Offshoring, R&D and Home Country Demand for Skilled Labor

Offshoring, R&D and Home Country Demand for Skilled Labor Patrik Karpaty, Örebro University, Sweden Ari Kokko Stockholm School of Economics, Sweden Patrik Gustavsson Tingvall, Stockholm School of Economics, Sweden Abstract 1 We investigate the effects of offshoring on firms R&D and demand for skilled labor using Swedish firm level data. Our results emphasize the importance of taking into account heterogeneous effects on firms. While there is no robust aggregate effect from offshoring of goods and services we find significant effects when separating between goods and material offshoring and between the manufacturing and the service sector. The results also indicate that it is more likely to find any effects in firms that undergo large structural changes in their international production distribution. To the extent that positive effects from offshoring on R&D and relative demand for skilled labor are present they are confined to firms that extensively change their material offshoring pattern. Using a dynamic approach and controlling for endogeneity in inputs gives support for a positive impact on the demand for skilled labor and R&D. There are also negative effects on home R&D that accrue to firms with initially low R&D intensities while offshoring in R&D-intensive firms increases domestic R&D. Acknowledgements: Financial support from the Nordic Innovation Center (NICe) and Forskningsrådet för arbetsliv och Social vetenskap (FAS) is gratefully acknowledged. E-mail: Patrik.gustavsson@hhs.se ; Patrik.karpaty@oru.se 1

1. Introduction During the last years, offshoring has received increasing attention in both the public debate and in academic research. Although many economists argue that offshoring is a normal part of international trade, the debate in the rich countries clearly reveals that there are worries about its consequences. These concerns are mostly connected to a fear of job losses as firms move production abroad. While low-skilled workers are obviously most vulnerable to the competition from offshoring, it poses a challenge also for other groups. Hence, as OECD (2005) points out, not only unskilled jobs but also ICT jobs and other tasks requiring skilled labor perhaps even R&D are at stake. It is therefore interesting to note that existing studies on offshoring and labor demand have mainly focused on detecting the expected shifts in demand from unskilled to skilled workers: the effects on the composition of skilled labor and R&D in the home country have largely been ignored. This is unfortunate. R&D is one of the main determinants of firm competitiveness, and it is of prior interest to ask whether offshoring in general, some specific kinds of offshoring, or offshoring in certain types of firms, lead to a relocation of R&D from the home country to other locations. To shed some light on these questions, we use highly disaggregated data from Swedish industry to decompose the relation between offshoring and labor demand in several ways. Firstly, we distinguish between several categories of labor, and examine whether offshoring affects the distribution of skilled labor between R&D and other advanced tasks. Secondly, we look separately at the consequences of offshoring in the manufacturing and service sectors. Thirdly, we distinguish between service offshoring and material offshoring. Since offshored services are intangibles that typically cannot be stored, service offshoring may require closer commutation between the mother firm and the foreign subcontractor than what is necessary for material offshoring. Fourthly, in an attempt to deal with the problem to correctly measure offshoring we adopt a dynamic approach. We seek to answer the question, are possible effects from offshoring confined only to firms that undergo large changes in their international production distribution. This paper is related to the literature on offshoring and labor demand and the offshoring models proposed by Grossman and Helpman (2002a, 2002b, 2003) and Grossman and Rossi- Hansberg (2006). As mentioned, a number of earlier studies have analyzed the relation between labor demand and offshoring. Amiti and Wei (2005b, 2005b) found small 2

employments effects of US service offshoring. Other US studies, pointing in the same direction are Landefeld and Mataloni (2004) and Borga (2005), who look at foreign and domestic employment in US multinationals. Crinò (2006), looking at relative labor demand, found outsourcing to favor skilled US workers. For a survey on US experiences, see e.g. Mankiw and Swagel (2006). A frequently cited study is the McKinsey Global Institute Report (2003), which concludes that the job losses caused by offshoring are marginal in comparison with the job losses caused by business cycles. Results similar to those for the US are also found in Swedish studies (e.g. Andersson and Karpaty 2007, Ekholm and Hakkala 2006, Mattila and Strandell 2006), German studies (Falk and Kobel, 2002), and for the UK (Amiti and Wei 2005a, Görg et al. 2005b). The magnitude of job losses caused by offshoring is estimated by Aubert and Sillard (2005) who conclude that 0.35 percent of total employment in the relevant industries were lost due to offshoring. However, there is reason to believe that the job losses and wage effects caused by offshoring may be non-trivial in some industries. For instance, in some early studies, Feenstra and Hanson (1996, 1999) found that about 15 percent of the increased relative wage for skilled workers in the US during the period 1979-1990 could be attributed to offshoring activities. Hence, for the advanced countries, it seems clear that offshoring tends to shift labor demand from relatively unskilled to skilled workers, although the magnitude of the shift seems to be relatively small. 2 Less is known about how offshoring may alter home country R&D and the demand for various types of skilled labor. The remainder of the paper is organized as follows: In section 2 we overview the relevant theory of offshoring and discuss measurement problems. Section 3 presents the data and the empirical methodology. A descriptive analysis is provided in section 4 and regression results are given in section 5. Section 6 concludes. 2 The empirical support from Sweden suggests that the negative effects on relatively unskilled workers are very small (Ekholm and Hakkala 2006); Andersson and Karpaty 2007). Only a couple of thousands employees should have been affected so far according to their estimates. 3

2. Offshoring: theory and measurement 2.1 Theory The theory of offshoring is based on the idea that firms, under certain circumstances, benefit from a vertical defragmentation of the production process. 3 In a closed economy setting, firms may choose between in-house production and subcontracting (or outsourcing), as discussed by McLaren (2000) and Grossman and Helpman (2002b). The optimal choice depends on how strong are the motives for internalization this, in turn, is discussed in the propertyrights theories of Grossman and Hart (1986): for surveys, see e.g. Spencer, (2005), Trefler (2005) and Helpman (2006). In an open economy context, firms may also locate production abroad. This also entails a choice between in-house production in the form of foreign direct investment, FDI and subcontracting in this case, outsourcing. Outsourcing allows the firm to avoid the fixed set-up cost that is related to FDI. However, outsourcing requires the firm to formulate a contract with a foreign subcontractor. One purpose of the contract is to protect the mother company from leakages of know-how and strategic information. Such contracts can be rather costly to formulate. Therefore, most offshoring models focus on the tension between taking on the fixed FDI cost or the transactions costs needed for subcontracting. This tension may be characterized by the Grossman and Helpman (2003) North-South outsourcing model. In their set-up, final goods producers are located in the high-wage North and intermediate goods suppliers are located in the low-wage South. For a given Northern firm, the probability of international outsourcing increases with thickness of the intermediate good market. The thicker the Southern market and the more standardized the product, the greater is the probability of finding a Southern supplier that fulfills the Northern producer s requirements. Other theoretical contributions that considers the choice between FDI and outsourcing are Grossman and Helpman (2002a), Antras (2003), and Feenstra and Hanson (2005). One conclusion from these studies is that sensitive tasks operations that expose strategic information are not as easily outsourced to outsiders as standardized job tasks: the costs for subcontracting increase with the complexity of the outsourced activity. Another aspect of outsourcing is that it affects relative labor costs and efficiency. Grossman and Rossi-Hansberg (2006) show how falling costs of outsourcing, or in their 3 The term offshoring refers here to the relocation of production of goods or services to another country. Offshoring does not separate whether the production is external or internal to the firm. We measure offshoring by looking at intermediate goods and services imports. 4

terminology trade in tasks, have an impact on factor prices through three different mechanisms: productivity improvements, supply effects, and a relative-price effect. The net effect of outsourcing on the wages of different labor categories can be either positive or negative, depending on the relative size of these effects. Hence, it is largely an empirical question how increased offshoring alters labor demand and relative wages. 2.2 Definitions A number of different terms appear in the offshoring debate. Outsourcing involves the subcontracting of an internal company function to an outside firm. However, outsourcing is more than an arms-length transaction. It also involves a transfer of management control, decision making, and often also transfers of firm-specific knowledge to the external supplier. Hence, outsourcing involves elements such as two way information exchanges, coordination, and trust. In an open economy, outsourcing can take place domestically or across international borders, and it can involve external actors or other plants belonging to the outsourcing corporation. In the public debate, most attention is focused on situations where there is a change of suppliers from domestic to foreign sources. 4 This phenomenon has been referred to as international outsourcing by Görg et al. (2005), while Criscuolo and Leaver (2005) talk about offshore outsourcing, and the United States Government Accountability Office (2004) uses offshoring. To avoid confusion, we will use the term offshoring to refer to outsourcing to foreign locations, in line with the definition by Baghwati, Panagariya and Srinivasan (2004): purchases of services abroad with the supplier and buyer remaining in their respective locations, including trade with either a foreign affiliate or an external overseas supplier. 5 When we specifically mean offshoring within the corporation, we will refer to FDI. Other contributions to the literature also discuss more specific phenomena such as outsourced offshoring (offshoring to independent foreign firms), domestic outsourcing, and insourcing (domestic outsourcing within the corporation): we will avoid using these terms. 6 2.3 Measures An exact measure of offshoring would require detailed information about the firm s input transactions over time. To be precise, inputs that were never produced internally should not be 4 Hagsten et al. (2006). 5 See also Bjerring Olsen (2006). 6 See Ekholm (2006) for a careful discussion about the concept offshoring. 5

included in measures of outsourcing, nor should medium and long term investment goods. It would also be necessary to keep track of production locations, movements of plants, and other changes in operations. Such detailed information is rarely available. Data and measurement problems have triggered researchers to find various proxies. One commonly used proxy suggested by Feenstra and Hanson (1996, 1999) is to define a narrow measure of offshoring as the ratio of imports from the firm s own industry to value added, and a wider measure as the ratio of total imports to value added: Crinò (2006), Amiti and Wei (2005a, 2005b, 2006) and Ekholm and Hakkala (2006) are examples of studies using this approach. As the availability of firm level data has improved, it has also become possible to use firm level measures of offshoring, such as the ratio of imports of materials or services to total sales, total wages, or other aggregates. 7 One weakness of such data is that there is typically no information on whether these tasks were previously executed within the firm. To some extent, this problem can be handled by looking at changes in imports from year to year. This is an approach that will be deployed in our analysis. 8 3. Data and empirical strategy 3.1 Empirical strategy We base our analysis of the impact of offshoring on relative labor demand on the approach of Berman et al. (1994). The model we use is based on a Translog cost function where different categories of labor (defined according to skill levels) are variable factors of production and physical capital is treated as a fixed factor. 9 The key assumption is that technological change and offshoring have different effects on the productivity of different skill-groups, and may alter both the relative demand for different types of labor and the firm s R&D-intensity. Relative labor demand can be estimated using the following regression equation: 7 See e.g. Görg et al. (2005), Andersson and Karpaty (2007), Criscuola and Leaver (2005), and Hagsten et al. (2006). 8 A possible drawback by looking at changes in import is that price and currency fluctuations may drive a wedge between changes in nominal and real import values. 9 This method has also been applied by e.g. Machin and Van Reneen (1998). Hansson (2000), and Ekholm and Hakkala (2007) and Andersson & Karpaty (2007). 6

Θ jit = J K γ js ws ) + β j ln( Qi ) + β j ln( Ki ) + ξ jk Zik + λt + s ln( ε (1) k it where is firm i s wage share allocated to workers belonging to skill group j, w s is the wage for workers belonging to skill group j, Q is firm output, Z is a set of variables capturing factor biased technological change such as e.g. offshoring, λ t is a period dummy capturing periodspecific skill upgrading effects that are common to all firms, and ε it is the error term. Given that the wages for workers of a given skill group should be the same for all firms, the inclusion of time dummies will be linearly dependent. We therefore follow the common strategy of dropping wages from the regression analysis. A positive (negative) that capital and labor are complements (substitutes) to each other. Estimates of β j indicates ξ jk show whether technological change is biased for ( ξ > 0 ) or ( ξ < 0 ), where labor belongs to skill group j. To capture technological change, we use a set of offshoring variables complemented with a set of control variables that affect the firm s technological choices. To be precise, we will use the firm s export share, profit ratio and the degree of competition facing the firm as supplementary control variables. Offshoring is measured as the ratio of the firm s imports of services and/or materials to total sales. To analyze how offshoring affects labor demand and firm R&D we explore effects on two dependent variables. Firstly, on the basis of their educational attainment, labor is divided into two skill groups lower secondary plus upper secondary and tertiary education. These are used to define our first dependent variable. 10 Secondly, because of the lack of information on the wage cost of R&D-workers, we use the ratio of R&D to total wage costs as a proxy for the proportion of the labor force allocated to R&D. Finally, offshoring can influence the demand for skilled labor either toward job tasks such as information, marketing, and accounting, or towards R&D. Using the ratio of R&D-expenditures to total wage cost for skilled labor as a dependent variable, we hope to detect shifts in the relative demand for two categories of skilled labor. Since offshoring is defined as the relocation of production at home to a foreign sub contractor (or an affiliate abroad) it seems natural to think of the potential for relocation of labor and R&D at home to be larger in firms that restructure their production more rapidly 10 Since the estimates for the less-skilled group is simply a mirror of the estimates for the skilled labor group they are not included in the analysis. Beyond the scope of this paper is a possibility to divide labor into labor with lower secondary plus upper secondary education in the analysis. 7

than other firms. 11 Following this reasoning, changes in the amount of intermediate inputs could possibly help us further explore the causality between offshoring and the activities that remain at home. For this reason we divide each firm s number of years in the panel into two periods. For each period we compute the average offshoring intensity for the two periods. Two separate dummies then indicate whether the firm doubles its imports of intermediate goods or services respectively. 12 We carry out the estimation of equation (1) with two panel data estimation techniques that provide consistent estimators: In order to control for correlations between the residuals, we estimate the models using GLS. Since we assume unobserved heterogeneity and simultaneity in offshoring and other inputs, we also utilize system GMM. The choice of using system GMM and not the differenced GMM is due to an assumption that the lagged levels of the regressors are poor instruments for the first-differenced regressors. 13 Therefore we use the augmented version system GMM. The system GMM should also provide more efficient estimates when the number of firms (N) is large and the number of years (T) is small as with our data. The dynamic approach chosen requires a possibility to follow firms for some years. We chose to exclude firms that survive less than three years. 14 3.2 Data The analysis is based on three register-based data sets from Statistics Sweden. The financial statistics data set (FS) contains detailed information on all Swedish private sector firms with at least 50 employees. Examples of variables included are R&D, value added, capital stock (book value), number of employees, total wages, ownership, profits, sales and industry sector. A detailed description of the variables is found in the appendix. 11 Moreover, the measure of offshoring that we use (import of intermediate goods) is just a proxy for offshoring and may also include the import of intermediates that never were produced at home. 12 We also include firms with zero offshoring in the first period and positive offshoring in the second period. About 20% of the firms with material offshoring and 15% of the firms with service offshoring went through such large restructuring during the period 1997-2005. 13 The difference GMM tends to produce inefficient estimates since it relies on first differences to eliminate unobserved firm-specific effects and then uses lagged instruments to correct for simultaneity in the firstdifferenced equations, Mairesse and Hall (1996). 14 Due to the number of lags (2-4 years) used in the analysis one may argue that we actually require an even longer time frame for each firm (5 years). Two points can be made in support of or choice to include firms that survive at least three years though. Firstly, since the panel is rather short in itself (7 years) we need to economize on observations. For example, only 15-20 % of the firms in the final sample undertake large changes in their import of intermediate goods or services. Using a balanced panel this number, as well as the representativity of the whole sample will fall drastically. Secondly, there is a greater risk for selection bias if the surviving firms are very different (superior) to those that survive less years in our panel. 8

Second, the regional labor market statistics data set (RAMS) includes labor data on all establishments for the period 1990-2005. RAMS is used to describe the labor force at the establishment level with respect to educational level and demographics. Data on imports of materials are collected at the firm level by Statistics Sweden for all firms with an import value above a 2.2 million SEK (approx. 24,000 EUR). This data is only available for a shorter time period, 1997-2007. Material imports are classified according to country of origin and item. The product classification is very detailed and allows us to distinguish between different types of goods at the five-digit level according to NACE Rev 1.1. Moreover, Statistics Sweden classifies imported materials into Major Industrial Groupings (MIG). 15 The MIG codes classifies material imports into broader classes depending on their intended use, distinguishing between production inputs, short, medium, and long term investment goods and consumption goods. This allows us to use a definition that is closer to what we aim at measuring, namely offshore production. In the analysis we will use the narrow definition of intermediate goods import that covers production inputs only. Data on imports of services are provided by the Swedish Central Bank and cover the period 1992-2002. The service import data is categorized into eight aggregated service items: financial, legal science, bookkeeping, technical, computer (ICT), transports, insurances, licenses & Royalties. In this paper we do not limit our self to any specific type of service import. 16 The data sets have been matched by unique identification codes. A longitudinal data set has been compiled by combining the identification codes and data for multiple points in time. Due to different time frame for the variables used we have to limit the main analysis to cover only the years 1997-2002. An analysis on a longer time period is only available for material imports, not service imports. 17 15 MIG - European Community classification of products: Major Industrial Groupings (NACE rev1 aggregates). 16 The reason for not excluding any kind of service import is that we only have information on an aggregate level on what kind of services that is imported. This means that we cannot easily separate out services that typically are intermediates in the business at home. 17 The reason to not use updated data for the service import is due to new collection methods by Statistics Sweden. Imports of services are now only collected for a stratified sample of firms, which makes it difficult to follow single firms before and after 2003. 9

4. Descriptive statistics Figure 1 and 2 illustrate how the aggregate offshoring intensities in the service and manufacturing sectors developed over the period 1997-2002. It can be noted that offshoring of materials is larger than offshoring of services but that service offshoring in the service sector is increasing in importance. 18 Table A1 provides the absolute number and percentage shares of offshorers in Sweden in 2000. More than half of the sample firms import intermediate materials or services. This high share is largely related to the fact that smaller firms (less than 50 employees) are left out of the analysis. In Criscuolo et al. (2008), who look at a broader sample of Swedish firms (including also those with 20-50 employees), service offshoring appears in only one-fifth of the firms. 19 Table A1 also shows that the offshoring firms in the sample account for a large proportion of employment and value added in both manufacturing and services: in other words, offshoring firms are on average larger than firms that do not offshore. Another distinctive feature of the offshoring firms is that they are relatively R&D-intensive. This indicates the importance of firm-level scale economies for these firms. Scale economies are created by investments in R&D and product development, resulting in knowledge capital that can be used in the firms producing affiliates all over the world. 20 Table A2 shows the distribution of offshoring firms across 15 industries at the two digit level in 2000. The overall picture is characterized by large heterogeneity between different industries. The largest share of offshorers is found in manufacturing (sector codes 15-36) and in the wholesale and retail trade industries. Unsurprisingly, offshoring is less common in sectors like hotels and restaurants, post, telecom, and storage, i.e. for firms operating in nontradable sectors. Moreover, earlier literature stresses the need to take into account firm level heterogeneity. This means that we should expect offshoring intensities to vary across firms depending on productivity, composition of skilled workers, and other firm-level characteristics. 18 This result is in sharp contrast to previous findings for Sweden by Andersson and Karpaty (2007) who argued that service offshoring in the manufacturing sector had almost tripled. The divergence in results is explained by differences in the definition of offshoring intensity. In their study, Andersson and Karpaty (2007) used the inputs of intermediates rather than total sales as the denominator. 19 In Criscuolo et al (2007) the definition of offshoring is limited to service offshoring and does not include materials (narrow) offshoring. If we exclude the materials offshoring, the number of offshorers is reduced and constitutes less than 50% of the sample. 20 Economies of scale at the firm level are different from economies of scale at the plant level. Economies of scale at the firm level are generated by for example investments in R&D that materialize into knowledge capital. This capital can later be used by all producing affiliates that belong to the group in Sweden or in other countries. Economies of scale at the plant level means that the unit cost falls with the size of production in the local plant. 10

Table A3 compares some of these firm characteristics for offshorers and non-offshorers in manufacturing and services in 2000. By subtracting log deviations averages for the offshoring firms from the corresponding variables of the non offshoring firms for each variable it seems clear that offshorers have on average higher labor productivity measured as value added per employee, deflated by the industry producer price index than non-offshoring firms. At the same time, offshorers are also more human capital intensive proxied by the proportion of employees with more than secondary education and more capital intensive (measured as the ratio of the book value machinery and buildings to employment) than non-offshorers. Moreover, offshoring firms are larger (in terms of employment) and pay higher wages to both low-skilled (not in table) and high-skilled employees. They have much higher R&D intensities (measured by the ratio of R&D expenditures to total sales) and purchased materials than their non-offshoring counterparts. In Table A4, the offshorers are grouped according to ownership: the three categories we use are domestic local firms, domestic MNEs, and foreign owned firms. As expected, MNEs are more frequently found to be offshorers than purely local firms. 5. Results Table 1 presents the results of regression analysis examining the impact of offshoring on labor composition and R&D intensity in Swedish industry. The first model in Table 1 looks at the relative demand for skilled labor. The results suggest that offshoring leads to a shift in labor demand in the home country from less and medium-skilled labor categories to highly skilled labor. For a developed economy like Sweden, this is in line with expectations and results from other studies on labor demand and offshoring. [Regression Table 1 about here] However, as pointed out by e.g. Blinder (2006), the increased possibilities to engage in offshoring mean that jobs that were previously not threatened by internationalization may now face competition from abroad. In the service sector, it is not only relatively simple services like call-centers that are offshored today: even sophisticated activities like R&D are 11

included in the offshoring discussion. This suggests that that there is a need to examine more directly how offshoring affects R&D expenditures at the firm level. Model 2 and model 3 in Table 1 suggest that there is no effect on R&D, neither measured as R&D expenditures to total wage costs nor measured as R&D expenditures to the skilled labor wage bill (including technicians, experts, managers, economist and other qualified personal) from offshoring. 21 These two observations suggest that the general picture in the Swedish industry is that offshoring has very small effects on the relative importance of R&D in their home country operations. Hence, the contract theoretic arguments proposed by Grossman and Helpman (2003) that core activities like R&D are unlikely to be offshored because the subcontracting costs increase with the degree of sophistication and strategic value of activities are supported by Swedish data. 22 However, offshoring may affect the demand for skills differently in different sectors. For example, OECD (2005) estimates that within the IT and ICT sector, up to 20 percent of all jobs can potentially be affected by offshoring. Hence, there are reasons to expect a heterogeneous response pattern across firms and industries. In Table 2, we therefore extend the analysis in several directions. First, we separate material offshoring from offshoring of services. As pointed out above, it is likely that imports of services are closer to the definition of offshoring than import of materials. For example, for a manufacturing firm there may be a wide set of inputs that never have been produced in-house, and it is therefore questionable whether such import can be classified as offshored production. In the case of services where material inputs are of second order this measurement problem is smaller. To allow for a comparison between the manufacturing and the service sector we run separate regressions on these sectors. 21 The nominator is R&D expenditures and the denominator is either total wages or wages to high skilled labor. If the denominator increases in the latter case, for instance due to increased wages to skilled labor, the nominator must also increase to keep the ratio unchanged. This is very likely since a large share of the R&D expenditures constitutes wages to high skilled labor (wages to highly skilled labor is therefore double counted). Put differently, an insignificant effect on a firms R&D intensity (measured as R&D expenditures relative to skilledwages) could be a proportional increase (decrease) in wages to high-skilled labor and R&D expenditures. Keeping the high-skilled wages and R&D expenditures constant when less skilled production is offshored, the relative demand for skilled labor increases (composition effect) but R&D relative to skilled-wages does not change. If R&D intensity is instead weighted by the total wages, R&D intensity will increase as less-skilled labor (and total wages) is offshored, but remain constant if the decrease in total wages is accompanied by a proportional decrease in R&D expenditures. 22 In the theoretical model wages should be controlled for when we estimate the translog cost function. We have chosen to drop the wages since the inclusion of time dummies will be linearly dependent. This is a fairly common approach. We have additional estimates where wages have been included and the results from these remain qualitative unchanged. The results are available upon request from the authors. 12

So far we have treated the offshoring variables the import of intermediate goods and services in the firm as exogenous. However, the offshoring decision is likely to be the outcome of a process where economic variables are involved. It may be the case that offshoring and capital input are endogenous giving rise to biased estimates. To handle causality issues and endogeniety we choose a system GMM approach. Blundell and Bond (1998) propose an improved GMM for shorter panels with possible persistence in the series. The model suggests that lagged time differenced regressors should be used as instruments for the endogenous variables. 23 Blundell and Bond also show that this system panel estimator that simultaneously considers variables in both differences and levels, produces estimates that are both consistent and efficient. 24 As argued previously, the static analysis may be an ambiguous approach where we expect the effects (if any) from offshoring on R&D and skill composition to be small. We extend the analysis in Table 2 by allowing for differential effects on labor composition and R&D in firms that at last double their offshoring intensity during the period of analysis. 25 The picture becomes more complex when we differentiate between material and service offshoring, and between effects in the manufacturing and service sectors. First, we see in Table 2 that offshoring indeed tends to shift labor demand away from low and medium skilled workers. However, this is only significant at the 10% level and for large changes in material offshoring in manufacturing and for large changes in services offshoring in the service sector. The estimations reveal both a positive and a negative effect on the R&D intensity from large changes in service offshoring, depending on how R&D intensity has been computed. The combination of a negative and weakly significant effect for R&D divided by skilled-wages, but a positive and strongly significant effect for R&D divided by less skilled-wages suggests that an absolute fall in less skilled-wages is driving the results. Less-skilled production is relocated abroad while the core activities remain at home in firms that undertake large increases in their offshoring activity. The negative effect is very small in magnitude though. Presumably, the decline in R&D expenditures is simply relocation to the foreign supplier of 23 We use a similar template of the system GMM model in all IV estimations. The variables that we assume endogenous are capital input, offshoring of materials and services and the lagged dependent variable. We use the Herfindahl index, profit, export and total sales as predetermined variables. The lag length of the model is defined in the interval 2-4 years. 24 The differenced linear generalized methods of moments (GMM) estimator uses time-differenced variables in order to remove permanent unobserved heterogeneity, (Arellano and Bond 1991). When there is relatively little persistence in the series the lagged levels may be valid instruments for endogenous variables. However, when time series are short or when there is persistence over time, the Arellano and Bond GMM estimator suffers from poor precision (Blundell and Bond, 1998). 25 Various tests with different lower bounds for growth finally resulted in this threshold value. 13

smaller improvements of existing products or procedures, while the core activities remain at home. We suspect that if longer time series were available the possibilities to evaluate effects for each firm would increase substantially. We have the possibility to use a longer timer period for material but not for services offshoring. In Table A6 in the Appendix we replicate Table 2 (except for service offshoring) but using a longer time series (1997-2005). The results support the hypothesis that the effect on labor composition in the manufacturing sectors is larger for firms that carry out large changes in their production abroad. 26 Unfortunately we are unable to analyze services offshoring after 2002. Another possible distinction between firms concerns the overall R&D intensity of operations. The sample covers all firms above the threshold size value (50 employees) whether or not they perform R&D. In Sweden, the bulk of private R&D is heavily concentrated to MNEs. About 95 percent of all private R&D is performed within MNEs. Moreover, two-thirds of all private R&D is concentrated to the top 10 R&D firms. Given this asymmetry, it is relevant to ask whether R&D-intensive firms behave in the same manner as other firms. For this reason, we again estimated equation (1) but now adding interactions between the top 10% R&D firms (in each sector) and offshoring. 27 As previously we again interact the offshoring variables with large and small changes in offshoring of materials and services. These results are reported in Table 3. For the most R&D-intensive firms in the manufacturing sector we find two contradictory effects: Service offshoring is in favor of skilled labor but material offshoring is in favor of the less skilled labor. These effect are weakly significant and only confined to firms making large changes in either material or services offshoring and that belongs to the top 10% R&D intensive firms. Perhaps more interestingly is the change from insignificant to significant and positive effects for the R&D intensities (either relative to total wage costs or relative to skilled-labor wage bill) in the service sector. These effects are strongly significant and confined to firms that both belong to the top 10% R&D intensive firms and is making large changes in material and service offshoring during the examined period. 26 Moreover, the Hansen test suggests that the longer times series produces more efficient instruments since we can no longer reject the hypothesis about exogenous instruments. The problems with serial correlation (of second order) are also smaller. 27 The top 10 % R&D firms were computed for each industry at the two digit level. Our sample includes 20 industries in the manufacturing sector and 14 industries in the service sector. 14

6. Summary and conclusions During the last years, offshoring has received increasing attention in the public discourse, with the fear of job losses motivating much of the debate. The fact that even relatively skill intensive jobs are at stake has often been seen as a matter of national concern. At the same time, empirical work has shown that the aggregate effects on labor demand seem to be small. However, even if the overall effects are relatively small, it is possible that the impact may be relatively large in some sectors. In this paper we extend existing work on offshoring and focus on how offshoring affects the skilled labor composition and location of R&D at the firm level. In addition, we explore whether the effects of offshoring differ between the manufacturing and service sectors. The result suggests that offshoring typically shifts firm s relative labor demand from less skilled to skilled workers in both sectors. Offshoring may also raise the home firms R&D intensity, but these effects are confined to the services industries only. We do not find any significant effects for international relocation of R&D in the manufacturing industry. The fact that advanced operations like R&D are not offshored is in line with theoretical work suggesting that the contract cost increase with the complexity of the outsourced activity. However, the positive effects of offshoring on skilled labor demand are different in the two sectors. Offshoring of goods had a positive effect on the demand for skilled labor in the manufacturing sectors, while the corresponding effect in the services sector were confined to services offshoring. A closer look at offshoring and firms with different R&D intensities confirms that offshoring is not a reason to reduce the demand for neither skilled labor nor R&D among the most R&D-intensive firms. Instead, among these firms we find offshoring to be associated with a concentration of R&D at the domestic market. This applies for the services sectors, and for both material offshoring and service offshoring. Firms with lower initial R&D intensities, by contrast, tend to see a falling demand for skilled labor (but not R&D) as a result of services offshoring. Since these firms are typically less sophisticated than the most R&D-intensive firms, the results suggest that offshoring may be used as a tool for offshoring simple development tasks. More qualified R&D, undertaken by the more R&D-intensive firms, tends to stay in the domestic market. 15

Table 1. The impact of offshoring on firms R&D and labor composition. 1997-2002. FGLS heteroscedastic panels, panel specific AR (1) error. Simple model specification. Variable Model 1 Model 2 Model 3 Rel. demand high education R&D / tot wage bill Total offshoring 0.031 (0.002) *** -0.005 (0.052) R&D/skill labor wage bill -0.011 (1.159) ln(output) 0.031 (0.0001) *** 0.027 (0.002) *** 0.0547 (0.005) *** ln(capital stock) -0.019 (6.1e-05) *** -0.001 (0.0002) *** 0.017 (0.003) *** Time dummies yes Yes yes Log likelihood 62 210 6 330-19742 No of obs. 19 348 19 348 15 520 Note. Standard error within parenthesis ( ). *, **, ***, indicates significance on the 10, 5 and 1 percent level respectively. Offshoring intensity defined as firm import of inputs / sales. Firms with at least 50 employees. 16

Table 2: Dynamic Panel Data Estimates of Equation (1). 1997-2002. System GMM. Variable Model 1 Model 2 Model 3 Rel. demand high education Manufacturing sectors R&D / tot wage bill Material offshoring -0.037 (0.016) ** 2.658 (2.569) Material offshoring large changes Service offshoring 0.038 (0.029) Service offshoring large changes 0.029 (0.016) * -2.478 (2.494) -0.063 (0.080) 0.373 (4.397) -5.428 (12.293) R&D/skill labor wage bill 5.083 (5.737) -4.520 (5.580) -0.297 (9.844) 9.942 (27.516) Hansen test (df), p-value 0.000 0.183 0.012 AR(1) test (p-value) 0.000 0.018 0.926 AR(2) test (p-value) 0.377 0.890 0.184 No. of obs. 7507 7507 7485 Material offshoring -0.003 (0.004) Material offshoring large changes 0.0001 (0.017) Service sectors 0.002 (0.028) -0.002 (0.004) 0.355 (0.109) *** 0.005 (0.010) Service offshoring -0.025 (0.014) * -0.206 (0.078) *** 0.002 (0.009) Service offshoring large changes 0.033 (0.049) 0.479-0.037 (0.118) *** (0.020) * Hansen test (df), p-value 0.003 0.000 0.233 AR(1) test (p-value) 0.000 0.000 0.000 AR(2) test (p-value) 0.069 0.000 0.917 No. of obs. 8160 8160 8160 Notes. System GMM for firms that survive at least three years. Standard error within parenthesis ( ). *, **, ***, indicates significance on the 10, 5 and 1 percent level respectively. Offshoring intensity (firm import; inputs / sales. Firms with at least 50 employees. Firm level variables: Firm output, capital stock, Herfindahl index, export ratio and profit margin are included but due to space limitations not shown. Detailed result description is available on request. 17

Table 3. R&D-intensive firms and firms with large changes in offshoring, Manufacturing sector: The impact of offshoring on firms R&D & labor composition 1997-2002. Sys. GMM. Variable Model 1 Model 2 Model 3 Rel. demand high education R&D / tot wage bill Material offshoring -0.051 (0.015) *** 0.621 (2.427) Material offshoring large changes Material offshoring Top 10% R&D Material offshoring Top 10% R&D & large changes 0.055 (0.018) *** -0.896 (2.676) 0.029 (0.019) Serv offshoring 0.039 (0.029) Serv offshoring large changes Serv offshoring Top 10% R&D Serv offshoring Top 10% R&D & large changes -2.724 (2.926) -0.058 (0.033) * -0.275 (4.911) -0.833 (4.334) -0.255 (0.114) ** 11.578 (17.145) -0.148 (0.091) * 8.359 (13.586) 0.346 (0.163) ** -36.239 (24.249) R&D/skill labor wage bill 0.161 (5.591) -1.229 (6.023) -7.426 (7.246) -1.410 (10.956) -1.401 (9.700) 19.548 (38.212) 9.357 (30.267) -66.065 (54.898) Hansen test (df), p-value 0.000 0.003 0.015 AR(1) test (p-value) 0.000 0.033 0.028 AR(2) test (p-value) 0.377 0.750 0.446 No. of obs. 7507 7507 7473 Notes. Standard error within parenthesis ( ). *, **, ***, indicates significance on the 10, 5 and 1 percent level respectively. Offshoring intensity (firm import; inputs / sales. Firms with at least 50 employees. Firm level variables: Firm output, capital stock, Herfindahl index, export ratio and profit margin are included but due to space limitations not shown. Detailed result description available on request. 18

Table 4. R&D-intensive firms and firms with large changes in offshoring, Service sector: The impact of offshoring on firms R&D and labor composition. 1997-2002. Sys. GMM. Variable Model 1 Model 2 Model 3 Rel. demand high education Material offshoring -0.003 (0.004) Material offshoring large changes Material offshoring Top 10% R&D Material offshoring Top 10% R&D & large changes -0.015 (0.023) 0.004 (0.019) 0.023 (0.034) R&D / tot wage bill 0.014 (0.020) -0.024 (0.106) Serv offshoring -0.027 (0.011) ** -0.061 (0.050) Serv offshoring large changes Serv offshoring Top 10% R&D Serv offshoring Top 10% R&D & large changes 0.030 (0.024) 0.014 (0.029) -0.021 (0.046) R&D/skill labor wage bill 0.015 (0.031) -0.029 (0.161) 0.310 (0.082) *** 0.292 (0.125) ** 0.336 (0.160) ** 0.419 (0.242) * 0.032 (0.095) -0.059 (0.076) 0.050 (0.144) 1.059 (0.136) *** 1.461 (0.205) *** 1.639 (0.213) *** 1.126 (0.323) *** Hansen test (df), p-value 0.016 0.000 0.000 AR(1) test (p-value) 0.000 0.000 0.000 AR(2) test (p-value) 0.069 0.000 0.000 No. of obs. 8160 8160 8047 Notes. Standard error within parenthesis ( ). *, **, ***, indicates significance on the 10, 5 and 1 percent level respectively. Offshoring intensity (firm import; inputs / sales. Firms with at least 50 employees. Firm level variables: Firm output, capital stock, Herfindahl index, export ratio and profit margin are included but due to space limitations not shown. Detailed result description available on request. 19

Figure 1. Import of intermediate goods and services in the service sector, (import/sales)1997-2002 5,0% services offshoring materials offshoring narrow materials offshoring wide 4,5% 4,0% 3,5% 3,0% 2,5% 2,0% 1,5% 1,0% 0,5% 0,0% 1997 1998 1999 2000 2001 2002 Figure 2. Imports of intermediate goods and services in the manufacturing sector, (import/sales)1997-2002 10,0% services materials offshoring narrow materials offshoring wide 9,0% 8,0% 7,0% 6,0% 5,0% 4,0% 3,0% 2,0% 1,0% 0,0% 1997 1998 1999 2000 2001 2002 20

Appendix Table A1. The number of offshorers and their percentage shares in Sweden, 2000 Actual number of firms in the sample Share Offshorers (%) Non-offshorers Offshorers Firms Employment Value Added R&D MANUFACTURIN G 204 1587 88.6 95.7 97.2 99.7 SERVICES SECTOR 980 1052 51.7 71.0 71.5 96.7 Notes: Notes numbers and shares reported refer to 2000 nd and relate to both outsourcing of international services intermediates in general and international goods outsourcing according to the narrow definition. The sample is truncated at 50 employees. Table A2. Distribution of offshoring firms by sector 2000 Sector code Sector Non-OFF 15-16 Food and beverages and tobacco 17 115 87% 17-19 Textile; Wearing apparel and Leather 0 42 100% OFF Percentage offshorers 20 Wood and wood products 30 116 79.4% 21-22 Pulp, paper and paper products. Publishing and printing 71 185 72.2% 23-25 Fuel Chemicals and chemical products Rubber and plastic products 1 199 99.5% 26 Other non-metallic mineral products 3 48 94% 27-28 Basic and fabricated metals 34 271 88.8% 29-33 Machinery and equipment (nec); Office machinery and computers; Electrical machinery nec; Radio, TV; Med., precision and optical instruments; 35 473 93.1% 34-35 Motor vehicles, trailers and semi-trailers; Other transport equipment 13 138 91.4% 36-37 Furniture, manufacturing nec and Recycling. 0 0 0 50-52 Wholesale and retail trade; repair of motor vehicle, motorcycles and personal and household goods 260 526 66.9% 55 Hotels and Restaurants 81 41 33.6% 60-63 Transport and storage 181 120 39.9% 64 Post and telecommunications 17 18 51.4% 70-74 Real estate; renting and business activities. 441 347 44% Notes: Authors calculation from SCB data. The sample is truncated at 50 employees. 21

Table A3: Characteristics of offshoring firms relative to non offshorers, year 2000 Variables Advantage of offshorers vs non-offshorers (1) Value Added per employee 0.18*** (0.52) (2) Gross output per employee 0.44*** (0.69) (3) Skill intensity 0.22*** (0.02) (4) R&D intensity 0.01*** (0.09) (5) Employment 0.46*** (0.98) (6) Purchased Materials per employee 0.63*** (1.38) (7) Capital per employee 0.52*** (1.35) (8) Average Wages 0.13*** (0.26) Observations 2203 Notes: figures reported are log deviation from the average non-offshoring firm and represent differences in means for services offshorers and non services offshorers respectively. Standard deviations in brackets. The sample is truncated at 50 employees. Table A4. Offshoring and multinationality, (%) Nonoffshorer offshorer Proportion of domestic firms (%) 79.0 36.1 Proportion of domestic MNEs (%) 12.6 30.4 Proportion of Foreign MNEs (%) 8.4 33.4 Notes: Data for year 2000, truncated at 50 employees. 22

Table A5. Variables Variable Service offshoring Material offshoring wide def. Material offshoring narrow def. Output Profit margin Export ratio Capital stock Definition Import of services Import of production inputs and non-durable consumption goods. Import of production inputs. Total sales Profit / sales Export / sales Book value, buildings and machinery Data source: Statistics Sweden: Financial statistics, Regional labor market statistics, material import statistics and service import statistics. Table A6: Dynamic Panel Data Estimates of Equation (1). 1997-2005. Variable Mod1 Mod2 Mod3 Mod4 Manufacturing sector Rel. demand low & md. education Rel. demand high education Material offshoring 0.035 (0.011) *** -0.035 (0.011) *** -0.695 (1.304) Material offshoring large changes -0.033 0.031 0.437 (0.011) *** (0.011) *** (1.364) R&D / tot wage bill R&D/skill labor wage bill -1.796 (2.915) 1.169 (3.049) Hansen test (df), p-value 0.005 0.129 0.951 0.966 AR(1) test (p-value) 0.000 0.000 0.000 0.000 AR(2) test (p-value) 0.018 0.017 0.587 0.692 No. of obs. 12378 12378 12378 12355 Service sector Material offshoring 0.002 (0.004) Material offshoring large changes -0.002 (0.005) -0.023 (0.014) * 0.022 (0.014) -0.013 (0.021) 0.088 (0.062) -0.017 (0.027) 0.065 (0.080) Hansen test (df), p-value 0.005 0.007 0.000 0.000 AR(1) test (p-value) 0.000 0.000 0.000 0.000 AR(2) test (p-value) 0.308 0.308 0.072 0.253 No. of obs. 14430 14430 14430 14273 Notes. System GMM for firms that survive at least three years. Standard error within parenthesis ( ). *, **, ***, indicates significance on the 10, 5 and 1 percent level respectively. Offshoring intensity (firm import; inputs / sales. Firms with at least 50 employees. Firm level variables: Firm output, capital stock, Herfindahl index, export ratio and profit margin are included but due to space limitations not shown. Detailed result description is available on request. 23