Offshoring and Wages: Evidence from Norway

Offshoring and Wages: Evidence from Norway Ragnhild Balsvik and Sigurd Birkeland September 3, 2012 Preliminary and incomplete Abstract We use matched employer-employee data from Norwegian manufacturing combined with firm level trade data from the customs declarations for the period 1996-2007 in order to investigate the impact of both firm and industry level offshoring on individual level wages. We estimate worker-level Mincer wage equations where we include both industry level and firm level offshoring measures. To address the potential problem of endogenous firm-level offshoring, we use an instrumental variable approach. Our preliminary findings suggest that to the extent that wages are affected by offshoring, it is the wages of employees in offshoring firms that are affected, while the offshoring intensity of other firms in the industry has little impact on the wages of workers in a given firm. Offshoring to low income countries has a positive impact on wages, although there is no significant impact on the wages of high skilled workers. Keywords: Offshoring, outsourcing, wages, matched employer-employee data JEL-codes: F16, J23 Norwegian School of Economics and Business Administration, Helleveien 30, 5045 Bergen, Norway; email: ragnhild.balsvik@nhh.no University of Bergen, Department of Economics, Postboks 7802, 5020 Bergen, Norway; email: sigurd.birkeland@econ.uib.no

1 Introduction An important feature of the ongoing globalization process is the disintegration of the production process into different parts conducted at different locations. As a result, firms are increasingly engaged in offshoring, either by importing intermediate inputs from affiliates or external suppliers abroad, or by relocating parts of their production to affiliates abroad. 1 Along with the documented increase in offshoring, there has also been a marked trend in several countries of increased wage and income inequality. 2 A pertinent question has been to what extent offshoring can explain rising within-country inequality, or more generally, how offshoring affects labour markets in the offshoring countries. In this paper, we document both the development in earnings inequality among Norwegian manufacturing workers and the development in firm-level offshoring activity over the period 1996-2007. We then investigate the impact of firm level offshoring on individual wages. From a theoretical point of view, the potential effects of offshoring on labour market outcomes are not obvious. Relocating parts of production abroad will ceteris paribus reduce employment at home, but the associated cost savings may make the offshoring firms more productive and competitive, increase their market shares and hence increase employment and wages. Theoretical models that distinguish between skilled and unskilled labour could, depending on assumptions, produce both increasing and decreasing wage inequality as a result of offshoring tasks that are low-skill intensive. 3 As the theoretical models generate different predictions concerning the links between offshoring and labour market outcomes, this has been investigated in a number of empirical studies. The majority of the existing empirical studies use industry level data on offshoring, see Crino (2009) for a recent survey of the empirical literature. The empirical literature on the international trade activities of firms has documented large withinindustry heterogeneity in both exporting and importing, this feature has so far received little attention in the empirical literature on offshoring. The within-industry variation in both the level of offshoring activity and in the types of offshoring (what kinds of products that are being imported from where), raises the question of how to interpret the results of empirical studies of offshoring using industry level offshoring measures. The definition of narrow offshoring at the industry level used in the literature usually defines the industry at a relatively aggregate level such that the industry produces at least some of its own intermediate inputs. This means that some of the firms in the industry are likely to be suppliers of other firms classified in the same industry, while other firms will be competitors. This means that changes in offshoring within the industry could have many different effects depending on which firms increase their offshoring. Consider the simple example where a firm starts to import intermediate inputs, 1 Hummels, Ishii, and Yi (2001) and Feenstra and Hanson (2003) provide evidence of this development for several countries. 2 For a survey and debate of the trends in the U.S., see Autor, Katz, and Kearney (2008). Similar trends are found for the UK and other anglophone countries, as documented by Machin and Reenen (2007). Several developing countries have also seen a rise in income inequality, see Goldberg and Pavcnik (2007). 3 One example is the different predictions that could be generated from the models of Feenstra and Hanson (1996b) and Grossman and Rossi-Hansberg (2008). 2

the production of which is unskilled intensive. Whether or not the firm itself previously produced this input, starting to import means that costs are reduced and profits increase. A natural prediction is then that this increases the wages of workers remaining in the firm. If the firm used to produce the offshored input itself, the change in the composition of labour demand within the firm, and the potential threat of further offshoring of low-skilled activities could lead to increased wage dispersion within the firm. If the firm used to source this intermediate input from domestic suppliers, and the change in offshoring merely involves replacing domestic suppliers with foreign suppliers, this main effect on labour demand could perhaps be felt by the supplying firm. Thus, there are clear reasons to believe that the effects of offshoring on wages of employees could depend on whether one looks for wage effects in offshoring or non-offshoring firms. In order to capture this aspect we use information both about industry level and firm level offshoring activity. We calculate both firm- and industry-level offshoring measures following the definition of narrow offshoring coined by Feenstra and Hanson (1996b). We are only aware of a handful of papers that use firm level trade data or firm level offshoring information to study the impact of firm level offshoring on wages (Andersson and Karpaty (2007), Moser, Urban, and di Mauro (2010) and Hummels, Jorgensen, Munch, and Xiang (2011)). We use matched employer-employee data from Norwegian manufacturing combined with firm level trade data from the customs declarations. We estimate worker-level Mincer wage equations where we include both industry level and firm level offshoring measures. Since we are using individual level earnings data, we can plausibly argue that the industry level offshoring measure can be regarded as exogenous relative to individual earnings. This is particularly so since our industry offshoring measure does not include the offshoring activity of the firm employing the worker. A similar argument cannot be used for the firm level offshoring measure. It is likely that unobserved firm level shocks are correlated both with firm level offshoring and wage setting. To address the problem of endogenous firm-level offshoring, we construct instruments that are correlated with offshoring, but arguably exogenous to changes in the productivity and wage structure of the firm. Our preliminary findings suggest that to the extent that wages are affected by offshoring, it is the wages of employees in offshoring firms that are affected, while the offshoring intensity of other firms in the industry has little impact on the wages of workers in a given firm. Offshoring to low income countries has a positive impact on wages, although there is no significant impact on the wages of high skilled workers. The rest of the paper is structured as follows. Section 2 contains a brief overview of the existing empirical literature on offshoring and labour market outcomes. Section 3 presents the data we use in our analysis, provides a discussion of the measurement of offshoring and shows the development in earnings inequality and offshoring in Norwegian manufacturing. In section 4 we present our empirical methodology, our results are found in section 5, while section 6 reports robustness checks and 7 briefly concludes. 3

2 Related empirical literature From a theoretical point of view, the potential effects of offshoring on labour market outcomes are not obvious. 4 Relocating parts of production abroad will ceteris paribus reduce employment at home, but the associated cost savings may make the offshoring firms more productive and competitive, increase their market shares and hence increase employment and wages. 5 Theoretical models that distinguish between skilled and unskilled labour could, depending on assumptions, produce both increasing and decreasing wage inequality as a result of offshoring tasks that are low-skill intensive. Feenstra and Hanson (1996b) set up a model where a final good is produced by a continuum of tasks that each vary in its required skill intensity. To perform all tasks, both capital and labour are required. Firms in skill intensive (rich) countries will offshore the least skill intensive tasks to (poor) countries that are abundant in low skilled labour. If offshoring from the rich country becomes more feasible (modelled as an increase in the capital stock of the poor country), the skill threshold of offshored tasks will increase. The result is decreased relative demand for unskilled labour and an increase in wage inequality in the offshoring country. 6 Grossman and Rossi-Hansberg (2008) introduce costs of offshoring tasks to a foreign country. In their model, a reduction in the cost of offshoring tasks that are performed by low skilled workers acts as an increase in the productivity of low skilled labour. Such a cost reduction will benefit the low-skill intensive sector more than the high-skill intensive sector, resulting in an increase in the economy-wide demand for low skilled labour. 7 The models of trade in tasks by Feenstra and Hanson (1996b) and Grossman and Rossi-Hansberg (2008) are based on assumptions of competitive labour and product markets. Examples of alternative models of outsourcing with imperfect markets and firm level bargaining between firms and unions can be found in Kramarz (2008) and Lommerud, Meland, and Straume (2009). In Kramarz (2008), firms and unions bargain over both employment and wages in a two stage game where the firms decide on the level of offshoring in the first stage. In this model, offshoring reduces the rents the firm has to share with the unions in the bargaining process, and hence the model predicts that firms who face strong unions should outsource more than firms facing weak unions in order to discipline workers. The result is that increased outsourcing reduces employment in the firm while the effect on wages is ambiguous. Kramarz (2008) tests his model using detailed firm and worker level data from France, and finds evidence consistent with its predictions. Interestingly, using a similar model Lommerud, Meland, and Straume (2009) conclude that lower union power leads to increased outsourcing, and that outsourcing has a positive effect on the 4 For a recent review of the literature on the link between trade and inequality, see? 5 Evidence consistent with a positive effect of offshoring on productivity is found by Görg and Hanley (2005) and Görg, Hanley, and Strobl (2008) for Irish firms, Andersson, Karpaty, and Kneller (2008) for Swedish firms and Hijzen, Inui, and Todo (2010) for Japanese firms. 6 The main point of the model is to show that increased offshoring also increases wage inequality in the poor country, as the newly offshored tasks are more skill intensive than previous tasks, and therefore increase the relative demand for skills in the poor country. This result is in contrast to the standard Heckscher-Ohlin model that predicts that trade in goods would increase income inequality in rich countries and reduce inequality in poor countries. 7 If the country is large, the positive productivity effect on low skilled wages could be overturned by a lower relative price of the final product. 4

wages for those workers remaining in-house. 8 The empirical literature looking at the labour market effects of offshoring originated from the twin observation of increases in the skilled/unskilled wage-gap and the trend of increasing fragmentation of production across borders. 9 The early empirical papers often start from the empirical specification suggested by Berman, Bound, and Griliches (1994), in which the representative firm is assumed to have a translog cost function with fixed capital and variable labour inputs of different skills. From such a specification one can arrive at an estimating equation where the wage-bill share of each skill group is a function of capital, output and wages for the different skill groups, and a measure of offshoring intensity. Examples of papers using variants of this estimation approach are Feenstra and Hanson (1996b), Falk and Koebel (2002), Strauss-Kahn (2004), Hijzen, Görg, and Hine (2005) and Ekholm and Hakkala (2008). 10 The mentioned studies all use industry level data both for the wage-bill (or employment) shares of different skill groups and for offshoring. The offshoring measures used in these studies follow the definition of narrow offshoring used by Feenstra and Hanson (1996b), where the industry level narrow offshoring measures are based either on aggregate trade data or input-output tables (or a combination of the two data sources) and calculates the share of imports from the same industry relative to domestic production of that industry. The main idea behind these studies is that offshoring could induce a within-industry shift in relative labour demand towards high skilled labour if the fragmentation of production relocates unskilled-intensive stages of production to countries relatively abundant in unskilled labour. In general, the studies mentioned above all find that relative labour demand within industries tilt in favour of high skilled workers as a result of increased industry-level offshoring. Feenstra and Hanson (1996b) conclude that increases in their narrow offshoring measure can explain between 30 to 50% of the increase in the wage share of nonproduction workers in US manufacturing during the period from 1972-1990. For France, Strauss-Kahn (2004) finds that offshoring explains from 11 to 25% of the decline in the share of unskilled workers in French manufacturing employment for the 1977-1993 period. A similar direction of results are also found by Falk and Koebel (2002) for Germany, Hijzen, Görg, and Hine (2005) for the UK and Ekholm and Hakkala (2008) for Sweden. 11 Offshoring may affect labour markets, not only through an effect on the wage gap between skilled and 8 The main reason for the diverging predictions of Kramarz (2008) and Lommerud, Meland, and Straume (2009) lies in whether the costs of imports are included in the profits at the stage of bargaining between the firm and its workers. In Kramarz model outsourcing has the effect of reducing the rents that the firm shares with its workers, while in Lommerud, Meland, and Straume (2009) outsourcing increases the rents to be shared. As stronger unions are able to get a larger share of rents, an increase in outsourcing is not profitable for firms with strong unions. 9 Note that the literature on the wage effects of offshoring is primarily interested in within-industry developments in the wage gap. There is also a literature documenting the relationship between import competition and wage differences between industries. For examples; see Lundin and Yun (2009) and Du Caju, Rycx, and Tojerow (2011). 10 See the review by Crino (2009) for further examples of this strand of the literature. 11 These studies focus on offshoring of material inputs, as does the bulk of the empirical literature on offshoring. Data on imports of services have so far not been as available as data on imports of goods, thus there are relatively few studies of labour market effects of services offshoring. The studies that exist conclude that services offshoring typically do not seem to affect overall employment levels (Amiti and Wei (2005), Amiti and Wei (2006)), while there is some evidence that services offshoring tilts relative labour demand among white collar workers towards the high skilled (Crino (2009)). As our data do not allow an analysis of services offshoring, we focus on material offshoring. 5

unskilled workers, but also through an impact on labour demand elasticities. The substitution of foreign labour for domestic through offshoring could flatten the domestic labour demand curve, as first noted by Rodrik (1997). Importantly, the threat of offshoring could increase labor demand elasticities, even if actual levels of offshoring do not change. Higher elasticities would make wages and employment more volatile in response to productivity shocks, and could reduce the bargaining power of labour. Slaughter (2001), using industry level data, investigates the role of offshoring in explaining the increased demand elasticity for production workers in US manufacturing industries over the period 1961-1991, and finds no significant effect of offshoring. Senses (2010) uses plant level data on labour demand in the US for the period 1972-2001, and finds that industry level demand elasticities for production workers are positively associated with industry level measures of exposure to offshoring. Hijzen and Swaim (2010) use industry level data for 11 OECD countries for the period 1980-2002, finding that offshoring increases the elasticity of labour demand to some extent, the effect being stronger in countries without extensive legislation on employment protection. Most empirical papers in the offshoring and labour markets literature have used industry level data for wages, employment and offshoring measures. Recently, more detailed data sets have been used for the analysis of these questions. Geishecker and Görg (2008) use individual data from the German socioeconomic household panel to estimate individual Mincer wage equations with industry-level offshoring measures as the variables of main interest. Their approach estimates the short-run effects of offshoring, and they find that offshoring reduced the real wage for workers in the lowest skill categories while it increased real wages for high-skilled workers over their sample period from 1991 to 2000. Similar results are found for Denmark by Munch and Skaksen (2009). A few papers use data on individual workers and their job transitions in combination with industry level measures of offshoring. Egger, Pfaffermayr, and Weber (2007) use a sample of Austrian male workers over the period 1988 to 2001, finding that an increase in offshoring intensity negatively affects the probability of staying in or changing into the manufacturing sector. Munch (2010) use data from Denmark for the period 1990-2003 and estimates the probabilities of job-to-job and job-to-unemployment transitions. He finds that offshoring increases the unemployment risk for workers, in particular relatively low skilled workers, but the quantitative effect is small. The estimates are interpreted as evidence of small short-run adjustment costs to increases in offshoring. To the best of our knowledge, only a handful of empirical papers use firm level information about imports to construct firm level measures of offshoring intensities. Andersson and Karpaty (2007) follow the approach of Berman, Bound, and Griliches (1994) and estimate firm level relative labour demand for different skill groups in Swedish manufacturing, using firm level offshoring as their main control variable. Their approach identifies the within-firm effects of increased offshoring to different regions of the world. They find that the relative demand for high skilled labour is positively affected by service 6

offshoring and offshoring of goods to Asia, but negatively affected by offshoring to high income countries. Their results indicate very small elasticities, much smaller than found by Ekholm and Hakkala (2008) who use the same methodology with industry level data. One possible reason for this difference is that estimates using industry level offshoring measures capture the effect on overall relative labour demand in the industry, while when using firm level offshoring information the effects are identified by the firms that increase offshoring. If the main reason for observing increased offshoring at the industry level is that some firms replace domestic suppliers (in the same industry) with imports from abroad, the largest effect will likely be felt by workers in the domestic suppliers, and not in the firms that increase their offshoring. The industry level studies will pick up this latter effect, while the firm level studies will not. Moser, Urban, and di Mauro (2010) examine the effects of offshoring on net employment using German establishment level data for the years 1998-2004. In contrast to Andersson and Karpaty (2007), they do not have access to firm level trade data. Instead, they construct a binary offshoring variable on the basis of a survey question in the data. The question asks whether firms have not at all, partly or predominantly received intermediate inputs from abroad, and the answers are used to generate their offshoring variable that takes the value one if the firm states that they have increased their share of intermediate inputs from abroad from one survey to the next. The authors argue that firm level changes in offshoring are endogenous to employment, and they use difference-in-differences matching techniques to control for the selection into treatment (offshoring). They find that the overall effect of offshoring on employment is positive, which suggests that any negative direct effects from relocation of production (downsizing) are compensated by positive indirect productivity effects at the firm level. They also argue that the treated firms are primarily increasing their offshoring by replacing domestic suppliers with foreign ones, as they find no increase in total intermediate inputs in total turnover. This also indicates that the direct downsizing effect in treated plants is small, explaining the dominance of the positive productivity effect on employment. 12 Hummels, Jorgensen, Munch, and Xiang (2011) use Danish matched employer employee data with firm level import and export information to investigate the effect of shocks to exports and imports on wages and employment at the firm and worker level. They start from a production function framework with capital and a composite input consisting of labour and imported inputs. An increase in imported inputs reduces the marginal product of labour as long as there is some substitutability between imports and labour. This effect will decrease both employment and wages. As the increase in imports also increases the marginal product of capital, the firm would want to adjust all its inputs. Thus, the authors distinguish between the wage elasticity depending on whether capital is constant or allowed to adjust to take account of the productivity effect. The model predicts that labour types that are highly substitutable with imports will experience decreased demand and hence falling wages as a response to increased imports. Based on 12 Böckerman and Riihimäki (2009) use Finnish firm level data on employment and offshoring and conclude that intensive offshoring do not decrease overall firm level employment nor have any effect on the employment of low-skilled workers. 7

this framework Hummels, Jorgensen, Munch, and Xiang (2011) estimate individual wage equations using spell-fixed effects where the impact on wages from changes in imports is identified by time variation in imports within firms. As the choice of imports and exports is likely to be correlated with wage setting and employment, they construct instruments for firm level imports and exports. They find that an exogenous increase in imports is associated with lower employment, primarily because of lower employment of lowskilled workers. They also find that high skilled workers get higher wages while low skilled workers receive lower wages as a result of increased imports. 13 3 Data 3.1 Data sources We use three different annual data bases for the years 1996-2007, all of which are censuses that can be linked to each other by firm or plant identifiers. All the data sources are administered by Statistics Norway. Our starting point is the Norwegian Manufacturing Statistics, which is collected at the plant level. We then link the administrative files containing the whole population of residents aged 16-74 to the plant level data. The administrative files contain, among other things, information on age, gender, identification of current employer, weekly work-hours, annual earnings and detailed education codes. Weekly work-hours are recorded as a categorical variable in four groups, with the longest work-hours being 30 hours or more per week. Finally, we use the Norwegian customs data that record all customs declarations above 1000 NOK (corresponding to approximately 200 US dollars). The customs declarations give information about imports and exports at the firm level by country and product code. We have firm identifiers in the manufacturing statistics that enable us to link the imports and exports of a firm to each of its plants, but unfortunately we cannot identify the imports and exports of single plants of multi-plant firms. We therefore aggregate the manufacturing statistics to the firm level. We restrict our sample in the following way. After matching workers to the plants in the Manufacturing statistics, we drop workers who never work full-time and workers earning less than the 10th percentile 80% or more of their years in a manufacturing plant. We then aggregate to the firm level and keep only firms that employ at least two full-time workers each year. In order to reduce the possible mistakes from discrepancies between the employer identification numbers in the manufacturing statistics and individual data, we drop firms with large differences between the number of matched workers and the number of employees as recorded in the manufacturing statistics (a difference of more than 400 employees). We also drop firms where the change in number of employees from one year to the next is very different depending on whether we calculate this according to the number of matched individuals or according to 13 Kramarz (2008) investigates how wages and employment of French manufacturing workers are affected by the sourcing strategies of their employing firms. The analysis focuses on how the strength of labour unions may affect the choice of offshoring. Using a theoretical model he predicts that firms facing strong unions will outsource more, and as a result lower their employment. Kramarz finds evidence consistent with this prediction in the French data. 8

the information about the number of employees in the manufacturing statistics (a difference of more than 300). Our data after these procedures consists of a firm-level panel with 11 900 different firms giving 81 000 firm-year observations. These firms employ in total over the period 425 000 different workers giving rise to 2,4 million worker-year observations. The firms in the panel account for around 80% of total manufacturing production and employment over the sample period. 14 Table 1 shows for each year the total number of firms and workers in our base sample as well as the minimun, average and maximum number of matched workers to firms. Further restrictions on the sample for regressions are detailed in section 5. Table 1: Firms, workers and firm size Firm size Firms Workers Min Mean Max 1996 6 469 209 710 2 32 3878 1997 6 626 221 849 2 33 3897 1998 6 910 230 129 2 33 3825 1999 6 732 228 891 2 34 3636 2000 6 914 223 615 2 32 3506 2001 6 814 222 197 2 33 3345 2002 6 828 215 394 2 32 3326 2003 6 780 206 476 2 30 3161 2004 6 783 203 336 2 30 2823 2005 6 809 204 827 2 30 2759 2006 6 817 210 781 2 31 4091 2007 6 604 210 042 2 32 4391 3.2 The wage gap in Norwegian manufacturing As mentioned in the introduction, one of the main issues in the empirical literature on offshoring has been to what extent offshoring is related to the increasing wage gap between low-skilled and high-skilled workers observed in many countries during the last two decades. The Norwegian labour market is, compared to most other countries, characterized by a compressed wage structure (OECD 2008). In manufacturing there is centralized bargaining with additional firm level wage hikes on top of the centrally negotiated agreements. Traditionally, centralized bargaining has ensured a relatively compressed wage structure. Evidence on the development of wage inequality in Norway during our sample period is provided in detail by Dale-Olsen and Nilsen (2009). They conclude that the spread in earnings in Norway has clearly increased from 1995 to 2006, as measured by the ratio of the 95th to the median of the earnings distribution. This holds whether one measures earnings as hourly wages or annual earnings, but the increase in inequality is moderate compared to the development in the US and the UK. In our data we have information about annual labour earnings, and figure 1 shows the development in earnings ratios 14 The cleaning procedures do not seem to hit disproportionally in certain years or certain industries. 9

among manufacturing workers. The development in manufacturing in figure 1 shows an increase in earnings dispersion, particularly at the top of the distribution. 15 Figure 1: Earnings inequality: percentile ratios 1.4 1.6 1.8 2 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 95 50 ratio 90 50 ratio 50 10 ratio Source: Own calculations based on register data from Statistics Norway Figure 2 shows the development in the average earnings of workers with 13 or more years of education relative to the average earnings of workers with less than 13 years of education over the sample period. This earnings ratio has been around 1.5 for the whole period from 1996 to 2007. Over the same period there has been an increase in the corresponding employment ratio of these two groups of workers, increasing from 0.14 to 0.22. Figure 2: Skilled to unskilled wage- and employment-ratios 0.5 1 1.5 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Earnings ratio Employment ratio Source: Own calculations based on register data from Statistics Norway Skilled workers defined as workers with 13 or more years of schooling 15 This is in line with the findings of Dale-Olsen and Nilsen (2009) for the overall Norwegian earnings distribution. 10

3.3 Offshoring In the theoretical literature on offshoring, as well as in the general discussion of international outsourcing and production fragmentation, the concept of offshoring captures the movement of parts of the production process from within the firm at home to some foreign entity. Thus, as a result of offshoring, firms should be expected to increase their trade in intermediate goods as in the framework of Feenstra and Hanson (1996a) or trade more in tasks as in the framework of Grossman and Rossi-Hansberg (2008). 16 The concept of offshoring includes both trade with foreign affiliates and arm s length transactions, the latter often called international outsourcing. Implicitly, the underlying idea is that some parts of the production process remain onshore, and in particular that the production of the final product sold to consumers remains onshore. As different parts of the production process may differ in the skill intensity of labour used for the process, offshoring parts of the production process could lead to a change in the composition of labour demand as well as a change in overall labour demand. When a firm moves part of the production process abroad, this will show up in the international trade statistics as imports of goods or services to the firm. Hence, firms imports serve as a natural proxy for offshoring, while moving the assembly of the final good to a foreign country will not be regarded as offshoring since this does not generate imports of intermediate inputs to the firm. Until recently, the main source of information regarding trade flows was only available at the industry level. Following the seminal papers by Feenstra and Hanson (1996a, 1999) most of the empirical literature on offshoring has used imports of intermediates as a proxy for offshoring. Feenstra and Hanson (1999) argue that imports from the same narrowly defined industry best capture the idea of offshoring as such imports best represent products that could have been produced by the industry (they use auto parts to the automobile industry as an example). This narrow measure of offshoring is usually obtained from the diagonal elements of the import sheets of the input-output tables constructed by national statistical agencies. In addition the term broad offshoring has been used of total imports, or as the differences between total imports and narrow imports. 17 Although imports of intermediates serve as the most common proxy for offshoring, these imports are usually normalized by some measure of the size of the industry. The reason is that changes in levels of import do not identify changes in offshoring activity per se. It is implicit in the definition that there is a movement of production taking place, and as such increases in imports coinciding with increases in production should be treated as mere up-scaling, something which should not give rise to changes in the composition of labor demand. While the original Feenstra-Hanson measure weights the imports with input use, some have later argued that the more appropriate normalization would be to weight imports 16 A distinction that according to Grossman and Rossi-Hansberg (2008), is largely semantic. 17 Several studies have use the distinction between narrow and broad offshoring,see e.g. Feenstra and Hanson (1999), Slaughter (2001), Kramarz (2008), Ekholm and Hakkala (2008), Geishecker and Görg (2008) and Munch (2010). We concentrate on narrow offshoring in the current paper. 11

by production. 18 Horgos (2009) discusses different offshoring measures, and argues that the measures weighting imports of intermediates by either input usage or production are very capable of capturing outsourcing activities. He also notes that one reason for the different results found in the literature may be due to the different measures used, and that the weights used to normalize the imports may influence the findings in such a way that comparisons across normalizations cannot be made. With access to firm level trade data or firm level offshoring information it is possible to construct firm level offshoring measures as for example in Andersson and Karpaty (2007) and Moser, Urban, and di Mauro (2010). The former study uses customs data for Sweden in order to construct firm level offshoring intensities, while the latter study uses German survey data to construct an indicator variable for whether the firm has increased its offshoring activity or not. 19 Hummels, Jorgensen, Munch, and Xiang (2011) also uses firm level import data in their analysis of outsourcing and wages, but they do not normalize imports to production or input use at the firm level. 20 Studies using firm level information on imports find significant within industry variation in both the level of offshoring activity and in the types of offshoring (what kinds of products that are being imported). Heterogeneity between firms of the same industry in terms of their offshoring activity raises the question of how to interpret the results of empirical studies of offshoring using industry level offshoring measures. As argued by Castellani, de Benedictis, and Horgos (2011), we could observe an increase in industry level offshoring intensity when firms within the industry replace previously domestically sourced inputs with imported inputs even though the use of intermediate inputs in each firm has not changed. They further argue that industry level offshoring indices could also increase due to entry and expansion of foreign multinationals that import from their affiliates abroad, without any previous domestic production being offshored. The empirical studies on the effects of offshoring on wages (e.g. Geishecker and Görg (2008)) or relative labour demand (e.g. Ekholm and Hakkala (2008)) using industry level measures of offshoring captures a net effect at the industry level of the changes in aggregate industry level offshoring activity. This is in contrast to the few studies that use firm level information about offshoring, e.g. Andersson and Karpaty (2007), Moser, Urban, and di Mauro (2010) and Hummels, Jorgensen, Munch, and Xiang (2011). These studies are only able to identify effects of offshoring on the firms that change their offshoring activity. The same caveat as Castellani, de Benedictis, and Horgos (2011) discuss for the industry level offshoring measure also applies to a firm level offshoring measure. When firm level offshoring is defined as imports of goods from the same industry relative to the production of the firm, this measure can increase if the firm replaces previous domestic suppliers with foreign suppliers without 18 See e.g. Geishecker, Görg, and Munch (2010) and Strauss-Kahn (2004) 19 There is an emerging literature focusing in general on the link between firm imports and wages or productivity, see for instance Martins and Opromolla (2009) and Haller. This literature springs out of the literature showing that exporting firms are more productive that non exporting firms (see Wagner for a survey), showing that this often applies even more so to firms that import. 20 Rather Hummels, Jorgensen, Munch, and Xiang (2011) use a production function framework where they also control for other firm level inputs. 12

moving any of the firm s own production processes abroad. The definition of narrow offshoring at the industry level used in the literature usually defines the industry at a relatively aggregate level such that the industry produces at least some of its own intermediate inputs. This means that some of the firms in the industry are likely to be suppliers of other firms classified in the same industry, while other firms will be competitors. This means that changes in offshoring within the industry could have many different effects depending on which firms increase their offshoring. Consider the simple example where a firm starts to import intermediate inputs, the production of which is unskilled intensive. Whether or not the firm itself previously produced this input, starting to import means that costs are reduced and profits increase. A natural prediction is then that this increases the wages of workers remaining in the firm (Lommerud, Meland, and Straume (2009)). If the firm used to produce the offshored input itself, the change in the composition of labour demand within the firm, and the potential threat of further offshoring of low-skilled activities could lead to increased wage dispersion within the firm. If the firm used to source this intermediate input from domestic suppliers, and the change in offshoring merely involves replacing domestic suppliers with foreign suppliers, this should not affect the skill composition of labour demand in the offshoring firm. The effect of this type of offshoring is felt primarily by the previous domestic suppliers who see the demand for their products decrease. For these firms we would expect to see a ceteris paribus change in the skill composition of labour demand and a negative effect on wages, in particular for the low-skilled. 21 Compared to competitors that do not offshore, firms that cut costs by offshoring gain, but as long as the market for their product remains the same we should mainly se reallocation of labour between firms. With the firm level data that is typically available it is hard to observe to what extent a firm moves a production process out of the firm and out of the country. In addition it is not observable which firms that are suppliers of other firms within the same industry. Thus, information about imports is typically the best observable proxy for offshoring that is available. Based on the discussion above there are clear reasons to believe that the effects of offshoring on wages is likely to depend on whether you work in a firm that offshores itself or not. In order to capture this aspect we use information both about industry level and firm level offshoring activity. Using the import data from the Norwegian customs declarations we can calculate both firm- and industry-level offshoring measures following the definition of narrow offshoring coined by Feenstra and Hanson (1996b). According to this definition, used widely in the offshoring literature, narrow offshoring is defined as imports of goods from industry j abroad relative to domestic production of industry j: O nar jt = imp jt prod jt. (1) 21 Of course, if the inputs that are offshored previously were produced by domestic firms in a different industry, this would not be picked up by a narrow offshoring measure. 13

Most of the empirical work on offshoring rely on input-output tables to calculate the offshoring measures at industry level, e.g. Feenstra and Hanson (1996b), Geishecker and Görg (2008). In this paper we use the customs data to calculate the nominator of equation 1 and the production values in the manufacturing statistics to calculate the denominator of equation 1. 22 Given our firm level trade data, we are also able to calculate the narrow offshoring measure for each firm according to O nar ijt = imp ijt prod ijt, (2) where i indexes the firm. Note that in the customs data we are not able to distinguish between imports from affiliated and unaffiliated parties, thus, in line with the existing literature, both imports from foreign subsidiaries and arm s length transactions are included in our offshoring measure. Figure 3: Narrow offshoring: OECD vs non-oecd 1996.02.04.06.08.1 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Narrow offshoring to OECD Narrow offshoring to non OECD Source: Own calculations based on Customs data and Manufacturing statistics Since our data contains information about the country of origin for imports, we show in figure 3 the development of narrow offshoring to OECD and non-oecd countries, respectively. The figure shows that the level of offshoring to countries outside the OECD is much lower than offshoring to OECD countries, but the increase from 1996 to 2007 has been much larger for offshoring to countries outside the OECD. In line with the trend in many other other countries, narrow offshoring has increased also in Norway, from around 11% in 1996 to 14% in 2007. The level of offshoring seems a bit higher than what Geishecker, Görg, and Munch (2007) report for Denmark, Germany and the UK. 23 Moving to the offshoring activities of individual firms, there is large between-firm variation in offshoring intensities. Figure 4 shows firm level measures of narrow offshoring calculated according to 22 We use a correspondence from Statistics Norway which gives the product codes in the customs data a Nace code. We use both the 2-digit and 4-digit Nace level. In the manufacturing statistics we have Nace codes at the plant level. For multi-plant firms we aggregate to the firm-level based on the Nace code with largest production value. 23 Note that the offshoring measures in Geishecker, Görg, and Munch (2007) are based on data from input-output tables. 14

equation 2. In terms of narrow offshoring, the median firm in our sample has no narrow offshoring during our sample period. For firms in the higher percentiles of the annual distribution of offshoring intensities, figure 4 shows both a clear increase in offshoring intensities, and large between-firm heterogeneity in offshoring. 24 Figure 4: Firm level narrow offshoring: by percentile of firm-year distribution 0.1.2.3 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 75th percentile firm 90th percentile firm 95th percentile firm Source: Own calculations based on Customs data and Manufacturing statistics 4 Empirical Strategy With our individual level data we estimate Mincerian wage equations (Mincer, 1974). Our approach is similar to that of Geishecker and Görg (2008), the main difference being that in addition to industry level offshoring measures, we also include firm level offshoring measures in our wage equations. Thus, we estimate variants of the following benchmark log wage equation: y ijnt = α + β x X ijt + β z Z jt + θ 1 O nt + θ 2 O jt + µ ij + µ t + µ n + µ nt + ɛ ijt, (3) where the dependent variable y is the log of annual earnings for individual i in firm j, industry n, year t. X contains a vector of individual level control variables containing potential experience, gender, years of education or skill dummies. Variables associated with the firm or the job-spell are collected in the vector Z: firm size measured by the log of employment, a dummy for exporting status and tenure in the firm. Our main variables of interest are the offshoring intensities O nt and O jt as defined by equations 1 and?? respectively. To account for time invariant industry level wage determinants, or aggregate business cycles affecting all industries similarly, we use a full set of industry and time dummies, µ n and µ t. In 24 Similar figures for the 3 largest 2-digit industries show a similar pattern of heterogeneity, thus the variation in figure 4 is not driven by differences in offshoring activity between industries. 15

addition, to take account of industry specific time trends, for instance related to technological change, we also include industry-year dummies µ nt.further, we estimate equation 3 using worker-firm, or spell fixed effects, thus accounting for both time invariant firm and worker characteristics that might affect earnings. Since our dependent variable of interest is defined at the worker-year level, while our offshoring variables are defined at the industry-year or firm-year level, this could lead to contemporaneous correlation in the error terms that could lead to distorted standard errors, as discussed in?. We therefore cluster standard errors at the industry-year level. In order to make sure that the firm and industry measures of offshoring are not strongly collinear in small industries where potentially very few firms could account for most of the offshoring, we subtract the firms own offshoring from the industry level offshoring measure. Thus, the industry level offshoring measure contains the offshoring intensity for all the other firms in the industry, not including the firm that the worker is employed in. We will in the following refer to this as industry level offshoring, even though it is strictly speaking the offshoring intensity for the remaining firms in the industry. Since we are using individual level earnings data, we can plausibly argue that the industry level offshoring measure can be regarded as exogenous relative to individual earnings (Geishecker and Görg (2008)). This is particularly so since our industry offshoring measure does not include the offshoring activity of the firm employing the worker. A similar argument cannot be used for the firm level offshoring measure. It is likely that unobserved firm level shocks are correlated both with firm level offshoring and wage setting, which means we cannot assume that the error term in equation 3, ɛ ijt, is i.i.d. We therefore need to consider potential solutions to this simultaneity problem. To address the problem of endogenous firm-level offshoring, we follow Hummels, Jorgensen, Munch, and Xiang (2011) and construct instruments that are correlated with offshoring, but exogenous to changes in the productivity and wage structure of the firm. We use world export supply as an instrument for firm level offshoring. The idea is that changes in the world export supply of a product from country j reflects a trade shock to the Norwegian firms importing this product from country j. This trade shock could reflect changes in the costs or desirability of offshoring that is exogenous to the firm. To construct the instrument we use data from the COMTRADE database at the 3 digit SITC product level, finding total world exports of each 3 digit product category from each country-year cell. In order to capture the relevant world export supply variation for a given firm we calculate a scalar capturing the world supply of the country-product combinations that the firm imports. In order to construct a firm level instrument for firm f s overall narrow offshoring in year t we multiply the world export supply of each country-product combination in year t with the country-product offshoring intensity of the firm in year t-1. Aggregating over all country-product combinations (c,p) we get our firm-level and time-varying instrument: W ES f,t = c,p imp f,t 1,c,p prod f,t 1 worldexport c,p. (4) 16

5 Results The model to be estimated is a basic Mincerian wage equation, outlined in section 4 and specified in equation (3). Equation (5) is a reiteration thereof, though with more precise variable definitions and functional form assumptions: log y ijnt =β x1 Experience it + β x2 (Experience it ) 2 + β x3 T enure ijt + β x4 (T enure ijt ) 2 + β x5 MedSkill it + β x6 HighSkill it + β x7 P arttime <30h,ijt + β x8 P arttime 30<p<35h,ijt + β z1 Medskillshare jt + β z2 Highskillshare jt + β z3 log Exports jt + β z4 (log Exports jt ) 2 + β z5 log P roduction jt + β z6 (log P roduction jt ) 2 + θ 1 Offshoring jt + θ 2 Offshoring nt + α + µ t + µ nt + µ ij + ɛ ijt. (5) We use log yearly earnings as the dependent variable. The subscripts i, j, n, and t refer to individuals, firms, industries, and time respectively. Individual specific, (possibly) time variant explanatory variables are denoted with subscripts it indicating the lack of relation to the firm j and industry n in which the individual is employed. Of interest here are mainly potential experience and its square, skill classification (dummies for medium and high skill) and part time dummies. In addition to this we include tenure and its square as a spell specific variable measuring the time the individual has spent in the particular firm. Firm level variables are denoted by subscripts jt. We use skill shares (share of employees of either medium or high skill), exports and its square, and production and its square to control for time varying factors at the firm level that possibly can influence wage setting and potentially can be related to the offshoring decision. The estimation equation also includes various fixed effects captured in the µ-s; time fixed effects, industry and time interactions (fully flexible time trends for each industry, also including industry fixed effects), and finally a spell fixed effect. ɛ is an error term that we allow to be correlated within industry-year clusters, but otherwise assumed to fulfill the usual iid assumptions. 25 In table 2 we present four variations of equation (5) by comparing results with and without inclusion of spell fixed effects and by employing two definitions of the offshoring measure; based on imports from within the same 2 digit and 4 digit industry classifications respectively. Columns 1 and 3 are based on the broader offshoring measure (2 digit) and columns 2 and 4 are based on the narrower measure. Columns 1 and 2 are without spell fixed effects, thus not controlling for spell heterogeneity (or individual and firm specific time invariant heterogeneity), whereas columns 3 and 4 do. 25 In the last part of the analysis we use an IV-approach utilizing lagged trade patterns to identify exogenous changes in offshoring intensities. In order to make the results comparable we restrict the sample used in the first part (where we treat offshoring as an exogenous variable) to be the same as the IV sample. 17