we find evidence of intersectoral R&D spillovers between manufacturing and non-manufacturing sectors? (2) What is the magnitude of international R&D spillovers in each sector? (3) Do R&D investments of NIEs influence productivity of other countries? (4) What are the social rates of return to sectoral R&D investments considering these spillover effects?

III. DATA DESCRIPTION

The annual time-series data for manufacturing and nonmanufacturing sectors were collected for 14 OECD economies--Australia, Belgium, Canada, Denmark, Finland, France, Germany, Italy, Japan, Netherlands, Norway, Sweden, United Kingdom, and the United States--and 3 East Asian NIEs--Korea, Singapore, and Taiwan from 1980 to 1995. (5) The sample periods and countries were mainly determined by the availability of R&D investment data.

TFP series for each sector were calculated based on the estimation of each sectoral production function. Sectoral production function regressions used in the TFP derivations and the data sources for the related variables are described in the appendix. Because it is fairly difficult to measure the level of successful R&D output, I have taken the expenditures on R&D activity as a proxy for the level of R&D effort and then accumulated real investment series over time to construct R&D capital stock series. Aggregate and manufacturing R&D investment data for the OECD economies were taken from Basic Science and Technology Statistics (OECD) and Basic Science and Technology Indicators (OECD). Aggregate and manufacturing R&D expenditures for the three NIEs were obtained from the local yearbooks, UNESCO Statistical Year-book, and National Science Foundation (1993). Additional sources were used to update the series up to 1995. (6) The current values of R&D expenditures were deflated by gross domestic product (GDP) deflators of each country and converted into 1990 U.S. dollar constant series. (7) Real nonmanufacturing R&D investment series were estimated as the difference between the real aggregate R&D and the real manufacturing R&D investments. Real manufacturing ([R.sup.d.sub.mjt]) and nonmanufacturing ([R.sup.d.sub.ojt]) R&D capital stocks were estimated from each corresponding real R&D investment series for country i in year t. Because the initial R&D stocks ([R.sup.d.sub.ij0]) were not available, they were estimated using the following formula:

(7) [R.sup.d.sub.ij0] = [I.sup.d.sub.ij0]/ (([[gamma].subij] + [g.sub.ij]]),

where [I.sup.d.sub.ij0] is the initial R&D investment for sector i in country j. The R&D investment growth rates ([g.sub.ij]) were calculated using the earliest five years of R&D investment data available and the retirement rates ([[gamma].sub.ij]) are assumed to be 10% (8) Given the estimated initial R&D capital stock, the R&D investments are then accumulated to form the subsequent stock series based on the perpetual inventory method ( [R.sup.d.sub.ij0] = [1 - [[gamma].sub.ij]][R.sup.d.sub.ijt-1] 1 + [R.sup.d.sub.ijt-1] Because the effect of R&D investments may take time to realize, R&D stocks include only up to one-year lag of R&D investments.

Foreign R&D stocks represent the accumulated foreign innovative efforts that potentially influence the productivity of each country. As in the previous studies, international trade is considered as the main channel of R&D spillovers. (9) Foreign R&D stocks in this study are estimated based on the version proposed by Lichtenberg and de la Potterie (1998) as it has the advantage of invariance property regarding the level of data aggregation while capturing the real R&D intensity of other countries embodied in the import flows [R.sup.f.sub.ijt] denotes the foreign R&D stock of sector i for country j which is import-weighted sum of real R&D intensity of other countries.

(8) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

where [v.sup.ij.sub.st] is the sector i imports of country j from country s and [Y.sub.ist] is the total gross output of sector i in country s in year t. The manufacturing bilateral import flows were obtained from Worm Trade Data Base of Hamburg Institute of International Economics, which, in turn, stores UN and OECD statistics. The aggregate bilateral import flows were taken from World Bank Institute's Trade and Production Database. Nonmanufacturing import flows were obtained as the difference of aggregate and manufacturing import flows.

To evaluate the differential international spillover effect arising from the R&D stocks for the two groups of economies, the foreign R&D stock variable solely based on OECD R&D stocks ([R.sup.fg.sub.ijt]) and the variable solely based on three NIE R&D stocks ([R.sup.fh.sub.ijt]) were constructed. Furthermore, to examine whether R&D stocks of NIEs have spillover effects only within their group, foreign R&D stock variable ([R.sup.fp.sub.ijt]) was created based on OECD R&D stocks in the case of the OECD economies, whereas it is based on all other country R&D stocks in the case of the NIEs. Table 1 provides average annual growth rates of TFP and R&D stocks discussed in this section.

IV. ECONOMETRIC METHOD

This study uses a pooled two-sector (manufacturing and nonmanufacturing) annual time-series data set of the 14 OECD economies and 3 East Asian NIEs from 1980 to 1995 to estimate the magnitude of international and intersectoral R&D spillovers on the TFP of manufacturing and nonmanufacturing sectors. Since most time-series exhibit nonstationarity, group mean panel unit root test developed by Im et al. (2003) is performed on the log levels and log differences of TFP, domestic R&D stocks, and alternative specifications of foreign R&D stocks for each sector ([R.sup.f.sub.ijt], [R.sup.fg.sub.ijt], [R.sup.fh.sub.ijt], and [R.sup.fp.sub.ijt]). The Im et al. (2003) test is more general arid powerful than other tests, such as the panel unit root tests of Levin and Lin (1992, 1993), because the former test allows for heterogeneous auto-correlation in each individual time series in the panel for the alternative hypothesis. In Table 2, group mean test statistics ([[bar]]t.sub.nt]) show that one cannot reject the null hypothesis of common unit autoregressive root in the level series for all variables, whereas one can reject the null hypothesis in all the difference series for both sectors at 1% significance level, indicating that the logs of the variables are all first-order integrated.

The regressions in nonstationary level series without testing for cointegration will result in spurious results. This study considers estimation using the stationary log difference series as in Engelbrecht (1997) and in Coe et al. (1997) rather than employing cointegration estimation method pursued by others. (10) The following equations are the baseline empirical equations used in this study based on equation (6).

(9) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCCII.]

(10) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCCII.]

where [[alpha].sub.it] and [[alpha].sub.ij] are, respectively, time-specific and country-specific constants for sector i. The error term [[epsilon].sub.ijt] is assumed to be distributed i.i.d. (0, [[sigma].sub.i]). The method of estimation is ordinary least squares (OLS) method. Because we have a panel of time-series data, a two-way fixed effect model along time and country dimensions is considered to allow for missing country-specific and time-specific factors. Equations (9) and (10) allow estimations of the influence of sector i's domestic R&D on own sectoral TFP ([[beta].sub.i1]), the spillover effect from foreign R&D based on OECD R&D of its own sector ([[beta].sub.i2]), the spillover effect from foreign R&D based on NIE R&D of its own sector ([[beta].sub.i3]), intersectoral R&D spillover effect ([[beta].sub.i4]), the spillover effect from foreign R&D based on OECD R&D of the cross-sector ([[beta].sub.i5]), and the spillover effect from foreign R&D based on NIE R&D of the cross-sector ([[beta].sub.i6]). (11) Cross-equation parameter restrictions on R&D effects are not imposed, which will allow differential R&D effects in the two sectors.

First, to see whether the two different groups of economies--the OECD and the East Asian economies share a common TFP equation for each sector, a series of F-tests are performed on the unrestricted model of the following specification where parameters for the two groups are allowed to differ, by introducing a set of corresponding parameters ([[beta].sup.h.sub.ik], k = 1 ... 6) and a group dummy ([dum.sup.h]) for the three East Asian economies.

(11) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCCII.]

(12) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCCII.]

Hypothesis tests of [[beta].sup.h.sub.ik] = 0 (k = 1 ... 6) are performed to see whether the two groups of economies share common parameters in the TFP equations.

Based on the test results, common parameters are used for the both groups of economies. Various empirical models with alternative definitions of R&D stocks based on equations (9) and (10) are estimated under OLS. For each estimation, each type of fixed effects are included or excluded, depending on the F-test results on each type of fixed effect constants (refer to Coe et al. 1997 and Engelbrecht 1997).

Regressions using stationary log difference variables have advantages because they eliminate econometric issues involving nonstationarity. However, a concern regarding this method is that it increases noise in the data and raises the possible reverse causation from TFP to the respective R&D variables. (12) To alleviate potential endogeneity bias, all R&D stocks are constructed to include only up to their one-year lagged value of R&D investments as discussed in section III. Furthermore, in the sensitivity analysis of section V, instrumental variables (IV) method regressions were performed using lagged values of own sectoral R&D stock, oil prices relative to wheat, and male and female life expectancies as instruments on each sectoral TFP equation to check the robustness of the main results. (13)

Because the TFPs in the two sectors may be influenced by common exogenous factors not considered in the baseline model, the error terms for the two sectors may have contemporaneous correlations. Thus seemingly unrelated regressions (SURs) with and without group-heteroscedasticity adjustments are also considered.

V. EMPIRICAL RESULTS AND IMPLICATIONS

Hypothesis Tests of Common Parameters

Hypothesis test results of common parameters for each variable based on equations (11) and (12) are provided in Table 3. In both sectors, the null hypothesis of all [[beta].sup.h.sub.ik] = 0 (k = 1 ... 6) cannot be rejected as well as the null hypotheses for individual parameter restrictions of [[beta].sup.h.sub.ik] = 0 at a significance level of 1%. (14) The results indicate that the hypothesis of a common TFP equation for the two groups of economies--OECD and East Asian NIEs--cannot be rejected in either sector.

Intersectoral Spillovers

Tables 4 and 5 provide OLS estimates of various specifications of manufacturing and nonmanufacturing TFP equations based on equations (9) and (10), respectively. Each domestic sectoral R&D is a significant factor in its own sectoral TFP equation in all specifications. Intersectoral R&D spillover effects are estimated by introducing the respective cross-sector R&D stocks into the TFP equations. The results show asymmetry of intersectoral spillover effects in the two sectors. Models in Table 4 indicate that domestic nonmanufacturing R&D has no intersectoral influence over domestic manufacturing TFP, whereas models in Table 5 illustrate that there is a strong presence of positive intersectoral spillovers from manufacturing to nonmanufacturing sectors. The result suggests that although there are bilateral industrial linkages between the two sectors through the use of intermediate goods, there exists an asymmetry in the nature of transaction where the manufacturing sector provides relatively more technology-embodied intermediate goods for the nonmanufacturing sector than the converse.

In terms of TFP elasticity in the nonmanufacturing sector, the estimates reveal that cross-sector domestic manufacturing R&D has a smaller impact (0.047 to 0.057) than own sectoral domestic nonmanufacturing R&D (0.071 to 0.087) but higher impact than cross-sector foreign manufacturing R&D (0.041 to 0.045). These estimates are consistent with the intuition that the direct R&D effects are stronger than the indirect spillover effects.

International R&D Spillover

Models in Table 4 show strong evidence of significant international R&D spillovers to manufacturing TFP from the foreign manufacturing R&D stocks created based on R&D stocks of all economies ([R.sup.f.sub.mjt]), based on R&D stocks of all countries but with different definitions for OECD and NIEs ([R.sup.fp.sub.mjt]), and based on OECD R&D stocks only ([R.sup.f.g.sub.mjt]). However, I find that the foreign manufacturing R&D stock created solely from the three NIE R&D stock ([R.sup.fh.sub.mjt]) is found to be not statistically significant in models (iv) and (vi).

The results support the view that the NIEs are lagging behind the technological frontier and the nature of R&D performed in the NIEs may be effective only in increasing NIEs' own productivity or efficiency to approach the frontier, whereas those R&D specific to each NIE's needs may not be useful in nature for the advanced economies, which are closer to the technological frontier. Because model (ii) with [R.sup.fp.sub.mjt] does not improve the goodness of fit over the model (iii) with [R.sup.fg.sub.mjt], it does not provide evidence that NIE R&D spillovers exist among its own group of countries.

This study tests whether foreign manufacturing R&D stocks have cross-sector international R&D spillover effect on non-manufacturing TFP by introducing four alternative specifications of foreign manufacturing R&D stocks ([R.sup.f.sub.mjt], [R.sup.fp.sub.mjt], [R.sup.fg.sub.mjt], and [R.sup.fh.sub.mjt]) in non-manufacturing TFP equations in Table 5. All specifications of foreign manufacturing R&D stocks except the one created solely based on the three NIE R&D stock ([R.sup.fh.sub.mjt]) are statistically significant.

In Table 5, international R&D spillovers to nonmanufacturing sector arising from foreign nonmanufacturing R&D stocks created based on R&D stocks of all economies ([R.sup.f.sub.ojt]), based on R&D stocks of all countries but with different definitions for OECD and NIEs ([R.sup.fp.sub.ojt]), based solely on OECD R&D stocks [R.sup.fg.sub.ojt], and based solely on NIE R&D stocks ([R.sup.fh.sub.ojt]) are considered. However, none of the specifications of foreign nonmanufacturing R&D that were introduced have significant impact on nonmanufacturing TFP.

Table 4 considers potential presence of cross-sector international R&D spillovers on manufacturing TFPs arising from foreign nonmanufacturing R&D stocks by introducing four alternative specifications of foreign nonmanufacturing R&D stocks ([R.sup.f.sub.ojt], [R.sup.fp.sub.ojt], [R.sup.fg.sub.ojt] and [R.sup.fh.sub.ojt]). The results indicate that all alternative specifications of foreign nonmanufacturing R&D considered in this study fail to show statistical significance in influencing manufacturing TFP.

The results from regressions of Tables 4 and 5 consistently indicate that foreign manufacturing R&D has international spillover effects in both sectors, whereas foreign nonmanufacturing R&D has no influence on either sector. The elasticity of the international manufacturing R&D spillover is greater in the manufacturing sector (0.065 to 0.076) than that in the nonmanufacturing sector (0.041 to 0.045), which may be due to stronger linkages in production process and greater degree of assimilation of production technology between manufacturing sectors than between cross-sectors. In comparison with the magnitude of own sectoral R&D effects (0.115 to 0.124 for manufacturing and 0.071 to 0.087 for nonmanufacturing), both international spillover effects are relatively smaller. The results do not show any evidence of international nonmanufacturing R&D spillovers. It may be due to the fact that there are media other than trade that may better proxy the nonmanufacturing R&D spillover intensity. Alternatively, this effect may be absent because the technology improvements in this sector are more focused on process innovation, which may not be easily transferable.

Sensitivity Analysis

As each sectoral TFP and respective sectoral R&D may be determined endogenously, IV regressions were performed using lagged values of own sectoral R&D, oil prices relative to wheat, and male and female life expectancies as instruments on each sectoral TFP equation. The results are presented in models (i) and (ii) of Table 6. In Table 6, SURs are also considered on the equations based on the models (vi) of Table 4 and (vi) of Table 5. Models (iii)-(vi) of Table 6 provide SUR estimates with and without countrywise heteroscedasticity adjustment. The IV and SUR estimates do not differ substantially from the corresponding OLS estimates and consistently confirm the qualitative aspects of the main finding.

Because the R&D effects may take time to materialize, the R&D stocks were constructed to include only up to one-year lagged values of R&D investments. In addition, estimations with lagged values of foreign R&D variables were considered. However, none of the lagged variables had statistically significant positive coefficients (results available on request).

Elasticities and Rates of Return

Given the coefficient estimates of foreign manufacturing R&D in the two TFP equations, the manufacturing and nonmanufacturing TFP elasticities in each country j with respect to manufacturing R&D of individual foreign country s, [e.sub.j,st] (15) are calculated. Tables 7 and 8 provide elasticities by individual source country of R&D based on SUR estimates of (iii) and (iv) in Table 6 given 1995 values of the variables. The U.S. manufacturing R&D stocks generally have the highest impact on all non-European countries, and Germany and France R&D stocks have strong influence on European countries. The United States is highly influenced by R&D stocks of Japan and Canada. It is shown that the NIEs are most heavily influenced by the R&D stocks of Japan and the United States. The relative impact reflects the trade intensity and the relative size of the countries.

Table 9 provides the rates of return to manufacturing and nonmanufacturing R&D stocks by influenced sectors. (16) Column (ii) shows that the rates of return to domestic-manufacturing R&D in domestic manufacturing sector, [delta][Y.sub.mjt]/[delta][R.sup.d.sub.mjt], where [Y.sub.ijt] is GDP of sector i in country j), for the three NIEs are substantially higher (57.2% to 143.7%) than those of the G-5 economies (18.6% to 24.5%).

Columns (iii), (iv), and (v), provide the implied rates of return to manufacturing R&D of country j in domestic nonmanufacturing sector, [delta][Y.sub.ojt]/[delta][R.sup.d.sub.mjt] ( = [[beta].sub.o4] * [Y.sub.ojt/[R.sup.d.sub.mjt]), in all foreign manufacturing sectors, [[SIGMA].sub.s[not equal to]j] [delta][Y.sub.mst]/[delta][R.sup.d.sub.mjt] ( = [[SIGMA].sub.s[not equal to]j] [e.sub.ms,jt] * [[Y.sub.mst]/ [R.sup.d.sub.mjt]), and in all foreign nonmanufacturing sectors [[SIGMA].sub.s[not equal to]j] [[delta][Y.sub.ost]/[[delta] [R.sup.d.sub.mjt] ( = [[SIGMA].sub.s[not equal to]j] [e.sub.os,jt] * [[Y.sub.ost]/[R.sup.d.sub.mjt]. These returns are returns beyond the private returns to manufacturing R&D in column (ii), which may not have been fully appropriated by the manufacturing sector. The column (i) which is the sum of columns (ii), (iii), (iv), and (v) measures the social returns to manufacturing R&D. It is surprising to find that the investments in manufacturing R&D for non-G-5 economies have very high social rates of return ranging from 82.8% to 316.7%. In the G-5 economies, social rates of return (68.0% to 95.4%) are relatively lower but still substantially higher than the private returns. (17)

In both groups of the countries, the returns to manufacturing R&D in the nonmanufacturing sector (shown in column [iii]) are equivalent to or up to four times greater than those in the manufacturing sector (shown in column [ii]). The higher returns reflect the fact that the nonmanufacturing sectors are two to four times larger in size than the manufacturing sector. Productivity enhancement arising from manufacturing to nonmanufacturing R&D spillover effect is magnified by the size of the recipient sector. The returns to OECD manufacturing R&D in foreign manufacturing sector (shown in column [iv]) are significant and commensurate to the returns in their domestic manufacturing sector, whereas the returns in foreign nonmanufacturing sector (shown in column [v]) are very small. However, these return estimates of international spillovers are downward biased because they are relative to the number of countries included in the data. To summarize, the social returns to manufacturing R&D are two to six times greater than the private returns.

Average rates of return to individual foreign manufacturing R&D in country j's domestic manufacturing sector, (1/[N.sub.j]) * [[SIGMA].sub.s[not equal to]j] [[delta][Y.sub.mjt]/[[delta][R.sup.d.sub.mst] ( = [1/[N.sub.j] * [[SIGMA].sub.s[not equal to]j] [[delta][Y.sub.ojt]/ [[delta][R.sup.d.sub.mst] where [N.sub.j].= 13 for OECD economies and 14 for the NIEs), and in country j's domestic nonmanufacturing sector, (1/[N.sub.j]) * [[SIGMA].sub.s[not equal to]j] [[delta] [Y.sub.ojt]/[[delta][R.sup.d.sub.mst] ( = [1/[N.sub.j] * [[SIGMA].sub.s[not equal to]j] [e.sub.mj,st] * [[Y.sub.ojt]/[R.sup.d.sub.mst]]), are presented in columns (vi) and (vii). The calculated returns are very small and insignificant, especially for the NIEs. (18) These findings are due to the fact that constant elasticity assumption is taken and that the sizes of the influenced economies are small relative to the existing foreign R&D stock.

The returns to domestic nonmanufacturing R&D stock in the domestic nonmanufacturing sector, [delta][Y.sub.ojt]/[delta][R.sup.d.sub.ojt] ( = [[beta].sub.01]/[Y.sub.ojt]) in column (viii) show that they are significantly large (68.0% to 209.8%).

VI. CONCLUDING REMARKS

This article examines the significance of International and intersectoral R&D spillovers on the TFP of the manufacturing and nonmanufacturing sectors in the 14 OECD economies and 3 NIEs based on two-sector pooled time-series regressions allowing for differential R&D impacts in each sectoral TFP equation.

First, the study shows an evidence of significant intersectoral R&D spillover effect from manufacturing to nonmanufacturing sectors. In all specifications of TFP equations, the manufacturing R&D had strong impact on the nonmanufacturing TFP. Asymmetry in the intersectoral R&D effects was also identified, as the nonmanufacturing R&D investments do not show any significant influence on the manufacturing TFP. These findings have an implication that the productivity advances in manufacturing sector due to R&D will bring about productivity increases in all sectors. The explanation may lie in the strong industrial link in which the manufacturing R&D enhances the technological contents of both consumer and intermediate goods that are used in both sectors. Different innovative intensities in the two sectors would give rise to a gap in sectoral productivity growths, so presence of intersectoral R&D spillover will help achieve balanced productivity growths in both sectors of an economy.

Second, the article provides strong evidence that foreign manufacturing R&D efforts have contributed to both domestic manufacturing and nonmanufacturing TFPs when international trade is considered as the channel of technology diffusion.

Third, empirical results indicate that the manufacturing R&D capital stocks of the three East Asian NIEs are not effective in raising the productivity of other economies, whereas the OECD R&D stocks are. This empirical result provides an indirect evidence that the NIEs are far behind in the technological frontier and the nature of their domestic R&D may be effective only in increasing NIEs' own productivity or efficiency to approach the frontier, and those R&D specific to NIEs' needs may not be useful in nature for the advanced economies because the latter countries are closer to the technological frontier.

Finally, given these positive spillover effects, the social returns to manufacturing R&D are estimated to be two to six times greater than the private returns in the manufacturing sector alone. The rates of return are significantly higher for the three NIEs than for the G-5 economies.

Complementarity among R&D and other intangibles may exist, so it would be interesting to investigate the role of sector-specific human capital in enhancing the R&D spillovers. It would provide better insight into how countries facilitate the technological diffusion. I leave the study of these issues to future research.

APPENDIX: DERIVATION OF TFP FROM ESTIMATION OF SECTORAL PRODUCTION FUNCTIONS

In calculating TFP, many of the previous studies have taken a competitive labor market assumption where the output elasticity with respect to labor in equation (3), 1 - [[beta].sub.i], is set equal to the average of the actual labor shares. This study relaxes the competitive labor market assumption and estimates the output elasticities with respect to labor in each sectoral production function based on a pooled time-series data set of the OECD economies (1970 to 1995) and the three NIEs (1980 to 1995). Regressions are in log difference of the variables to avoid nonstationarity problem in the level series.

d log [Y.sub.ijt] = [c.sub.ij] + [[beta].sub.i]d log [K.sub.ijt] + (1 - [[beta].sub.i]) x d log [L.sub.ijt] + [[epsilon].sup.p.sub.ijt], i = m or o,

where [c.sub.ij] denotes the con slant growth rate of technical progress of sector i in country j, and [[epsilon].sup.p.sub.ijt] is the error term with i.i.d. (0, [[sigma].sup.p.sub.i]). The variables included are real GDP ([Y.sub.ijt]), utilized physical capital stock ([K.sub.ijt]), and total labor hours worked ([L.sub.ijt]) of sector i in country j. Because [Y.sub.ijt], [K.sub.ijt], and [L.sub.ijt] are all endogenously determined in a system of structural equations, a simple OLS regression will be subject to an endogeneity problem. Thus, IV methods were used in the estimation of the two structural equations with countrywise heteroscedasticity adjustment. The IV estimates of [[beta].sub.i] are 0.454 and 0.346 for the manufacturing and nonmanufacturing sectors, respectively. The results show evidence of technical progress in both sectors for the OECD economies and the three East Asian NIEs. Using these estimates of [[??].sub.i], the TFP series for each sector were estimated using equation (4).

The data sources used in these regressions are described. The manufacturing data for the real GDP, gross fixed capital stock, and the number of persons employed for the OECD countries are obtained from OECD STAN Database for Industrial Analysis and OECD ISDB database. The respective aggregate economy data are from OECD Outlook Database. Because the physical capital stocks were not available for the aggregate economies of the OECD countries, the log-linear extrapolations involving the earliest 10 years of gross fixed capital formation data available were used to estimate the earlier investments and the initial capital stocks. They were then accumulated to form the stock series based on the perpetual inventory method with the retirement rates, which were assumed to be 3%. As for the manufacturing sector of the three NIEs, the real GDP, the gross fixed capital formation, and the numbers of persons employed are obtained from UN National Accounts and UN Industrial Statistics. The manufacturing physical capital stocks of the NIEs were estimated in the analogous method as described. The data for the aggregate economies of the three NIEs are from Lau and Park (2002). The data sources are discussed in their paper. The manufacturing capacity utilization rates for the G-7 are from OECD National Accounts. The rates for the three NIEs and the non-G7 OECD economies were estimated using the peak-to-peak method scaled with the average of the G-7 rates. Utilized capital stocks were derived from the physical capital stock series multiplied by the manufacturing capacity utilization rates. The labor hours were derived from weekly hours worked per person multiplied by the number of persons employed. The weekly labor hours were taken from ILO Yearbook of Labor Statistics. All nonmanufacturing variables are derived from the difference between the aggregate and manufacturing series.

The instrumental variables used in the regressions include country dummies, male and female population, male and female life expectancies, arable land, and contemporaneous and lagged world prices of cotton, oil and iron ore relative to the world price of wheat. World population and male and female population are from UN Statistical Yearbook. World prices of cotton, oil, iron, and wheat are from International Financial Statistics, IMF.

ABBREVIATIONS

GDP: Gross Domestic Product

IV: Instrumental Variables

NIE: Newly Industrialized Economies

OECD: Organisation for Economic Co-operation and Development

OLS: Ordinary Least Squares

R&D: Research and Development

SUR: Seemingly Unrelated Regression

TFP: Total Factor Productivity

TABLE 1
Average Annual Rates of Growth: TFP and R&D stocks (%)

 R&D Capital Stocks

 Manufacturing
 TFP
 Non Foreign Foreign
 Manufac- manufac- Domestic OECD NIE
 turing turing R&D R&D (a) R&D (a)

Australia 0.5 0.6 4.7 6.4 19.6
Belgium 1.7 0.6 4.3 5.2 20.4
Canada 0.3 -0.2 5.6 5.9 22.9
Denmark 0.5 1.1 5.9 5.5 19.7
Finland 3.3 1.3 8.4 5.7 27.5
France 1.0 1.0 4.1 5.5 21.8
Germany 0.8 1.2 4.5 6.2 21.9
Italy 2.0 0.4 6.5 5.3 19.5
Japan 1.1 1.2 8.4 9.1 23.9
Netherlands 1.8 0.4 2.3 5.4 20.2
Norway 0.2 1.8 3.3 5.0 26.8
Sweden 2.4 0.9 6.0 4.6 16.7
U.K. 2.1 0.8 2.0 6.7 22.2
U.S. 2.0 0.0 2.9 9.4 22.1
Korea 2.6 2.2 24.9 13.5 25.7
Singapore 2.5 2.6 17.2 11.5 30.3
Taiwan 2.3 3.2 17.5 11.8 34.6

 R&D Capital Stocks

 Nonmanufacturing

 Foreign Foreign
 Domestic OECD NIE
 R&D R&D (a) R&D (a)

Australia 4.6 -2.7 29.2
Belgium 3.8 -0.6 8.6
Canada 4.8 0.6 10.5
Denmark 5.0 -0.3 13.7
Finland 7.8 5.8 4.5
France 4.5 -0.1 2.7
Germany 3.6 1.3 3.9
Italy 4.6 -0.4 0.5
Japan 5.3 1.2 7.2
Netherlands 3.2 -0.7 6.9
Norway 6.5 -5.2 17.8
Sweden 5.8 0.9 4.3
U.K. 2.3 -1.9 -3.4
U.S. 4.5 0.2 -9.5
Korea 11.6 6.1 1.4
Singapore 19.6 5.5 6.3
Taiwan 13.1 4.4 8.5

Notes: Data sources are described in section III and appendix. Sample
periods are 1980-95 for all economies.

(a) Foreign manufacturing R&D stock ([R.sup.f.sub.mjt]) is defined as
an import-weighted sum of real R&D intensity of other countries.
[R.sup.f.sub.mjt] = [[SIGMA].sub.s[not equal to]j] ([v.sub.mj,st] *
[R.sup.d.sub.mst]) / [y.sub.mst], where [v.sub.mj,st] is the
manufacturing imports of country i from country s, [y.sup.d.sub.mst] is
the total gross output of manufacturing sector in country s, and
[R.sup.d.sub.mst] is the domestic manufacturing R&D of country s at
time t. Foreign nonmanufacturing R&D stock ([R.sup.f.sub.ojt]) is
defined analogously. Foreign OECD R&D and foreign NIE R&D are weighted
sums of R&D stocks solely based on other OECD R&D stocks and solely
based on other NIE R&D stocks, respectively.

TABLE 2
Group Mean Panel Unit Root Tests: Variables in TFP Regressions (a)

 Mean
Variable [bar.t] (b) [bar.p] (c) Adjustment (d)

Level series

log(TF[P.sub.m]) -1.2077 1.2353 -1.4757
log(TF[P.sub.o]) -0.7470 1.1765 -1.4881
log([R.sup.d.sub.m]) (e) -0.8847 1.8824 -1.4414
log([R.sup.f.sub.m]) -0.9455 1.2941 -1.4689
log([R.sup.fp.sub.m]) -0.9616 1.4118 -1.4608
log([R.sup.fg.sub.m]) -0.9686 1.4118 -1.4608
log([R.sup.fh.sub.m]) -0.9416 1.4118 -1.4670
log([R.sup.d.sub.m]) (e) -1.1657 1.6471 -1.4644
log([R.sup.f.sub.o]) -1.2398 1.3529 -1.4682
log([R.sup.fp.sub.o]) -1.2487 1.3529 -1.4682
log([R.sup.fg.sub.o]) -1.2463 1.3529 -1.4682
log([R.sup.fh.sub.o]) -1.2249 1.2353 -1.4813

Difference series

dlog(TF[P.sub.m]) -2.7096 1.1177 -1.4965
dlog(TF[P.sub.o]) -2.7535 1.2941 -1.4850
dlog([R.sup.d.sub.m]) (e) -2.1278 1.2941 -1.4822
dlog([R.sup.f.sub.m]) -2.6405 1.1177 -1.4992
dlog([R.sup.fp.sub.m]) -2.6385 1.1177 -1.4992
dlog([R.sup.fg.sub.m]) -2.6390 1.1177 -1.4992
dlog([R.sup.fh.sub.m]) -2.0578 1.6471 -1.4614
dlog([R.sup.d.sub.m]) (e) -2.3986 1.7059 -1.4498
dlog([R.sup.f.sub.o]) -2.6551 1.1765 -1.4932
dlog([R.sup.fp.sub.o]) -2.6579 1.1765 -1.4932
dlog([R.sup.fg.sub.o]) -2.6559 1.1765 -1.4932
dlog([R.sup.fh.sub.o]) -3.4498 1.2941 -1.4822

 Group
 Mean
 Variance Statistic
Variable Adjustment (d) ([[bar.t].sub.nt])

Level series

log(TF[P.sub.m]) 1.0268 1.0907
log(TF[P.sub.o]) 1.0249 3.0182
log([R.sup.d.sub.m]) (e) 1.0068 2.2877
log([R.sup.f.sub.m]) 1.0307 2.1257
log([R.sup.fp.sub.m]) 1.0407 2.0177
log([R.sup.fg.sub.m]) 1.0407 1.9893
log([R.sup.fh.sub.m]) 1.0447 2.1195
log([R.sup.d.sub.m]) (e) 0.9879 1.2393
log([R.sup.f.sub.o]) 1.0386 0.9241
log([R.sup.fp.sub.o]) 1.0386 0.8882
log([R.sup.fg.sub.o]) 1.0386 0.8979
log([R.sup.fh.sub.o]) 1.0289 1.0421

Difference series

dlog(TF[P.sub.m]) 0.9558 -5.1162 ***
dlog(TF[P.sub.o]) 0.9656 -5.3223 ***
dlog([R.sup.d.sub.m]) (e) 0.9715 -2.7003 ***
dlog([R.sup.f.sub.m]) 0.9544 -4.8171 ***
dlog([R.sup.fp.sub.m]) 0.9544 -4.8084 ***
dlog([R.sup.fg.sub.m]) 0.9544 -4.8105 ***
dlog([R.sup.fh.sub.m]) 0.9969 -2.4631 ***
dlog([R.sup.d.sub.m]) (e) 1.0105 -3.8919 ***
dlog([R.sup.f.sub.o]) 0.9650 -4.8767 ***
dlog([R.sup.fp.sub.o]) 0.9650 -4.8883 ***
dlog([R.sup.fg.sub.o]) 0.9650 -4.8798 ***
dlog([R.sup.fh.sub.o]) 0.9715 -8.2306 ***

Notes: *** denotes that the null hypothesis of nonstationarity is
rejected at the 1% significance level. The critical values for the test
are available in Table 2 of Im et al. (2003).

(a) Panel unit root test (t-bar test) suggested by Im et al. (2003)
provides a test of the null hypothesis of common unit autoregressive
root. log(X) and dlog(X) are log and log difference of variable X.
TF[P.sub.m], [R.sup.d.sub.m], [R.sup.f.sub.m], [R.sup.fp.sub.m],
[R.sup.fg.sub.m], and [R.sup.fh.sub.m] are, respectively, TFP, domestic
R&D stock, foreign R&D stock based on all R&D, foreign R&D stock based
on all R&D in an alternative form, foreign R&D stock based solely on
OECD R&D, and foreign R&D stock based solely on three NIE R&D for
manufacturing sector. The variables for nonmanufacturing sector,
[TF[P.sub.o], [R.sup.d.sub.o], [R.sup.f.sub.o], [R.sup.fp.sub.o],
[R.sup.fg.sub.o], and [R.sup.fh.sub.o], are defined analogously. Sample
periods are 1980-95.

(b) Cross-section average of individual Dickey-Fuller t-statistics.

(c) Cross-section average of individual number of lagged differenced
terms in ADF([p.sub.i]) regressions.

(d) Cross-section averages of mean adjustment and variance adjustments
based on Table 3 of Im et al. (2003).

(e) Test results for log levels of [R.sup.d.sub.m] and [R.sup.d.sub.o]
suggest that they are not first-order integrated for the given sample.
It may be due to short-run dynamics in the data because the
longitudinal dimension is small. Extending the sample period to include
1970-95 for the G-7, the test results show first-order integration for
the two variables.

TABLE 3
Hypothesis Tests of Common Parameters in TFP Equations for OECD
and East Asian Economies

 Manufacturing TFP equation

Null Hypothesis F-Statistic p-Value

[[beta].sup.h.sub.m1] = 0 3.832 0.052
[[beta].sup.h.sub.m2] = 0 5.778 0.017
[[beta].sup.h.sub.m3] = 0 2.550 0.112
[[beta].sup.h.sub.m4] = 0 1.164 0.282
[[beta].sup.h.sub.m5] = 0 0.146 0.703
[[beta].sup.h.sub.m6] = 0 0.067 0.796
[[beta].sup.h.sub.mk] = 0 (for all k) 1.559 0.161

 Nonmanufacturing TFP equation

Null Hypothesis F-Statistic p-Value

[[beta].sup.h.sub.o1] = 0 2.147 0.144
[[beta].sup.h.sub.o2] = 0 5.292 0.022
[[beta].sup.h.sub.o3] = 0 0.654 0.420
[[beta].sup.h.sub.o4] = 0 0.896 0.345
[[beta].sup.h.sub.o5] = 0 0.283 0.595
[[beta].sup.h.sub.o6] = 0 3.347 0.069
[[beta].sup.h.sub.ok] = 0 (for all k) 2.830 0.011

Notes: The hypothesis tests are based on equations (11) and (12) of
section IV. Time-specific and country-specific constants were included
in the unrestricted and restricted models.

F-test statistics are based on F(1,211) for tests of individual
parameter restriction and F(6,211) for tests of all parameter
restriction, respectively.

TABLE 4
Manufacturing TFP Regressions (Pooled Full Sample Data of 1981-95 for
14 OECD Economies and 3 NIEs: 255 Observations) (a)

 (i) (ii)

Variable Coef. SE Coef. SE

dlog([R.sup.d.sub.m]) 0.124 0.062 ** 0.124 0.062 **
dlog([R.sup.f.sub.m]) 0.066 0.021 ***
dlog([R.sup.fp.sub.m]) 0.066 0.021 ***
dlog([R.sup.fg.sub.m])
dlog([R.sup.fh.sub.m])
dlog([R.sup.d.sub.o]) -0.096 0.087 -0.096 0.087
dlog([R.sup.f.sub.o]) -0.002 0.006
dlog([R.sup.fp.sub.o]) -0.002 0.006
dlog([R.sup.fg.sub.o])
dlog([R.sup.fh.sub.o])

[R.sup.3] 0.385 0.385
Adjusted [R.sup.2] 0.287 0.287
Durbin-Watson 1.669 1.670

Test of no time-specific
constants (b)
 F-statistics 5.048 5.056
 p-Values 0.000 0.000
Test of no country-specific
constants (b)
 F-statistics 2.179 2.171
 p-Values 0.007 0.007

 (iii) (iv)

Variable Coef. SE Coef. SE

dlog([R.sup.d.sub.m]) 0.124 0.062 ** 0.115 0.063 *
dlog([R.sup.f.sub.m])
dlog([R.sup.fp.sub.m])
dlog([R.sup.fg.sub.m]) 0.065 0.021 *** 0.074 0.023 ***
dlog([R.sup.fh.sub.m]) -0.009 0.011
dlog([R.sup.d.sub.o]) -0.097 0.087 -0.090 0.087
dlog([R.sup.f.sub.o])
dlog([R.sup.fp.sub.o])
dlog([R.sup.fg.sub.o]) -0.002 0.006 -0.002 0.006
dlog([R.sup.fh.sub.o])

[R.sup.3] 0.385 0.386
Adjusted [R.sup.2] 0.286 0.285
Durbin-Watson 1.670 1.684

Test of no time-specific
constants (b)
 F-statistics 5.063 5.052
 p-Values 0.000 0.000
Test of no country-specific
constants (b)
 F-statistics 2.172 2.163
 p-Values 0.007 0.007

 (v) (vi)

Variable Coef. SE Coef. SE

dlog([R.sup.d.sub.m]) 0.124 0.062 ** 0.115 0.063 *
dlog([R.sup.f.sub.m])
dlog([R.sup.fp.sub.m])
dlog([R.sup.fg.sub.m]) 0.067 0.021 *** 0.076 0.024 ***
dlog([R.sup.fh.sub.m]) -0.009 0.011
dlog([R.sup.d.sub.o]) -0.100 0.087 -0.093 0.088
dlog([R.sup.f.sub.o])
dlog([R.sup.fp.sub.o])
dlog([R.sup.fg.sub.o]) 0.000 0.006 -0.001 0.006
dlog([R.sup.fh.sub.o]) -0.003 0.003 -0.003 0.003

[R.sup.3] 0.387 0.388
Adjusted [R.sup.2] 0.285 0.284
Durbin-Watson 1.680 1.694

Test of no time-specific
constants (b)
 F-statistics 5.020 5.021
 p-Values 0.000 0.000
Test of no country-specific
constants (b)
 F-statistics 2.086 2.079
 p-Values 0.010 0.010

Notes: *, **, and *** are significant at 10%, 5%, and 1%, respectively.

(a) Dependent Variable: dlog(TF[P.sub.m]). All equations have been
estimated by OLS. dlog(X) is log difference of variable X. TF[P.sub.m]
[R.sup.d.sub.m], [R.sup.f.sub.m], [R.sup.fp.sub.m], [R.sup.fg.sub.m],
and [R.sup.fh.sub.m] are, respectively, TFP, domestic R&D stock,
foreign R&D stock based on all R&D, foreign R&D stock based on all R&D
in an alternative form, foreign R&D stock based solely on OECD R&D, and
foreign R&D stock based solely on 3 NIE R&D for manufacturing sector.
The variables for nonmanufacturing sector, [R.sup.d.sub.o],
[R.sup.f.sub.o], [R.sup.fp.sub.o], [R.sup.fg.sub.o], and
[R.sup.fh.sub.o], are defined analogously.

(b) Tests of no time-specific constants and no country-specific
constants were performed. The p-values are based on F(14,218) and
F(16,218), respectively. The p-values listed as 0.000 have p-values
less than 0.001. In all specifications, unreported time-specific and
country-specific constants were included based on the respective test
results. All regressions include an unreported year dummy for German
reunification.

TABLE 5
Nonmanufacturirng TFP Regressions (Pooled Full Sample data of 1981 -95
for 14 OECD Economies and 3 NIEs, 255 Observations) (a)

 (i) (ii)
Variable Coef SE Coef SE

dlog ([R.sup.d.sub.m]) 0.056 0.027 ** 0.054 0.027 *
dlog ([R.sup,f.sub.m] 0.044 0.015 ***
dlog ([R.sup.fp.sub.m]) 0.044 0.015 ***
dlog (R.sup.fg.sub.m])
dlog (R.dup.fh.sub.m])
dlog (R.sup.d.sub.o]) 0.073 0.037 *** 0.073 0.037 *
dlog ([R.sup.f.sub.o]) -0.002 0.004
dlog ([R.sup.fp.sub.o]) -0.002 0.004
dlog ([R.sup.fg.sub.o])
dlog ([R.sup.fh.sub.o])

[R.sup.2] 0.203 0.204
Adjusted [R.sup.2] 0.187 0.188
Durbin-Watson 1.614 1.615

Test of no time-specific
constants (b)
 F-statistics 1.215 1.215
 p-values 0.265 0.266

Test of no country-specific
constants (b) 1.999 1.986
 F-statistics 0.014 0.015
 p-values

 (iii) (iv)
Variable Coef SE Coef SE

dlog ([R.sup.d.sub.m]) 0.054 0.027 ** 0.047 0.029
dlog ([R.sup,f.sub.m]
dlog ([R.sup.fp.sub.m])
dlog (R.sup.fg.sub.m]) 0.044 0.015 *** 0.045 0.016 ***
dlog (R.dup.fh.sub.m])
dlog (R.sup.d.sub.o]) 0.073 0.037 ** 0.087 0.043 **
dlog ([R.sup.f.sub.o])
dlog ([R.sup.fp.sub.o])
dlog ([R.sup.fg.sub.o]) -0.002 0.004 -0.004 0.004
dlog ([R.sup.fh.sub.o]) 0.003 0.003

[R.sup.2] 0.204 0.004 0.212
Adjusted [R.sup.2] 0.188 0.191
Durbin-Watson 1.614 1.615

Test of no time-specific
constants (b)
 F-statistics 1.219 1.263
 p-values 0.263 0.232

Test of no country-specific
constants (b) 1.995 1.943
 F-statistics 0.015 0.018
 p-values

 (v) (iv)
Variable Coef SE Coef SE

dlog ([R.sup.d.sub.m]) 0.054 0.027 *** 0.50 0.029 *
dlog ([R.sup,f.sub.m]
dlog ([R.sup.fp.sub.m])
dlog (R.sup.fg.sub.m]) 0.041 0.016 ** 0.042 0.017 **
dlog (R.dup.fh.sub.m]) 0.003 0.007 0.003 0.008
dlog (R.sup.d.sub.o]) 0.071 0.037 * 0.082 0.043
dlog ([R.sup.f.sub.o])
dlog ([R.sup.fp.sub.o])
dlog ([R.sup.fg.sub.o]) -0.002 0.004 -0.003 0.005
dlog ([R.sup.fh.sub.o]) 0.003 0.003

[R.sup.2] 0.205 0.212
Adjusted [R.sup.2] 0.185 0.188
Durbin-Watson 1.611 1.612

Test of no time-specific
constants (b)
 F-statistics 1.218 1.263
 p-values 0.263 0.232

Test of no country-specific
constants (b) 1.949 1.899
 F-statistics 0.018 0.022
 p-values

Note: *, ***, and *** are significant at 10%, 5%, and 1%, respectively.

(a) Dependent Variable: dlog (TFP) All equations have been estimated
by OLD, dlog (X) is log difference of variable C. TF[P.sub.o],
[R.sup.d.sub.o], pR.sup.f.sub.o], pR.sup.fg.sub.o], and
[R.sup.fh.sub.o] are, respectively TFP domestic R&D stock, foreign R&D
stock based on all R&D stock based on all R&D in an alternative form,
foreign R&D stock based solely on OECD R&D, adn
foreign R&D based sole;y pn 3 NIE & R&D for nonmanufacturing sector.
The variables for manufacturing sector, [R.sup.d.sub.m],
[R.sup.f.sub.m], [R.sup.fg.sub.m], and [R.sup.fh.sub.m] are
defined analogously.

(b) Tests of no time-specific constants and no country-specific
constants were performed. The p-values are based on F(14,218) and
F(16,218), respectively. In all specifications, both time-specific and
county-specific constants were excluded based on the respective test
results. All regressions include and unreported constant and a
year-dummy for German reunification.

TABLE 6
IV and SUR Estimates of TFP Equations (Pooled Full Sample Data of
1981-95 for 14 OECD Economies and 3 NIEs: 255 Observations)

 IV

 dlog(TFP.sub.m]) dlog(TFP.sub.o])

 (i) (ii)
Dependent
Variable: Coef. SE Coef. SE

dlog([R.sup.d.sub.m]) 0.171 0.092 * 0.050 0.029 *
dlog([R.sup.fg.sub.m]) 0.080 0.025 *** 0.042 0.017 **
dlog([R.sup.fh.sub.m]) -0.007 0.012 0.003 0.007
dlog([R.sup.d.sub.m]) -0.108 0.092 0.082 0.043 *
dlog([R.sup.fg.sub.m]) 0.001 0.006 -0.003 0.005
dlog([R.sup.fh.sub.m]) -0.004 0.004 0.003 0.003
[R.sup.2] 0.402 0.212
Durbin-Watson 1.744 1.612
Heteroscedasticity
Adjustment No No

 SUR

 dlog(TFP.sub.m]) dlog(TFP.sub.o])

 (iii) (iv)
Dependent
Variable: Coef. SE Coef. SE

dlog([R.sup.d.sub.m]) 0.117 0.058 ** 0.055 0.026 **
dlog([R.sup.fg.sub.m]) 0.075 0.022 *** 0.041 0.016 **
dlog([R.sup.fh.sub.m]) -0.009 0.010 0.002 0.007
dlog([R.sup.d.sub.m]) -0.070 0.080 0.072 0.037 *
dlog([R.sup.fg.sub.m]) -0.003 0.003 0.001 0.002
dlog([R.sup.fh.sub.m]) -0.001 0.006 -0.002 0.004
[R.sup.2] 0.387 0.205
Durbin-Watson 1.691 1.614
Heteroscedasticity
Adjustment No No

 SUR

 dlog(TFP.sub.m]) dlog(TFP.sub.o])

 (v) (vi)
Dependent
Variable: Coef. SE Coef. SE

dlog([R.sup.d.sub.m]) 0.087 0.057 0.057 0.027 **
dlog([R.sup.fg.sub.m]) 0.078 0.021 *** 0.033 0.015 **
dlog([R.sup.fh.sub.m]) -0.011 0.010 0.002 0.006
dlog([R.sup.d.sub.m]) -0.010 0.075 0.075 0.038 *
dlog([R.sup.fg.sub.m]) -0.001 0.003 0.001 0.002
dlog([R.sup.fh.sub.m]) -0.002 0.005 -0.003 0.004
[R.sup.2] 0.395 0.189
Durbin-Watson 1.703 1.551
Heteroscedasticity
Adjustment Yes Yes

Notes: *, **, and *** are significant at 10%, 5%, and 1%, respectively.

(a) Models (iii) and (iv), and (v) and (vi) were estimated pair-wise
using SUR method. The specifications (i), (iii), and (v) are based on
models (vi) of Table 4, and they include both time-specific and
country-specific constants. The specifications (ii), (iv) and (vi), are
based on models (vi) of Table 5. dlog(X) is log difference of variable
X. TF[P.sub.m], [R.sub.m.sup.d], [R.sub.m.sup.fg], and [R.sub.m.sup.fh]
are, respectively, TFP, domestic R&D stock, foreign R&D stock based
solely on OECD R&D, and foreign R&D stock based solely on 3 NIE R&D
for manufacturing sector. The variables for nonmanufacturing sector,
TF[P.sub.o], [R.sup.d.sub.o], [R.sup.fg.sub.o], and [R.sup.fh.sub.o],
are defined analogously. All regressions include an unreported constant
and a year-dummy for German reunification.

TABLE 7
Manufacturing TFP Elasticities with Respect to Foreign Manufacturing
R&D by Individual Source Country

 Australia Belgium Canada Denmark Finland France

Australia 0.000 0.001 0.001 0.000 0.001 0.003
Belgium 0.000 0.000 0.000 0.000 0.001 0.016
Canada 0.000 0.000 0.000 0.000 0.000 0.001
Denmark 0.000 0.002 0.000 0.000 0.002 0.006
Finland 0.000 0.002 0.000 0.002 0.000 0.005
France 0.000 0.007 0.000 0.001 0.001 0.000
Germany 0.000 0.006 0.001 0.002 0.001 0.017
Italy 0.000 0.004 0.000 0.001 0.001 0.018
Japan 0.001 0.001 0.002 0.001 0.000 0.004
Netherlands 0.000 0.008 0.000 0.001 0.001 0.008
Norway 0.000 0.002 0.001 0.004 0.003 0.005
Sweden 0.000 0.003 0.000 0.005 0.005 0.007
U.K. 0.000 0.003 0.001 0.001 0.001 0.011
U.S. 0.000 0.001 0.019 0.000 0.000 0.005
Korea 0.001 0.000 0.001 0.000 0.000 0.002
Singapore 0.001 0.001 0.000 0.000 0.000 0.004
Taiwan 0.001 0.001 0.001 0.000 0.000 0.002

 Germany Italy Japan Netherlands Norway Sweden

Australia 0.007 0.002 0.014 0.001 0.000 0.004
Belgium 0.021 0.002 0.003 0.010 0.000 0.003
Canada 0.002 0.001 0.003 0.000 0.000 0.001
Denmark 0.022 0.002 0.002 0.005 0.002 0.019
Finland 0.017 0.002 0.006 0.003 0.001 0.019
France 0.024 0.007 0.004 0.005 0.000 0.003
Germany 0.000 0.006 0.007 0.008 0.001 0.005
Italy 0.025 0.000 0.002 0.005 0.000 0.003
Japan 0.007 0.002 0.000 0.001 0.000 0.002
Netherlands 0.024 0.002 0.003 0.000 0.000 0.007
Norway 0.013 0.002 0.003 0.003 0.000 0.022
Sweden 0.023 0.002 0.003 0.006 0.003 0.000
U.K. 0.018 0.003 0.006 0.006 0.001 0.005
U.S. 0.010 0.002 0.027 0.001 0.000 0.003
Korea 0.006 0.001 0.024 0.001 0.000 0.001
Singapore 0.005 0.001 0.026 0.001 0.000 0.001
Taiwan 0.007 0.001 0.029 0.001 0.000 0.001

 U.K. U.S.

Australia 0.007 0.034
Belgium 0.008 0.011
Canada 0.001 0.066
Denmark 0.007 0.006
Finland 0.008 0.010
France 0.010 0.014
Germany 0.009 0.014
Italy 0.008 0.008
Japan 0.004 0.050
Netherlands 0.011 0.012
Norway 0.009 0.009
Sweden 0.011 0.009
U.K. 0.000 0.019
U.S. 0.007 0.000
Korea 0.002 0.036
Singapore 0.004 0.031
Taiwan 0.002 0.029

Notes: The calculations are based on the SUR estimates of model (iii)
in Table 6.

TABLE 8
Nonmanufacturing TFP Elasticities with Respect to Foreign Manufacturing
R&D by Individual Source Country

 Australia Belgium Canada Denmark Finland France

Australia 0.000 0.000 0.000 0.000 0.000 0.001
Belgium 0.000 0.000 0.000 0.000 0.001 0.015
Canada 0.000 0.000 0.000 0.000 0.000 0.002
Denmark 0.000 0.001 0.000 0.000 0.001 0.002
Finland 0.000 0.000 0.000 0.000 0.000 0.001
France 0.000 0.009 0.001 0.001 0.001 0.000
Germany 0.000 0.011 0.001 0.003 0.002 0.032
Italy 0.000 0.004 0.000 0.001 0.001 0.018
Japan 0.001 0.001 0.003 0.001 0.000 0.004
Netherlands 0.000 0.007 0.000 0.001 0.001 0.007
Norway 0.000 0.000 0.000 0.001 0.001 0.001
Sweden 0.000 0.001 0.000 0.002 0.002 0.002
U.K. 0.000 0.005 0.001 0.001 0.002 0.016
U.S. 0.001 0.003 0.048 0.001 0.001 0.012
Korea 0.000 0.000 0.001 0.000 0.000 0.001
Singapore 0.000 0.000 0.000 0.000 0.000 0.002
Taiwan 0.000 0.000 0.000 0.000 0.000 0.001

 Germany Italy Japan Netherlands Norway Sweden

Australia 0.002 0.001 0.005 0.000 0.000 0.001
Belgium 0.021 0.002 0.002 0.009 0.000 0.003
Canada 0.002 0.001 0.005 0.000 0.000 0.001
Denmark 0.006 0.001 0.001 0.001 0.000 0.005
Finland 0.003 0.000 0.001 0.000 0.000 0.003
France 0.032 0.009 0.005 0.006 0.000 0.004
Germany 0.000 0.012 0.013 0.015 0.001 0.009
Italy 0.025 0.000 0.002 0.005 0.000 0.003
Japan 0.009 0.002 0.000 0.001 0.000 0.003
Netherlands 0.021 0.002 0.003 0.000 0.000 0.004
Norway 0.003 0.000 0.001 0.001 0.000 0.005
Sweden 0.008 0.001 0.001 0.002 0.001 0.000
U.K. 0.026 0.004 0.008 0.008 0.001 0.007
U.S. 0.024 0.006 0.067 0.003 0.001 0.006
Korea 0.004 0.001 0.017 0.001 0.000 0.001
Singapore 0.003 0.001 0.014 0.001 0.000 0.001
Taiwan 0.004 0.001 0.016 0.001 0.000 0.001

 U.K. U.S.

Australia 0.002 0.011
Belgium 0.008 0.011
Canada 0.002 0.094
Denmark 0.002 0.002
Finland 0.001 0.002
France 0.013 0.018
Germany 0.017 0.027
Italy 0.008 0.008
Japan 0.004 0.058
Netherlands 0.009 0.011
Norway 0.002 0.002
Sweden 0.004 0.003
U.K. 0.000 0.027
U.S. 0.017 0.000
Korea 0.001 0.025
Singapore 0.002 0.016
Taiwan 0.001 0.016

Notes: The calculations are based on the SUR estimates of model (iv) in
Table 6.

TABLE 9
Rates of Return to Manufacturing and Nonmanufacturing R&D by Sector of
Influence (%)

 Domestic Manufacturing R&D

 Intersectoral
 R&D spillovers

 Social Rate Rate of Return Rate of Return in
 Return to of in Domestic Domestic Non-
 Manufacturing Manufacturing Manufacturing
 R&D Sector Sector
 (i) (ii) (iii)

Australia 282.6 58.6 201.7
Belgium 136.0 28.6 54.5
Canada 263.3 44.1 116.2
Denmark 173.2 38.1 97.6
Finland 124.0 39.0 59.1
France 86.9 23.5 42.8
Germany 71.1 22.4 30.0
Italy 168.1 56.9 91.9
Japan 68.2 24.5 31.2
Netherlands 128.5 27.0 59.2
Norway 213.8 36.4 148.8
Sweden 82.8 17.4 31.3
U.K. 95.4 23.9 47.9
U.S. 68.0 18.6 38.2
Korea 119.2 57.2 62.0
Singapore 316.7 143.7 172.9
Taiwan 178.2 82.1 96.2

 International R&D Spillovers

 Rate of Return Rate of Return
 in Foreign in Foreign Non-
 Manufacturing Manufacturing
 Sector Sector
 (iv) (v)

Australia 19.9 2.3
Belgium 52.1 0.9
Canada 102.1 0.9
Denmark 36.8 0.7
Finland 25.1 0.8
France 20.0 0.6
Germany 17.9 0.8
Italy 18.8 0.6
Japan 10.6 1.8
Netherlands 41.2 1.1
Norway 27.8 0.7
Sweden 33.1 1.0
U.K. 22.8 0.7
U.S. 9.9 1.2
Korea 0.0 0.0
Singapore 0.0 0.0
Taiwan 0.0 0.0

 Foreign Manufacturing R&D Domestic Non-
 Manufacturing R&D

 Rate of Return Rate of Return Rate of Return
 in Domestic in Domestic Non- in Domestic Non-
 Manufacturing Manufacturing Manufacturing
 Sector Sector Sector
 (vi) (vii) (viii)

Australia 12.6 4.3 81.4
Belgium 5.1 1.1 119.5
Canada 5.5 3.9 85.5
Denmark 2.0 2.3 86.3
Finland 2.6 2.2 72.1
France 0.2 0.4 69.6
Germany 2.2 1.3 75.9
Italy 0.3 1.2 144.5
Japan 0.3 0.6 68.0
Netherlands 0.4 0.3 72.9
Norway 0.5 1.3 77.0
Sweden 0.2 0.5 68.9
U.K. 1.0 1.1 72.7
U.S. 0.2 0.2 72.1
Korea 0.4 0.4 122.7
Singapore 0.1 0.3 209.8
Taiwan 0.2 0.3 134.1

Notes. The calculations are based on the SUR estimates of models (iii)
and (iv) in Table 6.

(1.) In 1995, the average share of manufacturing R&D in total R&D expenditure was 48%, and the average share of manufacturing GDP in the aggregate GDP was 19% for the 14 OECD economies in this study. The two respective average shares for the three East Asian economies included in this study were 90% and 29%.

(2.) The 14 OECD economies included in this study are Australia, Belgium, Canada, Denmark, Finland, France, Germany, Italy, Japan, Netherlands, Norway, Sweden, United Kingdom, and the United States. The three East Asian economies are Korea, Singapore, and Taiwan.

(3.) Given a panel data set of 22 OECD countries, Coe and Helpman (1995) conclude that foreign R&D has beneficial effects on domestic productivity that are stronger the more open an economy is to foreign trade. Lichtenberg and de la Ponerie (1998) confirm the findings of Coe and Helpman (1995) using an alternative index of trade-intensity-weighted foreign R&D stock, which is invariant to the level of data aggregation. As R&D variables in levels exhibit non-stationarity, Edmond (2001) revisits the issue based on panel cointegration method and finds rather mixed results. Engelbrecht (1997) shows that R&D spillover effects are robust even when human capital is included as an additional variable. Frantzen (2000) also finds net impact of human capital, domestic and foreign R&D based on cross-country regressions using OECD country data.

(4.) Bernstein and Yan (1997) find that domestic spillovers generate greater effects compared with international spillovers between Canadian and Japanese manufacturing industries.

(5.) I thank an anonymous referee who suggested expanding the set of OECD countries in the study. The inclusion of more countries significantly improved the results.

(6.) The additional sources are Major Indicators of Industrial Technology (Korea Industrial Technology Association, Korea), National Survey of R&D in Singapore (National Science and Technology Board, Singapore), Taiwan Statistical Data Book (Council for Economic Planning and Development, Taiwan) and Indicators of Science and Technology (National Science Council, Taiwan).

(7.) Aggregate GDP deflators were used in deflating all series, as I assume that the prices of R&D investments in each sector are not correlated with the prices of each sector's representative goods.

(8.) Refer to Griliches and Lichtenberg (1984) for the related discussion.

(9.) Refer to Coe and Helpman (1995), Coe et al. (1997), Engelbrecht (1997), Frantzen (2000), and Keller (2002). Keller (1998) questions whether the pattern of international trade drives R&D spillovers and finds that the elasticity based on randomly created trade patterns can explain more of variation in productivity across countries. Exports rather than imports are shown to be important in receiving substantial R&D spillovers among OECD countries in Funk (2001). Other variables have been explored to proxy the R&D spillover intensity, such as technological proximity of functional composition of R&D in W. Park (1995), transaction intensity and proximity in the patent space in Frantzen (2002), foreign direct investments in Lichtenberg and de la Potterie (1998, 2001), and international student flows in J. Park (2004).

(10.) Regressions using cointegration method were considered in Coe and Helpman (1995), Edmond (2001), and Frantzen (2002). Autoregressive estimations were used in Madden et al. (2001).

(11.) The intensity of intersectoral link was incorporated by multiplying the coefficients from the input-output table to the intersectoral R&D variables. However, input-output tables were only available for a single year for the given sample period for each country. Since our regressions only consider log difference series in our regressions, multiplication of these country- and sector-specific constants to the relevant R&D stocks did not alter the original values of the log difference series without the intensity consideration.

(12.) I thank an anonymous referee for pointing out this issue.

(13.) The lagged value of R&D stock is considered an instrument under the assumption that economic agents are unable to anticipate future economic shocks to the environment. The lack of evidence of serial correlation in the error term in the regression results of section V suggests that the implied weak exogeneity assumption is indeed justified. World price of oil relative to wheat may be correlated with the decision on R&D investments because it will influence the cost of production. The male and female life expectancies may reflect the human capital environment which may also influence the decision on R&D investment because it can be considered complementary. The exogeneity tests based on Hausman specification tests were performed on the latter three variables, where the null hypothesis of exogeneity could not be rejected for all three.

(14.) F-tests with three groups of economies--G-7, non-G-70ECD, and NIEs--provided qualitatively the same results.

(15.) [e.sub.ij,st] is ([delta][TFP.sub.ijt]/[delta][R.sup.d.sub.mst]) * [R.sup.d.sub.mst]/[TFP.sub.ijt] = [[beta].sub.ik] * ([V.sub.mj,st] * [R.sup.d.sub.mst])/([Y.sub.mst] * [R.sup.fg.sub.mjt]), where k = 2 for i = m and k = 5 for i = o).

(16.) The calculations are based on the SUR estimates of (iii) and (iv) in Table 6 for the 1995 values.

(17.) Bernstein and Mohnen (1998) identified substantial international spillovers from the United States to Japan in the R&D-intensive sectors where the estimated social rate of return to U.S. R&D was twice its private return. The estimates in this study may be significantly higher due to the consideration of intersectoral R&D spillovers.

(18.) The rates of return to foreign R&D for NIEs are comparable to the previous estimates of Coe et al. (1997).

REFERENCES

Bernstein, J. I., and P. Mohnen. "International R&D Spillovers between U.S. and Japanese R&D Intensive Sectors." Journal of International Economics, 44(2), 1998, 315-38.

Bernstein, J. I., and X. Yam "International R&D Spillovers between Canadian and Japanese Industries." Canadian Journal of Economics, 30(2), 1997, 276-94.

Coe, D. T., and E. Helpman. "International R&D Spillovers." National Bureau of Economic Research Working Paper no. 4444, August 1993.

--, "International R&D Spillovers." European Economic Review, 39(5), 1995, 859-87.

Coe, D. T., E. Helpman, and A. Hoffmaister. "North-South R&D Spillovers." Economic Journal 107, 1997, 134-49.

Edmond, E. "Some Panel Cointegration Models of International R&D Spillovers." Journal of Macroeconomics, 23(1), 2001, 241-60.

Engelbrecht, H.-J. "International R&D Spillovers, Human Capital and Productivity in OECD Economies: An Empirical Investigation." European Economic Review, 41(8), 1997, 1479-88

Frantzen, D. "R&D, Human Capital and International Technology Spillovers: A Cross-Country Analysis." Scandinavian Journal of Economics, 102(1), 2000, 57-75.

--, "Intersectoral and International R&D Knowledge Spillovers and Total Factor Productivity." Scottish Journal of Political Economy, 49(3), 2002, 280-303.

Funk, M. "Trade and International R&D Spillovers among OECD Countries." Southern Economic Journal, 67(3), 2001, 725-36.

Griliches, Z., and F. Lichtenberg. "R&D and Productivity Growth at the Industry Level: Is There Still a Relationship?," in R&D, Patents, and Productivity, edited by Z. Griliches. Chicago: University of Chicago Press, 1984, 465-96.

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

Im, K. H., M. H. Pesaran, and Y. Shin. "Testing for Unit Roots in Heterogeneous Panel." Journal of Econometrics, 115, 2003, 53-74.

Keller, W. "Are International R&D Spillovers Trade-Related? Analyzing Spillovers among Randomly Matched Trade Partners." European Economic Review, 42(8), 1998, 1469-81.

--, "Trade and the Transmission of Technology." Journal of Economic Growth, 7, 2002, 5-25.

Lau, L. J., and J. Park. "The Sources of East Asian Economic Growth: Revisited." Working Paper, Stanford University, 2002.

Levin, A., and C.-F. Lin. "Unit Root Tests in Panel Data: Asymptotic and Finite-Sample Properties." UCSD Discussion Paper 92-93, University of California, San Diego, 1992.

--, "Unit Root Tests in Panel Data: New Results." UCSD Discussion Paper 93-56, University of California, San Diego, 1993.

Lichtenberg, F. R., and B. P. de la Potterie. "International R&D Spillovers: A Comment." European Economic Review, 42(8), 1998, 148341.

--, "Does Foreign Direct Investment Transfer Technology across Borders?" Review of Economics and Statistics, 83(3), 2001, 490-97.

Madden, G., S. J. Savage, and P. Bloxham. "Asian and OECD International R&D Spillovers." Applied Economics Letters, 8, 2001, 431-35.

National Science Foundation. "Human Resources for Science and Technology: The Asian Region." Surveys of Science Resources Series Special Report, NSF 93-303, 1993.

Park, J. "International Student Flows and R&D Spillovers." Economics Letters, 82, March 2004, 315-20.

Park, W. G. "International R&D Spillovers and OECD Economic Growth." Economic Inquiry, 33(4), 1995, 571-90.

JUNGSOO PARK, I thank Lawrence J. Lau and Michael J. Boskin for helpful discussions and Wolfgang Keller for providing maufacturing trade data. Two anonymous referees made useful suggestions.

Park: Assistant Professor, Department of Economics, State University of New York at Buffalo, Buffalo, NY 14260-1520. Phone 1-716-645-2121 x440, Fax 1-716-645-2127, E-mail: jungsoo@buffalo.edu