International outsourcing and the productivity of low-skilled labor in the EU.

Egger, Hartmut ; Egger, Peter

I. INTRODUCTION

In recent years, international outsourcing (international fragmentation of the value-added chain) has become one of the core interests in international economics, and it is now seen as an important source of the observed change in factor productivity and factor rewards in the recent past (see Feenstra and Hanson 1999). However, from a theoretical point of view, the impact on factor productivity and factor rewards is not clear-cut but critically depends on which factors are substituted by the international fragmentation of production, which sectors are engaged in international outsourcing, and the intersectoral and international mobility of factors. (1) In addition, factor and product market imperfections affect the outcome (at least in the short run).

Empirical research has predominantly been concerned with the effects of outsourcing on the U.S. labor market (Feenstra and Hanson 1999; Siegel and Griliches 1991; Slaughter 2000; etc.). Research on the importance for European economies has predominantly concentrated on the effects of trade with less developed countries on labor markets rather than focusing on direct measures of outsourcing (Anderton and Brenton 1999; Greenaway et al. 1999, 2000; Hine and Wright 1998; etc.). (2)

This article combines trade statistics for intermediate goods imports and information from input-output tables to construct a conceptually narrow measure of outsourcing (i.e., intermediate goods imports from the same industry (3)) to investigate its effect on the productivity of low-skilled labor. Due to the lack of skill-specific wages at the industry level of the European Union member countries, this has to be done in a primary production function approach. We choose a nested Constant Elasticity of Substitution (CES) framework, which is sufficiently flexible (see Perroni and Rutherford 1995) and less demanding than other flexible functional forms in terms of the parameters to be estimated. To investigate the robustness of our findings, we additionally estimate a translog specification. (4)

Interestingly, in the nonlinear CES framework we find a Hicks nonneutral augmenting effect of outsourcing on physical capital and high-skilled labor (both relative to low-skilled labor) of approximately the same size. In the short run, outsourcing exhibits a negative effect on real value added per low-skilled worker. This might be caused by imperfections on factor and product markets. In the long run, outsourcing increases real value added per low-skilled employee. The average annual change in the outsourcing to output ratio in EU manufacturing amounted to 3.2%. According to our simulations, the observed change in EU outsourcing since 1993 alone accounts for a long-run effect of about 6.0% of the observed change in the real value added per low-skilled worker.

II. THEORETICAL BACKGROUND

In general, the value added of industry i is given by

(1) [Y.sub.i] - [summation over j [not equal to] i] [D.sup.i.sub.i] - [n.summation over j=1] [O.sup.j.sub.i] = [Q.sub.i](V.sub.i]),

where [Y.sub.i] is the real value of commodity i produced in industry i, [D.sup.j.sub.i] is the real value of domestically sourced intermediate good j, and [O.sup.j.sub.i] describes the real value of the imported intermediate good j employed in the production of industry i. Throughout the analysis, we focus on the role of [O.sup.j.sub.i]. [Q.sub.i] is the value added produced at home with a vector of input factors [V.sub.i]. For notational simplicity, we omit time and country indices in the theoretical part of with a translog specification. the article. Using a CES specification for the production of [Q.sub.i], gives (5)

(2) [Q.sub.i]= [A.sub.i]{[delta][K.sup.*-[rho].sub.i] + (1 - [delta])[L.sup.*-[rho].sub.i]}.sup.r/[rho]],

where [A.sub.i] subsumes information about the level of technology, the degree of competition and the level of outsourcing activities. (6) [delta] and (1 - [delta]) are the weights of efficiency units of capital and labor in the production process, r indicates the degree of scale economies in the production of the value added. [K.sup.*] and [L.sup.*] denote levels of efficiency units of capital and labor. We define [K.sup.*.sub.i] [equivalent to] [a.sub.K] (O.sub.i][K.sub.i], where [a.sub.K]([O.sub.i]) is an efficiency measure and [K.sub.i] is the capital input used in industry i. (7) By allowing for two types of labor, we define [L.sup.*.sub.i] [equivalent to] [a.sub.H]([O.sub.i])[H.sub.i] + [a.sub.L]([O.sub.i])[L.sub.i], where [a.sub.H]([O.sub.i]) and [a.sub.L]([O.sub.i]) are again efficiency measures and [H.sub.i] and [L.sub.i] denote the amount of high-skilled and low-skilled labor employed in industry i, respectively. (8) [a.sub.K]([O.sub.i]), [a.sub.H]([O.sub.i]) and [a.sub.L]([O.sub.i]) depend on the intensity of outsourcing of industry i ([O.sub.i]), where [O.sub.i] [equivalent to] [[summation].sup.n.sub.j=1] [O.sup.j.sub.i]/[Y.sub.i]. This assumption can be justified in the following way. First, outsourcing decisions themselves may be driven by specialization effects, possibly increasing the factor productivity in different ways (see the discussion in Burda and Dluhosch 1998). Second, if production processes differ in terms of factor productivity, outsourcing of some production stages in general implies changes in the productivity of home-supplied production factors K, H, and L (see Arndt 1997). Finally, international outsourcing may also alter the degree of substitutability of factors in the production process, which is taken into account by allowing efficiency measures [a.sub.K], [a.sub.H, and [a.sub.L] to depend in different ways on [O.sub.i].

For our empirical analysis, we have to specify [A.sub.i], [a.sub.K]([O.sub.i]), [a.sub.H](O.sub.i], and [a.sub.L]([O.sub.i]). In the following, we assume [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.], [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.], and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]. After inserting, (2) can be rewritten as

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

This closes the theoretical discussion. In the empirical analysis, we focus on estimating the impact of international outsourcing on the average product of low-skilled labor [q.sub.i] [equivalent to] [Q.sub.i]/[L.sub.i] for different industries of the EU member countries.

III. DATA AND EMPIRICAL RESULTS

We use data from New Cronos (EUROSTAT) on education, employment, real value added, and real gross production with 1996 as the base year. Additionally, investment to value added ratios come from STAN (OECD). The construction of our narrow outsourcing measure requires data from the EU input-output tables (EUROSTAT) together with intermediate goods trade figures from United Nations. We use data for 21 NACE two-digit manufacturing industries in 12 EU members (EU-15 without the joining countries of 1995) for the period 1993-97 at constant prices and U.S. dollars. This database does not provide information on skill-specific wages nor on research and development and the use of computers at the required level of aggregation in the countries under consideration. However, any impact identical for all countries and industries can be captured by fixed time effects. Also, the panel is relatively short, and the variance of these variables over the covered six years is likely small. A time-invariant, country-specific effect of these variables can be accounted for by the inclusion of fixed country effects. Finally a time-invariant, industry-specific impact can be controlled for by the inclusion of fixed industry effects.

New Cronos provides information on the number of workers with a minimum of upper secondary education. We use this share of workers relative to the other ones as a measure of the high-skilled to low-skilled labor ratio (H/L) at the industry level. We cannot account for hours worked, because the required data are not available for all countries and industries under consideration. However, as long as the hours worked are not specific to factors and constant over time, they can be captured by the fixed country and industry effects.

Learner (1984) suggests to approximate capital stock data by the perpetual inventory method. Industry gross fixed capital formation is constructed by the use of investment to value added ratios at ISIC together with real value added from New Cronos at NACE. We follow Keller (2000) in the perpetual inventory construction of the time-(t), country(c), and industry-(/) specific capital stock:

(4) [K.sub.1986,ic] = 1/4 * ([I.sub.1986,ic] + [I.sub.1987,ic] + [I.sub.1988,ic] + [I.sub.1989,ic])/([g.sub.ic] + [delta])

[K.sub.t]>1986,ic] = (1 - [delta])[K.sub.t-1,ic] + [I.sub.tic],

with K denoting real capital stocks, I is real gross fixed capital formation (assuming that the nominal investment to value added ratio corresponds to the real one), g is the real average annual growth of industry specific investment between 1986 and 1997, and 6 is the depreciation rate. As suggested by Hofer et al. (1997), the latter is assumed at 10% for each industry, country, and year. (10)

Following Feenstra and Hanson (1999), we use a narrow measure of outsourcing ([O.sub.tic]), which is defined as

(5) [O.sub.tic] = [D.sub.ic][M.sub.tic]/[Y.sub.tic],

with [D.sub.ic] as the diagonal of the NACE two-digit input-output tables as a share of total intermediates usage by that industry for each EU economy in 1995 (assumed to be constant between 1993-97), (11) [Y.sub.tic] is real gross production, and [M.sub.tic] are NACE two-digit real intermediate goods imports. The latter are constructed from UN Broad Economic Categories at SITC five-digit (see Fontagne et al. 1996) and the available correspondence table between SITC five-digit and NACE three-digit provided from Statistics Austria. Notably, the constant coefficient [D.sub.ic] does not imply that outsourcing is constant over time. It means only that within this short time period we assume that the distribution of a typical industry i and country c manufacturing firm's varying level of outsourced inputs across industries remains constant.

The specifications include time dummies ([[lambda].sub.t]), country dummies ([[eta].sub.c]), and industry dummies ([[mu].sub.i]). These time-specific, country-specific, and industry-specific effects have also been considered in the theoretical discussion and were subsumed under parameter [J.sub.i]. Whereas [[mu].sub.i] controls for overall technological improvements equal for all industries and countries, [[eta].sub.c] and [[mu].sub.i] account (among others) for persistent differences between countries in the degree of competition induced by legislation (e.g., for mergers, market power) and persistent industrial differences within countries (e.g., chemistry and pharmaceutical industry versus food production). Due to the unbalancedness of the panel, we come up with 992 observations in the econometric analysis.

We estimate four specifications, labeled as Model 1, Model 2, Model 3, and Model 4. Model 1 is the nonlinear specification of the primary production framework, which uses contemporaneous outsourcing as an explanatory variable:

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

where k [equivalent to] K/L, h [equivalent to] H/L, [[beta].sub.k] [equivalent to] [[beta].sub.K] - [[beta].sub.L], [[beta].sub.h] [equivalent to] [[beta].sub.H] - [[beta].sub.L] and [[beta].sub.A] [equivalent to] [gamma] + r[[beta].sub.L]. From (6) it is clear that our specification has nice properties because it allows to distinguish between a simple "shift" and "biased" outsourcing effects on the average product of low-skilled labor. The distinction of these two effects may be especially interesting in view of the discussion on the cost-saving and (biased) factor substitution effects of international outsourcing in the Heckscher-Ohlin model. See Egger (2002) and Egger and Falkinger (2003) for an extensive analysis of shift and rotation of unit iso-cost curves in the context of international outsourcing. (12)

Model 2 is similar to Model 1, but it includes lagged rather than contemporaneous outsourcing to check the possible relevance of the endogeneity of this variable. This procedure is based on the insight that "variables that are predetermined in a model can be treated, at least asymptotically, as if they were exogenous in the sense that consistent estimates can be obtained when they appear as regressors" (Greene, 1997, p. 714). With panel data, this requires at least some longitudinal variation in the data. Model 2 reads:

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

Finally, Models 3 and 4 represent cross-sectional models, which are estimated on the variable means (indicated by subscript "*" instead of "t") in the time dimension:

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

Model 3 treats [O.sub.ic] as exogenous, and Model 4 takes its potential endogeneity into account and represents a two-step nonlinear least squares estimator (see Greene 1997, pp. 465 72, for a discussion; Murphy and Topel 1985, for the asymptotic properties; and later discussion for details on the instruments). Note that Models 1-4 for simplicity are labeled by identical parameter letters. There are no restrictions on parameters across equations.

Table 1 presents the regression results for the four estimated specifications. In Models 1 and 2, we use 0 as the starting value for most parameters except for r (1.2), p (0.5), and 8 (0.5), which are motivated by previous research on labor productivity. (13) First of all, there is both neutral and nonneutral productivity change due to outsourcing. The first parameter ([[beta].sub.A]) measures a composite of influences comprising a neutral technology shift effect, the impact of low-skilled labor and r. However, we also find a significant and positive relative physical capital augmenting and relative high-skilled labor augmenting effect of outsourcing ([[beta].sub.k], [[beta].sub.h] > 0). (14) We estimate an elasticity of substitution of 1/(1+[rho]) [nearly equal to] 0.54 for efficiency units of capital and labor, which is fairly low. The parameter estimate of r on the average implies decreasing returns to scale at the industry level (possibly due to free capacity in the short run). The low parameter estimate for [delta] reflects the fact that we could not use information on capital services, and that the capital stock variable represents a multiple of the required capital services. The exercise in Model 2 gives results that are very similar to those from Model 1. This provides insights that the parameter estimates in Model 1 are not severely affected by an endogeneity problem of the outsourcing variable, and the (short-term) exogeneity of outsourcing is not rejected. Therefore, we concentrate on the results from Model 1 in what follows as long as short-run relationships are considered.

With the parameter estimates at hand, we can investigate the marginal effect of outsourcing:

(9) [differential]ln[q.sub.tic]/[differential][O.sub.tic] = [[beta].sub.A] + r ([delta][[beta].sub.k][k.sup.-[rho]].sub.tic] + [1 - [delta]]

x [[beta].sub.h][h.sub.tic][[1 + [h.sub.tic]].sup.-[rho]-1])/([delta] [k.sup.-[rho]].sub.tic]

+ [1 - [delta]][[1 + [h.sub.tic]].sup.-[rho]]),

where [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] are used. Evaluated at the variable means, this effect is -0.181 and indicates that a one percentage point increase of the outsourcing intensity induces a decrease in the productivity of low-skilled labor of about 0.2%. (15)

The usual caveats apply, because capital services are accounted for by (estimated) capital stocks and labor inputs are measured by employment in heads. The former leads to a downward bias of the estimated capital coefficient ([delta]), and the latter omits the importance of the volume and the quality of hours worked (see Jorgenson et al. 1987, and Siegel and Griliches 1991, for an overview on this problem). Additionally, we cannot explicitly control for the effects of research and development. As mentioned earlier, we have to rely on the assumption that these effects are comprehensively accounted for by the fixed effects.

We follow the well-established literature on the estimation of short-run and long-run effects in static panel models (Baltagi 2001, Pirotte 1999). The fixed effects estimator is associated with short-run parameter estimates and the cross-sectional estimator approximates their long-run counterparts. (16) Model 4 instruments the potentially endogenous outsourcing intensity variable by (1) the average unit labor costs in the remaining EU members and the same industry and the log of the c.i.f./ f.o.b, ratio as a measure of transportation costs of a country's trade, (2) in final goods with the remaining EU economies, (3) in final goods with the Central and Eastern European countries, (17) (4) and in intermediate goods with the Central and Eastern European countries. Model 3 is rejected on the basis of the Hausman (1978) test. (18) Regarding the relatively small cross-sectional sample, the quality of the instruments is fairly good, which shows up in a canonical correlation coefficient of 0.2.

From a comparison of the parameter estimates of Model 1 with those of Model 4, we find the following main differences. First, the production of the value added exhibits diseconomies of scale in the short run, whereas the long-run regression comes up with constant scale economies, given by a value of r near unity in Model 4. The reason for this finding may be that firms produce under capacity in the short run, which can be adjusted in the long run, or that firms outsource the fixed cost intensive part of production (especially, the physical capital intensive one) to foreign economies. The latter may be related to relatively high foreign direct investment. Second, the elasticity of substitution of efficiency units of capital and labor is less pronounced within industries and countries (1/ [1 + [rho]] [nearly equal to] 0.54), that is, in the short run, than in the cross-section (1/[1 + [rho]] [nearly equal to] 0.62), associated with the long run.

The marginal long-run effect of narrow outsourcing from the cross-sectional regression (Model 4, using variable means over time) is positive and amounts to 1.813, which is much higher in absolute value as compared to its negative short-run counterpart.

IV. EXTENSION AND DISCUSSION OF THE RESULTS

The chosen approach is of course restrictive because of the strong assumption of a CES technology in the production of value added. Hence, it is convenient to relax the imposed restrictions by estimating a translog function (see Christensen et al. 1973). To capture the idea that outsourcing changes the efficiency of production factors both non-neutrally and neutrally, we can estimate the following three-factor translog model in primal form:

ln[q.sub.tic] = [[beta].sub.0] + [[beta].sub.1] [O.sub.tic] + [[beta].sub.2][O.sub.tic]ln[K.sub.tic] + [[beta].sub.3][O.sub.tic]ln[H.sub.tic] + [[beta].sub.4][O.sub.tic]ln[L.sub.tic] + [[beta].sub.5][([O.sub.tic]ln[K.sub.tic]).sup.2] + [[beta].sub.6][([O.sub.tic]ln[H.sub.tic]).sup.2] + [[beta].sub.7][([O.sub.tic]ln[L.sub.tic]).sup.2] + [[beta].sub.8]([O.sub.tic]ln[L.sub.tic]) x ([O.sub.tic]ln[K.sub.tic]) + [[beta].sub.9]([O.sub.tic]ln[L.sub.tic]) x ([O.sub.tic]ln[H.sub.tic]) + [[beta].sub.10]([O.sub.tic]ln[H.sub.tic]) x ([O.sub.tic]ln[K.sub.tic]) + [[lambda].sub.t] + [[mu].sub.i] + [[eta].sub.i] + [[eta].sub.c] + [[epsilon].sub.tic].

Table 2 summarizes the results from this regression (Model 5), which is closest to Model 1 in Table 1. Furthermore, we estimate the corresponding regression on time-averaged data (the cross-sectional estimator, Model 6).

Again, we focus on the marginal effect of narrow outsourcing on the productivity of low-skilled labor and compute the marginal effect on the average productivity of low-skilled labor [differential]ln[q.sub.tic]/[differential][O.sub.tic]. Similar to the CES case, the estimate is negative for the fixed effects model and amounts to -0.412 in Model 5 (evaluated at variable means). Hence the marginal effect is stronger under the translog specification than under the CES specification. The translog cross-sectional estimate of the marginal effect is positive, similar to the cross-sectional estimates in the CES case. It amounts to 1.176, which is somewhat smaller than under the CES technology assumption.

In qualitative terms, the results from the more flexible translog specification confirm the insights from the more restrictive CES specification. However, the adjusted [R.sup.2]; is remarkably higher for the CES models (compare Model 5 with Model 1 and Model 6 with Model 3). Therefore we use these estimates in the subsequent analysis. Overall, the results seem plausible from a theoretical point of view, and the difference in sign between the short-run (approximated by the fixed effects models) and the long-run effects (approximated by the cross-sectional estimates) may stem from the following sources.

(1) Outsourcing shifts part of production to foreign economies. First, for a given factor employment and without any changes in the employment of physical and human capital, this implies a decline in low-skilled labor productivity. Second, due to decreasing returns to scale (at least in the short run) at the industry level, a decline in the value added Q for a given outsourcing intensity results in an increase in the average product of low-skilled labor. In general, it seems to be plausible that the direct level effect of outsourcing is stronger than the indirect economies of scale effect, so that the overall impact of the production shift on the average product of low-skilled labor is negative. Moreover, the gain from international outsourcing has also its costs in terms of physical and human capital resources, associated with foreign direct investment and coordination activities, respectively. (19) For a given amount of low-skilled labor employment, a decline in the stock of capital used for the value-added production process (induced by foreign direct investment), has a negative impact on the low-skilled labor productivity. Using high-skilled labor for coordination activities rather than in the production of the value-added does not have a direct impact on the amount of high-skilled labor counted in the value-added process, but it reduces the value added Q for a given low-skilled labor employment. This implies a further negative impact on the low-skilled labor productivity.

(2) By maximizing their profits, firms want to adjust their factor employment. Whereas imperfections on human and physical capital markets are rather negligible, rigid markets for low-skilled labor in Europe are often mentioned as responsible for adjustment delays. In the short run, trade unions (under efficient bargaining), legal regulation (protection against dismissal), or social pressure prohibit perfect adjustment of labor employment, explaining the negative impact of international outsourcing on the low-skilled labor productivity. In the long-run, employment adjustments (due to higher fall-back profits of firms and lower fall-back income of workers in the bargain, due to the natural quit rate, etc.) can explain a positive impact of international outsourcing on the average product of low-skilled labor.

(3) The difference between the short-run and the long-run effects of international outsourcing may be magnified by product market imperfections. If international outsourcing has a cost-reducing effect, product market imperfections may first retard the output shift within one industry from firms producing in an integrated way to firms making use of outsourcing opportunities. Second, there may also be some delay in the adjustment of the output structure across sectors from those with relatively little advantage of international outsourcing to those with a high cost saving effect.

Taking into account (1)-(3), it seems plausible from a theoretical point of view that the marginal effect of the outsourcing intensity on the productivity of low-skilled labor exhibits a different sign in the short run and the long run.

In addition to the marginal effect of outsourcing on the average productivity of low-skilled labor, the effect on the marginal productivity of high-skilled relative to low-skilled labor is of interest, given the extensive discussion on the relationship between globalization and wage inequality. In our CES case, we obtain [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] so that [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]. (20) Evaluated at the mean, this marginal effect is 0.746 in the fixed effects model and 0.016 in the cross-sectional model. Hence, outsourcing of the EU economies induces a skill-biased effect similar to outsourcing of the United States (see Feenstra and Hanson, 1999). In a competitive economy, the increase in marginal productivity of high-skilled relative to low-skilled labor immediately leads to an increase in the high-skilled to low-skilled wage ratio. (21) The different sizes of the long-run and short-run effects have to be interpreted with care. Although the aforementioned results indicate that wage effects may be of lower relevance in the long run, there may be adverse employment effects in unionized European labor markets when firms respond to changes in outsourcing opportunities.

V. SIMULATING THE OUTSOURCING EFFECT ON THE AVERAGE PRODUCTIVITY OF LOW-SKILLED LABOR

Because the marginal effect varies over time, countries and industries in the fixed effects regression and over countries and industries in the cross-sectional model, it seems appropriate to undertake some simulations to quantify the importance of outsourcing for different industries at least in the period under consideration. In an experiment of thought, we take into account the parameter estimates of the CES specification (Models 1 and 4, respectively) and derive predictions from our model assuming that the outsourcing intensity ([O.sub.tic]) were constant since 1993. Over the period 1993-97, the observed real value added per low-skilled worker grew by 9.3% p.a. in the average country, manufacturing industry and year.

Table 3 provides more details on industry specific growth rates. The outsourcing intensity ([O.sub.tic]) grew by 3.2% on average. With growth rates of 7% and higher, the outsourcing intensity grew particularly fast in the manufacture of textiles, manufacture of wearing apparel, and tanning and dressing of leather industries. Also that intensity in the manufacture of medical, precision, and optical instruments grew at almost 10%. In a few industries, the outsourcing intensity declined. Specifically, the manufacture of furniture, manufacture of wood and products of wood, and manufacture of office machinery and computers industries with outsourcing intensity growth rates of in between -4% and -8.5% could be mentioned.

Assuming a constant [O.sub.tic] in the thought experiment means to focus on a situation where real intermediate imports and real production grew with the same annual rate. The last two columns of Table 3 present the difference between the model prediction for observed outsourcing intensity growth and the thought experiment with zero outsourcing intensity growth since 1993 for both the fixed effects (short-run) and the cross-sectional regression (long-run). According to our econometric results, in the short run the increase in the outsourcing intensity has lowered the average annual change in real value added per low-skilled worker by about 1.4% in the average industry, indicating that the short-run marginal effect of outsourcing is negative throughout the sample. In contrast, the long-run stimulus caused by the change in the outsourcing intensity is positive and amounts to about 6.0% measured in terms of the model prediction due to observed outsourcing. In accordance with our priors, this effect is highest in the paper and basic metals industries but less expected also in the radio, television, and communication equipment industries. This result shows that international outsourcing affects the low-skilled labor productivity in low-tech as well as in high-tech industries.

VI. CONCLUSIONS

This article presents first insights into the role of international outsourcing for the productivity of low-skilled workers in EU manufacturing. Because of its reliability, we follow Feenstra and Hanson (1999) in using a conceptually narrow measure of the cross-border fragmentation phenomenon. According to the unavailability of data on skill-specific factor rewards, we estimate a nested CES primary production function approach for 12 EU countries and 21 NACE two-digit industries over the period 1992-97. Our short-run evidence is inconsistent with the long-run findings by Feenstra and Hanson (1999) for the U.S. economy, because outsourcing seems to exert a significant negative marginal effect on real value added per low-skilled worker. However, to some extent this coincides with a similar but insignificant finding by Siegel and Griliches (1991; for the United States as well). In contrast, our long-run parameter estimates reveal a positive impact of outsourcing on real value added per low-skilled worker, which supports Feenstra and Hanson's (1999) finding and generally fits well into the literature on the productivity effects of outsourcing. There is evidence that international outsourcing augments physical capital and high-skilled labor (both relative to low-skilled labor) to roughly the same extent in the short run as well as the long run.

For our sample of countries and the underlying level of aggregation, no data on skill-specific factor rewards, hours per worker, and the actual capital services are available. Therefore, we have to rely on employment figures and estimates of the capital stocks as our controls. This might have some impact on the results. Our preliminary findings suggest that low-skilled labor productivity growth in European manufacturing--besides unobserved influences in the short run was mainly induced by the change in physical capital stocks and skill upgrading rather than fragmentation of production across borders. But the impact of international outsourcing becomes important in the long run. Future research--especially at the firm level--could help provide deeper insights into the role of outsourcing on productivity.

REFERENCES

Arndt, Sven W. "Globalization and the Open Economy." North American Journal of Economics and Finance, 8, 1997, 71-79.

Baltagi, Badi H. Econometric Analysis of Panel Data, Chichester: Wiley, 2001.

Barro, Robert J., and Xavier Sala-i-Martin. Economic Growth, New York: McGraw-Hill, 1995.

Burda, Michael C, and Barbara Dluhosch. "Globalization and European Labor Markets." CEPR Discussion Paper, 1992.

Christensen, Laurits R., Dale W. Jorgensen, and Lawrence J. Lau. "Transcendental Logarithmic Production Frontiers." Review of Economics and Statistics, 55, 1973, 28-45.

Deardorff, Alan V. "Fragmentation in Simple Trade Models." North American Journal of Economics and Finance, 12, 2001, 121-37.

Egger, Hartmut. "International Outsourcing in a Two-Sector Heckscher-Ohlin Model." Journal of Economic Integration, 17, 2002, 687-709.

Egger, Hartmut, and Peter Egger. "Cross-Border Sourcing and Outward Processing in EU Manufacturing." North American Journal of Economics and Finance, 12, 2001, 243-56.

--. "How International Outsourcing Drives Up Eastern European Wages." Weltwirtschaftliches Archiv, 138, 2002, 83 96.

--. "Outsourcing and Skill-Specific Employment in a Small Economy: Austria and the Fall of the Iron Curtain." Oxford Economic Papers, 55, 2003, 625-43.

Egger, Hartmut, and Josef Falkinger. "The Distributional Effects of International Outsourcing in a 2x2-Production Model." North American Journal of Economics and Finance, 14, 2003, 189-206.

Feenstra, Robert C., and Gordon Hanson. "Foreign Investment, Outsourcing, and Relative Wages," in Political Economy of Trade Policy." Essays in Honor of Jagdish Bhagwati, edited by Robert C. Feenstra, Gene M. Grossman, and Douglas A. Irwin. Cambridge, Mass.: MIT Press, 1996.

--. "The Impact of Outsourcing and High-technology Capital on Wages: Estimates for the United States, 1979-1990." Quarterly Journal of Economics, 114, 1999, 907-40.

Fontagne, Lionel, Michael Freudenberg, and Deniz UnalKesenci. Statistical Analysis of EC in Intermediate Products. Paris: EUROSTAT, 1996.

Glass, Amy J., and Kamal Saggi. "Innovation and Wage Effects of International Outsourcing." European Economic Review, 45, 2001, 67-86.

Greenaway, David, Robert Hine, and Peter Wright. "An Empirical Assessment of the Impact of Trade on Employment in the United Kingdom." European Journal of Political Economy, 15, 1999, 485-500.

Greenaway, David, Robert Hine, and Peter Wright. "Further Evidence on the Effect of Foreign Competition on Industry Level Wages." Weltwirtschaftliches Archiv, 136, 2000, 522-38.

Griliches, Zvi. R&D and Productivity: The Econometric Evidence. Chicago: University of Chicago Press, 1998.

Hine, Robert, and Peter Wright. "Trade with Low Wage Economies, Employment and Productivity in UK Manufacturing." Economic Journal, 108, 1998, 1500-10.

Hofer, Helmut, Christian Keuschnigg, and Wilhelm Kohler. "A Dynamic Applied General Equilibrium Model for the Austrian Economy with Special Emphasis on the Eastern Enlargement." University of Linz Working Paper No. 9724. University of Linz, 1997.

Jones, Ronald W. Globalization and the Theory of Input Trade. Cambridge, Mass.: MIT Press, 2000.

Jones, Ronald W., and Henryk Kierzkowski. "A Framework for Fragmentation," in Fragmentation." New Production Patterns in the Worm Economy, edited by Sven W. Arndt and Henryk Kierzkowski. New York: Oxford University Press, 2001.

Jorgenson, Dale W., Frank M. Gollop, and Barbara Fraumeni. Productivity and US Economic Growth, 1979-1985. Cambridge, Mass: Harvard University Press, 1987.

Keller, Wolfgang. "Trade Patterns, Technology Flows, and Productivity Growth." Worm Bank Economic Review, 14, 2000, 17-47.

Kohler, Wilhelm. "A Specific Factors View on Outsourcing." North American Journal of Economics and Finance, 12, 2001, 31-53.

Krusell, Per, Lee E. Ohanian, Jose-Victor Rios-Rull, and Giovanni L. Violante. "Capital-Skill Complementarity and Inequality: A Macroeconomic Analysis." Federal Reserve Bank of Minneapolis Research Department Staff Report, 239, 1997.

Layard, Richard, and Stephen Nickell. "Is Unemployment Lower if Unions Bargain over Employment." Quarterly Journal of Economics, 105, 1990, 737-87.

Learner, Edward E. Sources of International Comparative Advantage, Cambridge, Mass: MIT Press, 1984.

Murphy, Kevin M., and Robert H. Topel. "Estimation and Inference in Two-Step Econometric Models." Journal of Business and Economic Statistics, 3, 1985, 370-79.

Perroni, Carlo, and Thomas F. Rutherford. "Regular Flexibility of Nested CES Functions." European Economic Review, 39, 1995, 335-43.

Pirotte, Alain. "Convergence of the Static Estimation toward the Long-Run Effects of Dynamic Panel Data Models." Economics Letters, 63, 1999, 151-58.

Siegel, Donald, and Zvi Griliches. "Purchased Services, Outsourcing, Computers, and Productivity in Manufacturing." NBER Working Paper No. 3678, Cambridge, Mass, 1991.

Slaughter, Matthew J. "Production Transfer within Multinational Enterprises and American Wages." Journal of International Economies, 50, 2000, 449-72.

Venables, Anthony J. "Fragmentation and Multinational Production, " European Economic Review, 43, 1999, 935-45.

(1.) For a theoretical treatment of international outsourcing, see (among others) Arndt (1997), Deardorff (2001), Feenstra and Hanson (1996), Egger (2002), Egger and Falkinger (2003), Jones (2000), Jones and Kierzkowski (2001), Kohler (2001), and Venables (1999). Egger and Egger (2001) investigate both theoretically and empirically the impact of international outsourcing on the skill intensity in production.

(2.) For an assessment of the impact of international outsourcing on Eastern European wages see Egger and Egger (2002). Egger and Egger (2003) investigate the impact of outsourcing to Central and Eastern European countries on relative employment of high-skilled and low-skilled labor in Austrian manufacturing industries.

(3.) "We do not normally think of, say, the import of steel by a U.S. automobile producer as outsourcing. But it is common to consider the purchase of automobile parts by that company as outsourcing, especially if the parts were formerly made by the same company, or at least purchased in the United States" (Feenstra and Hanson 1999, p. 924).

(4.) In terms of the adjusted [R.sup.2], the translog model performs worse for our data. Krusell et al. (1997) stress the advantages of a CES production function as compared

(5.) Assuming a homogenous production function is in line with the theoretical literature dealing with international outsourcing in the orthodox Heckscher-Ohlin model. Compare among others Arndt (1997); Deardorff (2001); Jones (2000); Egger (2002); Egger and Egger (2001). Feenstra and Hanson (1996) consider a three-factors model and assume that capital and labor substitute in a Cobb-Douglas production function at the intermediate goods level. Final assembly is also formalized as a Cobb-Douglas technology. However, in favor of a more general specification we decided for a CES production function.

(6.) According to the availability of data, we have to assume that the effect of D,J is comprehensively accounted for by [A.sub.i].

(7.) See Griliches (1998) for a similar approach when estimating the productivity effects of research and development.

(8.) Feenstra and Hanson (1996) do not allow for any substitution of the two types of labor. Krusell et al. (1997) emphasize that the elasticity of substitution between skilled labor and unskilled labor is higher than the elasticity of substitution between skilled labor and capital. However, this is not a contradiction to our assumption of perfect substitution between efficiency units of the two types of labor. The advantage of our approach is stressed in the discussion following equation (6).

(9.) Compare Barro and Sala-i-Martin (1995, p. 54) for a similar formulation of efficiency units of capital and labor but outside the context of outsourcing.

(10.) This is similar to Leamer (1984), who suggests a rate of 13% at the aggregate country level.

(11.) This is the share of intermediates in total intermediate usage, which stems from producers of the same industry classification. Without any doubt, our measure is wider than Feenstra and Hanson's (1999), because the lowest level of aggregation of European data is two-digit rather than four-digit. However, we call the measure narrow, because we focus on outsourcing of goods production within the same two-digit industry. This is conceptually equivalent to Feenstra and Hanson (1999).

(12.) The discussion on the factor bias and sector bias of international outsourcing (compare Kohler 2001; for an overview) is also related to the shift and rotation of unit isocost curves in the factor price diagram as explored in Egger (2002) and Egger and Falkinger (2003).

(13.) However, we find that the present models face excellent convergence properties. Especially, the parameter estimates for [[beta].sub.A], [[beta].sub.k], and [[beta].sub.h] are not sensitive to the choice of parameter values, even in terms of their sign.

(14.) Unfortunately, we cannot control for other sources of skill and capital upgrading. However, given the short time period, the time variance of alternative influences should be rather small.

(15.) Using four-digit industry data, Siegel and Griliches (1991) find that productivity growth in U.S. manufacturing was negatively (but insignificantly) related to the change in the share of imported materials. However, the latter is only a broad measure of outsourcing as compared to the narrow one used in the present article. This result seems also consistent with the finding from a linearized specification of Model 1. The results from the Taylor approximated models are not presented for the sake of brevity. However, they are available on request.

(16.) Pirotte (1999) demonstrates that the cross-sectional estimates are close to the long-run effects also for fixed time and large cross-sectional dimension. This is especially a useful result, if the time dimension is too short to estimate a dynamic model, which is the case in our application.

(17.) Bulgaria, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovenia, and the Slovak Republic.

(18.) The corresponding test statistic amounts to 32.29 and is distributed as [chi square](7).

(19.) For a discussion of resource requirements for international outsourcing compare for example Jones and Kierzkowski (2001) and Glass and Saggi (2001).

(20.) We make use of ([differential]ln[Q.sub.tic]/[differential]ln [H.sub.tic])/([differential]ln[Q.sub.tic]/[differential]ln[L.sub.tic]) = [h.sub.tic][([differential][Q.sub.tic]/[differential][H.sub.tic])/ ([differential][Q.sub.tic]/[differential][L.sub.tic])], which motivates us to look at the ratio of marginal products instead of the ratio of the respective elasticities.

(21.) Even in the presence of trade unions, workers earn their marginal product as long as firms have the "right-to-manage" employment (e.g., Layard and Nickell 1990).

ABBREVIATION

CES: Constant Elasticity of Substitution

HARTMUT EGGER and PETER EGGER *

* We thank Michael Pfaffermayr and two anonymous referees for helpful comments and suggestions.

H. Egger: Socioeconomic Institute, University of Zurich, Zurichbergstr. 14, CH-8032 Zurich, Switzerland; and Leverhulme Centre for Research on Globalisation and Economic Policy. Phone 41-1-63-42303, Fax 41-163-44996, E-mail egger@wwi.unizh.ch

P. Egger: Ifo Institute of Economic Research, Poschingerstrasse 5, D-81679 Munich, Germany; University of Munich and Leverhulme Centre for Research on Globalisation and Economic Policy. Phone 49-89-9224-1238, Fax 49-89-985369, E-mail egger@ifo.de

TABLE 1

Regression Results for CES Specification (Dependent Variable Is Real
Value Added per Low-skilled Employee)

 Nonlinear Fixed Effects

Parameters (a) Model 1 Model 2

[[beta].sub.A] -0.605 ** (0.250) -0.613 ** (0.261)
[[beta].sub.k] 0.703 ** (0.287) 0.704 ** (0.300)
[[beta].sub.h] 0.704 ** (0.287) 0.704 ** (0.300)
[delta] 0.045 ** (0.020) 0.040 * (0.021)
r 0.860 *** (0.015) 0.860 *** (0.017)
[rho] 0.849 *** (0.142) 0.851 *** (0.167)
[[beta].sub.0] 0.020 (0.177) 0.020 (0.199)
Observations 992 755
Adjusted [R.sup.2] 0.935 0.934
Time effects (b) 7.19 *** 9.26 ***
Country effects (c) 36.59 *** 29.3 ***
Industry effects (d) 19.70 *** 16.07 ***

 Nonlinear Cross-Sectional

Parameters (a) Model 3 Model 4 (e)

[[beta].sub.A] -0.014 *** (0.004) -0.012 *** (0.005)
[[beta].sub.k] 0.021 *** (0.006) 0.015 * (0.008)
[[beta].sub.h] 0.010 *** (0.002) 0.016 *** (0.003
[delta] 0.406 *** (0.092) 0.100 (0.118)
r 0.997 *** (0.017) 0.987 *** (0.015)
[rho] 0.179 (0.112) 0.605 ** (0.306)
[[beta].sub.0] -1.300 *** (0.258) -2.032 *** (0.466)
Observations 225 225
Adjusted [R.sup.2] 0.885 0.900
Time effects (b) -- --
Country effects (c) -- --
Industry effects (d) -- --

(a) Standard errors in parentheses.

(b) Distributed as F(5, 949) in Model 1 and as F(4, 713) in Model 2.

(c) Distributed as F(11, 949) in Model 1 and as F(11, 713) in Model 2.

(d) Distributed as F(20, 949) in Model 1 and as F(20, 713) in Model 2.

(e) Instrumental variable regression assuming the outsourcing intensity
to be endogenous and using the following instruments: average unit
labor costs in the remaining EU and the same industry, log (c.i.f./
f.o.b.) of the country- and industry-specific EU final goods exports,
log (c.i.f./f.o.b.) of the country- and industry-specific EU
intermediate goods exports, log (c.i.f./f.o.b.) of the country- and
industry-specific final goods exports to the CEEC, log (c.i.f./f.o.b.)
of the country- and industry-specific intermediate goods exports to the
CEEC. The associated coefficient of the canonical correlation between
the endogenous variable and the set of instruments is 0.2.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

TABLE 2

Regression Results for Translog Specification (Dependent Variable Is
Real Value Added per Low-skilled Employee)

 Fixed Effects Between

Regressors (a) Model 5 Model 6

Outsourcing intensity: O 0.0009 (0.0009) 0.0088 ***
 (0.0031)
Log physical capital: (In K) x O 0.0002 (0.0003) 0.0031 ***
 (0.0011)
Log high-skilled labor: (In H) x O 0.0018 *** 0.0069 ***
 (0.0003) (0.0011)
Log low-skilled labor: (in L) x O -0.0022 *** -0.0101 ***
 (0.0003) (0.0009)
[(ln K).sup.2] x [0.sup.2] -1.49 x -2.52 x
 [10.sup.-7] [10.sup.-7]
 (1.44 x (7.30 x
 [10.sup.-7]) [10.sup.-7])
[(ln H).sup.2] x [O.sup.2] 1.43 x -1.59 x
 [10.sup.-6] *** [10.sup.-6]
 (2.75 x (1.75 x
 [10.sup.-7]) [10.sup.-6])
[(ln L).sup.2] x [O.sup.2] 2.34 x 5.41 x
 [10.sup.-6] *** [10.sup.-6]
 (2.80 x (1.36 x
 [10.sup.-7]) [10.sup.-6])
(ln L) x (ln K) x [O.sup.2] -5.36 x 3.23 x
 [10.sup.-7] [10.sup.-8]
 (3.98 x (2.45 x
 [10.sup.-7]) [10.sup.-6])
(ln L) x (In H) x [0.sup.2] -3.78 x 1.17 x
 [10.sup.-6] *** [10.sup.-6]
 (4.54 x (2.37 x
 [10.sup.-7]) [10.sup.-6])
(ln H) x (ln K) x [0.sup.2] 6.88 x 6.16 x
 [10.sup.-7] [10.sup.-8]
 (4.98 x (3.05 x
 [10.sup.-7]) [10.sup.-6])
Observations 992 225
Adjusted [R.sup.2] 0.884 0.587
Time effects (b) 4.70 *** --
Country effects (c) 185.81 *** --
Industry effects (d) 78.98 *** --

(a) Standard errors in parentheses.

(b) Distributed as F(5, 945).

(c) Distributed as F(11, 945).

(d) Distributed as F(20, 945).

** Significant at 5%.

*** Significant at 1%.

TABLE 3

Average Annual Growth of Outsourcing and Productivity of Low-Skilled
Labor in the EU (Average Annual Change in Percent, 1993-97)

 Productivity
 Outsourcing of Low-
NACE 2-Digit Industry Intensity Skilled Labor

Manufacture of food products and 3.15 4.94
beverages
Manufacture of textiles 7.03 9.45
Manufacture of wearing apparel; 15.02 12.31
dressing; dyeing of fur
Tanning, dressing of leather; 8.08 8.57
manufacture of luggage
Manufacture of wood and of products of -7.16 10.78
wood and cork, except furniture
Manufacture of pulp, paper and paper 1.44 15.92
products
Publishing, printing, reproduction of -0.83 5.66
recorded media
Manufacture of coke, refined petroleum -0.62 -0.02
products and nuclear fuel
Manufacture of chemicals and chemical 5.28 4.00
products
Manufacture of rubber and plastic 3.74 6.91
products
Manufacture of other non-metallic 4.66 1.82
mineral products
Manufacture of basic metals 2.33 14.34
Manufacture of fabricated metal -0.28 11.04
products, except machinery and
equipment
Manufacture of machinery and 1.70 17.17
equipment n.e.c.
Manufacture of office machinery and -4.31 12.02
computers
Manufacture of electrical machinery and 4.31 17.66
apparatus n.e.c.
Manufacture of radio, television and 1.20 13.85
communication equipment and apparatus
Manufacture of medical, precision and 9.83 6.98
optical instruments, watches and clocks
Manufacture of motor vehicles, trailers 6.02 7.82
and semi-trailers
Manufacture of other transport 1.18 8.16
equipment
Manufacture of furniture; manufacturing -8.47 10.79
n.e.c.
Total manufacturing 3.22 9.28

 Predicted Productivity
 Observed--Simulated (a)

 Cross-
 Fixed Effects Sectional
NACE 2-Digit Industry (Short-Run) (Long-Run)

Manufacture of food products and -0.16 1.16
beverages
Manufacture of textiles -0.14 6.31
Manufacture of wearing apparel; -0.03 0.32
dressing; dyeing of fur
Tanning, dressing of leather; -0.03 -3.95
manufacture of luggage
Manufacture of wood and of products of -0.04 10.87
wood and cork, except furniture
Manufacture of pulp, paper and paper -0.11 22.32
products
Publishing, printing, reproduction of -0.07 0.65
recorded media
Manufacture of coke, refined petroleum -0.01 14.83
products and nuclear fuel
Manufacture of chemicals and chemical -0.05 -1.55
products
Manufacture of rubber and plastic -0.03 0.14
products
Manufacture of other non-metallic -0.05 0.85
mineral products
Manufacture of basic metals -0.10 18.11
Manufacture of fabricated metal -0.11 2.29
products, except machinery and
equipment
Manufacture of machinery and -0.14 6.51
equipment n.e.c.
Manufacture of office machinery and 0.00 -0.46
computers
Manufacture of electrical machinery and -0.13 10.07
apparatus n.e.c.
Manufacture of radio, television and -0.04 17.38
communication equipment and apparatus
Manufacture of medical, precision and -0.01 3.45
optical instruments, watches and clocks
Manufacture of motor vehicles, trailers -0.1 11.17
and semi-trailers
Manufacture of other transport -0.01 0.24
equipment
Manufacture of furniture; manufacturing -0.05 1.78
n.e.c.
Total manufacturing -1.39 6.01

(a) Calculated as predicted productivity due to observed outsourcing
minus predicted productivity due to simulated counterfactual
outsourcing in terms of the former.