The influence of offshoring on wage inequality in Western Europe

Transcription

1 MASTER THESIS The influence of offshoring on wage inequality in Western Europe An emprical investigation Author: R.C.M. ZAGERS, BSc ANR: Supervisor: dr. G.C.L. VANNOORENBERGHE Number of words: 14,108 August 1, 2011 Tilburg University School of Economics and Management Department of Economics

2 Contents 1 Introduction Research questions and hypothesis Auxilliary research questions Main research question Hypothesis Outline of the thesis Theories about offshoring Offshoring theories Earlier empirical work and trends in offshoring and wages Model setup and data Model setup Basics of the model Endogeneity Regression equations Dataset Data Regression results OLS regressions Single destination regressions Separate destination regressions Explanation of the regression results Conclusion 36 A Regression equations 42 A.1 Single destination regression equations A.2 Separate destination regression equations B Regression tables 44 1

3 List of Tables 3.1 Variables used in regressions: narrow offshoring definition Variables used in regressions: wide offshoring definition Instrumental variables used in 2SLS regressions Industries featuring in regressions Descriptive statistics by industry, narrow definition offshoring Descriptive statistics by industry, wide definition offshoring Correlations of descriptive statistics by industry, narrow definition Correlations of descriptive statistics by industry, wide definition 35 B.1 OLS regression: single destination, narrow definition, country and year dummies B.2 OLS regression: single destination, wide definition, country and year dummies B.3 OLS regression: single destination, narrow definition, all dummies 47 B.4 OLS regression: single destination, wide definition, all dummies 47 B.5 OLS regression: separate destinations, narrow definition, country and year dummies B.6 OLS regression: separate destinations, wide definition, country and year dummies B.7 OLS regression: separate destinations, narrow definition, all dummies B.8 OLS regression: separate destinations, wide definition, all dummies

4 Chapter 1 Introduction Over the past twenty years globalization has gone on to take massive proportions. The mobility of firms has increased a lot as barriers to the movement of capital have been gradually lifted. This has caused many companies, especially multinational ones, to shift (parts of) their production from countries where labour is relatively expensive (mainly the U.S. and Western European countries) to countries where labour is much cheaper (such as Eastern European countries, India and China). Trade unions in Western Europe fear offshoring will cause workers to lose out and face falling wages and increasing wage inequality between low-skilled and high-skilled workers. The phenomenon of offshoring has attracted a lot of attention from researchers over the past twenty years, with the brunt of attention going to investigating what offshoring meant for wages and employment in the U.S. and Mexico after the implementation of the North American Free Trade Agreement (NAFTA) in 1994 (Feenstra & Hanson, 2009)[5]. Research on the effects of offshoring on wages and employment in Europe is rare though. Offshoring is also exerting an ever stronger influence on the position of labour in European manufacturing sectors as the accession of Eastern European countries from 2004 onwards have made offshoring of production to these countries more attractive. Brazil, Russia, India and China (the so-called BRIC countries) have flourished because of offshoring from developed nations amongst other factors. Offshoring of European production to the BRIC nations has become more common too and may add to the effects exerted by offshoring to Eastern Europe. One of the few papers written on offshoring in Europe deals with offshoring from Germany to Eastern Europe (Geishecker 2006). This paper suggests that offshoring has led to skill upgrading in manufacturing, with low-skill jobs being replaced by high-skill jobs. The effects on the wages of the lowly skilled were small. Wage levels have not fallen much, but this does not reveal how wage inequality between low-skilled and high-skilled workers has developed. This has led me to conclude that research into how offshoring affects wage inequality in Western Europe may be worthwile. This thesis aims to answer the question how wage inequality between lowskilled and high-skilled workers in Western Europe has been affected by offshoring of production in manufacturing to Eastern Europe and the BRIC nations. In order to give a proper answer to this question this paper will make use of a theory of tasking trade as developed by Grossman and Rossi-Hansberg 3

5 (2008)[10]. This theory and the works of Geishecker[7, 8, 9], amongst others, will shed light on how the research question could be answered best and which data are necessary for that purpose. 1.1 Research questions and hypothesis In order to get a firm grasp on how offshoring affects wages and wage inequality it is necessary to draw up some research questions and a hypothesis. This hypothesis can then be tested using an econometric regression. The results of this regression will reveal the effects of offshoring (amongst other effects) on wages and wage inequality. The regression results will then be used to evaluate the hypothesis. The research questions should both aid in formulating the hypothesis and help identify the mechanisms that are involved in the effects of offshoring on wages. This thesis will formulate auxilliary research questions that deal with the mechanisms that mediate between offshoring and wage as well as lead up to the main research question. In short two mechanisms have to be discussed: the factors that play a role in the offshoring decisions of firms on when to offshore, what tasks to offshore and to where these tasks will be offshored; and the way these decisions and the resulting offshoring translate into wage effects. Attention should also be paid to earlier research that has been conducted on this matter, which will also be a subject of the auxilliary research questions. The main research question will ask for the empirical results on the effects of offshoring on the wage level and wage inequality and will be pointed towards the regressions Auxilliary research questions Based on the criteria set in the previous paragraph the following auxilliary questions can be formulated: 1. What factors influence the decisions of firms on when to offshore, which tasks to offshore and to where these tasks will be offshored? 2. What mechanisms mediate between offshoring and wages and how do these mechanisms affect the wage level and wage inequality in Western European manufacturing? 3. What is currently known, from earlier research, of the effects of offshoring on wages and wage inequality in Western European manufacturing? 4. What are proper measures for offshoring, wages and wage inequality; and how can they be properly employed to estimate the effects of offshoring on wages? 5. What do econometric regressions say about the relation between offshoring and wage inequality? 6. Given the results of the econometric analysis, how does offshoring affect wage inequality? 4

6 1.1.2 Main research question The preceding discussions and the auxilliary questions lead to the formulation of the following main research question: How, in what direction, and to what extent, does offshoring of production from Western Europe to other parts of the world affect the wage level and wage inequality between low-skilled and high-skilled workers in Western European manufacturing industries? Hypothesis In public opinion it is sometimes alleged that offshoring leads to lower wages for and more wage inequality amongst workers. Because of that I would like the regressions run in this thesis to prove whether the following hypothesis is true: Offshoring causes the wage level in Western European manufacturing to fall and causes wage inequality in Western European manufacturing to rise 1.2 Outline of the thesis Now that the research question and the hypothesis have been formulated, I will spend the rest of this introductory chapter on explaining the layout of the thesis. Chapter 2 will explain the theories on offshoring. These theories will shed light on the factors that influence firm decisions on offshoring and its extent and direction and on the mechanisms through which offshoring affects wages and wage inequality. Chapter 2 will start with a theory from Grossman and Rossi-Hansberg (2008)[10] and will concentrate on which tasks will be offshored to which destination. The discussion of the effects of offshoring on wages and wage inequality will be started by looking at a literature review on theories and empirical studies on the effects of trade on wage inequality. This general discussion will be made more concrete by discussing a model from Baldwin and Robert-Nicoud (2007)[2]. This model makes it possible to explore mechanisms which might mediate between offshoring and its effects on wages and wage inequality and will serve as a theoretical bedrock of the regressions from chapter 4. From the theories of Grossman/Rossi-Hansberg and Baldwin/Robert-Nicoud, two predictions will be made with regards to offshoring and anticipating the regressions and associated data analysis: one about the mechanism behind offshoring itself, and one about the mechanism mediating between offshoring and its wage and wage inequality effects. Works from Geishecker (2006, 2008, 2010)[7, 8, 9] will be discussed in order to get an idea of scientific knowledge on the effects of offshoring on wages and wage inequality gained so far. Chapter 3 will deal with the methodology of estimating the effects of offshoring on wages and wage inequality and serves as a stepping stone to the actual regressions in chapter 4. In chapter 3 the basics of the econometric model 5

7 behind the regressions will be explained. Its first section will provide the rationale of the OLS regressions conducted and will introduce the regressors used to estimate the wage. It will deal with the rationale behind the regressors and tell what the coefficients of the regressors measure. At the end of the first section of chapter 3 the regression equations will be presented. In the second section of chapter 3 the dataset underlying the model will be discussed. It deals with the process of constructing the dataset and the variables that had to be dropped for a reliable regression. The last section of chapter 3 deals with the variables in more detail. It especially deals with the sources of the data and the construction of offshoring indicators (unlike other variables these indicators were not directly obtained from databases, but had to be constructed from other statistics). Chapter 4 contains the actual regressions, which are subdivided in single destination regressions (lumping all offshoring together), and separate destination regressions (spliting offshoring into three destinations). The regressions are discussed in the first section, whereas further data analysis is conducted in the second section. This section not only explains the outcomes of the regression but also tries to look for the actual mechanisms responsible for the relationships exposed in the regresssions. It does so by looking at descriptive statistics generated by the data, and correlations of these descriptive statistics. These correlations can be used to look at the plausibility of the predictions made in chapter 2. Chapter 5 concludes and tries to answer the research questions and the hypothesis formulated in this introductory chapter. 6

8 Chapter 2 Theories about offshoring 2.1 Offshoring theories In order to answer the research questions posed in the previous chapter, and before we will empirically determine whether the hypothesis from the previous chapter holds, some theories about outsourcing and offshoring will be reviewed. The first theory that will be discussed is one from Grossman and Rossi- Hansberg (2008)[10], who explained offshoring patterns through their task trade theory. The task trade theory is novel amongst trade theories in the sense that it does not look at trade in final products but rather at the tasks executed to produce these final products. It tries to explain why tasks are performed in a certain country and when it is worthwile for a firm to offshore production to another country. According to Grossman and Rossi-Hansberg firms face a trade off between executing tasks were they can be most efficiently done and keeping monitoring costs low. This trade off exists as moving production away from the headquarter country to countries were production is most efficients involves monitoring costs. Production is most efficient were output is largest, due to (external) economies of scale. Monitoring costs increase with the complexity of the task, as complex tasks require more monitoring of workers in order to be executed correctly. These monitoring costs are furthermore influenced by the relative wage. According to Grossman and Rossi-Hansberg a firm will outsource a task if the benefits from economies of scale in the destination country outweigh the cost of offshoring and the production costs in the destination country. In general tasks will be offshored to the country with the highest output, where economies of scale are more prevalent. Between countries with the same output and size tasks will be offshored to the country with lower wages. Grossman and Rossi-Hansberg discern three types of tasks: those with low offshoring costs, those with intermediate offshoring costs and those with high offshoring costs. Tasks with high offshoring costs will not be offshored, as the gains from economies of scale do not outweigh monitoring costs. Tasks with low or intermediate costs can be offshored, but will not necessarily move in one direction. A task that is being offshored to a country with larger output, will use up labour in that country and thus push up wages there. This could cause the 7

9 initial offshoring to be dampened, as part of that offshoring will then flow back to the source country. A proposition made by Grossman and Rossi-Hansberg is that tasks with low offshoring costs tend to be done in countries with lower wages and output, while tasks with intermediate offshoring costs tend to be done in countries with higher wages and output. This proposition arises from the fact that countries with a larger output generate more economies of scale, which makes offshoring more worthwile and enables monitoring costs (part of which are wages) to be higher. The model from Grossman and Rossi-Hansberg can explain for example why China is specialised in textiles, while the East Asian Tigers (Hong Kong, Singapore, South Korea and Taiwan) became specialised in electronic equipment and cars when wages rose. The model from Grossman and Rossi-Hansberg leads to the following prediction. Prediction 1: Tasks with low offshoring costs will be offshored to countries with lower wages and output, whereas tasks with intermediate offshoring costs will be offshored to countries with higher wages and output. Translated to this thesis it means: tasks with low complexity (tasks with a low skill intensity) will be offshored to countries with lower wages and output, whereas tasks with high complexity (tasks with a high skill intensity) will be offshored to countries with higher wages and output. The model of Grossman and Rossi-Hansberg is a good starter for this thesis as it reveals the preconditions for offshoring, but for a theoretical approach of the research question it is necessary to know how offshoring affects wages and wage inequality. The beginning of an answer on this question is given by a literature review on the relationship between trade and wage inequality written by Freeman (1995)[6]. In this literature review he starts with the observation that the living standards of less skilled workers has dropped in both Europe and the United States (in Europe this took the form of an increase in unemployment while in the U.S. it took the form of decreasing wages). Freeman then goes on to deal with the question of how large the role of import competition and production shifting to low-wage countries is in affecting the living standards (including wages) of less skilled workers compared to skilled workers and how trade economists have dealt with this question. Freeman outlines the central position of factor price equalisation in determining the extent of the effects of trade and offshoring on wages and wage inequality. If one assumes complete factor price equalisation to hold, one will arrive at the conclusion that trade and offshoring matter a lot to the development of wage inequality. If one relaxes this assumption, one will conclude that the role of trade and offshoring in affecting wage inequality is much smaller. The more factor price equalisation there is, the more the wages of similarly skilled workers across countries will be equalised as well. If industries employing relatively many low-skilled workers offshore more to and face more import competition from low-wage countries, this could lead to downward pressures on the wages of low-skilled workers in high-skilled countries. According to Freeman, the existence of huge differences in payment between high- and low-wage countries makes the assumption of factor price equalisation a strong and not very realistic one, but the current trade patterns of the U.S. and Europe (imports of products low in skill-intensity, exports of skill-intensive products) still lends some justification to factor price equalisation. 8

10 Freeman names two empirical approaches of tackling the question of how much and to what extent import competition and offshoring affect the wages of less skilled workers: factor content studies and price studies. Factor content studies focus on how trade and offshoring affect the effective demand for labour (split between domestic and foreign demand) at given wages and prices and use the wage elasticities from other studies to estimate the effects of trade and offshoring. According to Freeman most factor content studies estimate that 10 to 20% of the rise in wage inequality is due to trade. Factor content studies have been criticised for wrongly estimating the true effects of trade on wage inequality due to the fact that these studies assume wages to constant in the short run. As such they would perform better in estimating the effects of trade on labour demand and wages when there is wage rigidity (as in Europe), and worse when there are flexible wages (like in the U.S.). These drawbacks are exacerbated by the fact that factor content studies take imports to be exogenous, which ignores the effects generated by feedback loops. Price studies circumvent these drawbacks by estimating the price effects of trade rather than quantity effects. The conclusions drawn from price studies hinge on the assumptions that wages move along with prices. If prices of domestic goods fall as a result of import competition and offshoring, wages should fall as well. Most price studies show modest negative effects of trade on prices (and wages). According to Freeman price studies have there own drawbacks, such as the measurement errors in price data and aggregation problems. Combining the fact that both methods have drawbacks with the fact that studies do not provide clear results on the effect of trade on wage inequality led Freeman to conclude that studies on trade and wage inequality should be presented with humility. The following model developed by Baldwin by Baldwin and Robert-Nicoud (2007)[2] is an example of a theoretical model on offshoring and wages that is based on standard trade theories. While the previous model of Grossman and Rossi-Hansberg did not feature hypotheses on wages itself, the model developed by Baldwin and Robert-Nicoud (2007)[2] does, even though it is only a minor part of their model. As can be seen in the theorems below, the model from Baldwin and Robert-Nicoud features factor price equalisation, but only as a result of trade in tasks rather than trade in final goods. Baldwin and Robert-Nicoud start from the Heckscher-Ohlin model (HO) with two countries ( Home and Foreign ) and its four theorems. In the paper from Baldwin and Robert-Nicoud[2] we can find the theorems: 1. The Factor Price Equalisation theorem, which states that factor prices are equal across countries when measured in effective units of factors. That is, nominal factor prices differ only because of differences in technology levels. 2. The Heckscher-Ohlin theorem, which states that countries export goods of the type for which they are most properly endowed (i.e. countries which are relatively more endowed with labour export labour-intensive goods). 3. The Stolper-Samuelson theorem provides the link between goods prices and factor prices. For example when the price of a capital-intensive good 9

11 rises, the interest rate r rises more than proportionally, while wages w fall. 4. The Rybczynski theorem, which states that an increase in a country s endowment raises the production of goods intensive in that endowment (i.e. if a country s labour endowment increases relative to its capital endowment it more than proportionally raises its production of labourintensive goods) So far there seems to be nothing new about the model of Baldwin and Robert-Nicoud, as this is simply a summary of the well-known Heckscher- Ohlin model. The novelty about the model of Baldwin and Robert-Nicoud is that they have altered the Heckscher-Ohlin model in such a way as to include trade of tasks and intermediate goods next to final goods. Baldwin and Robert- Nicoud discern two types of tasks: labour-intensive L-tasks and tasks intensive in human capital (K-tasks). L-tasks are performed by unskilled workers (L-workers), while K-tasks are performed by skilled workers (K-workers). Goods that mainly consist of L-tasks are considered to be labour-intensive (Lintensive), whereas goods that mainly consist of K-tasks are considered intensive in human capital (K-intensive). Baldwin and Robert-Nicoud denote L-intensive goods with X and K-intensive goods with Y. For the rest of this section I will use the term capital-intensive to denote goods and tasks that are intensive in human capital. The mechanisms of this modified HO-model are the same as those of the standard HO-model, but the modification (such as the inclusion of tasks) may cause the standard theorems to not hold. Therefore Baldwin and Robert-Nicoud have drawn up propositions of their own (see below). The model sees offshoring as production conducted in foreign countries using superior Home country technologies, though paying Foreign factor prices, which are lower. This is akin to Foreign production factors moving to Home to be employed there, while being paid a Foreign wage. This is why Baldwin and Robert-Nicoud call their concept of offshoring shadow migration. The difference between the marginal product of labour generated by Home s superior technology and the lower wages paid to workers in Foreign are a cost-saving, with consequences for goods prices and factor prices. One could say that Home has a comparative advantage in technology as its technology is superior. Baldwin and Robert-Nicoud assume this comparative advantage will always hold, as Home s technology will not degrade if its firms move (offshore) production to Foreign. Foreign can be considered to have a comparative advantage in labour as its labour is both abundant and cheap. By offshoring production to Foreign Home firms want to gain from Foreign s comparative advantage in labour, whilst also gaining from Home s comparative advantage in technology (which is assumed to hold). As a result Home can improve its terms of trade through its offshoring firms. This may affect the wages of Home workers. In their paper Baldwin and Robert-Nicoud outlined these consequences in six propositions, the three most relevant of which I directly cite[2] below: 1. Offshoring of either type of labour changes the world price of final goods. The relative price of capital-intensive good Y rises if the shadow K- 10

12 migration is proportionally less than the shadow L-migration. This is a price effect 2. Offshoring can be viewed as shadow migration of Foreign L and K. The impact on Home production follows a Rybczynski-like pattern, if offshoring implies a very unbalanced ratio of K versus L shadow migration, but the output of both sectors may rise if the amounts of L and K shadow migration are fairly similar. This is a production effect. 3. Offshoring raises the real wage of Home L-workers if the offshoring implies cost savings that are sufficiently larger in the L-intensive sector than in the K-intensive sector; the real wage of K-workers rises less or actually falls; it falls if the cost-savings are sufficiently skewed towards the L-intensive sector. Apart from terms of trade effects, wages of Foreign L- and K-workers are unaffected. This is a wage effect. Of the three propositions cited above, the third proposition on wages is most relevant to the subject of this thesis, as it directly deals with the effects of offshoring on wages. Baldwin and Robert-Nicoud s first proposition is relevant to wages as the degree of shadow migration determines the prices of goods, which through the Stolper-Samuelson theorem influence wages (as per the third proposition). According to the third proposition the wages of workers in labour-intensive industries rise if cost savings from offshoring in the labour-intensive sector are substantially larger than in the capital-intensive sector. This insight results from combining the effects of shadow migration with the Stolper-Samuelson theorem. Less shadow migration of labour leads to an increase in the price of labour-intensive goods and, following the Stolper- Samuelson theorem, to an increase in the wages of unskilled workers (L-workers). Depending on the extent to which labour-intensive cost-savings exceed capitalintensive cost-savings and the extent to which shadow migration of labour is smaller than shadow migration of capital, the wages of K-workers rise slightly or fall. This paragraph s conclusion can be summarised in the following prediction. Prediction 2: If the costs savings are substantially larger in the labourintensive sector than in the capital-intensive sector, and if shadow migration (offshoring) of labour is smaller than that of capital, the wages of unskilled workers in the home country will rise. The wages of skilled workers in the home country will rise modestly when differences in cost-saving and shadow migration are small, but will fall when these differences are large. The opposite will hold if cost savings in the capital-intensive sector are larger than in the labour-intensive sector, and if shadow migration (offshoring) of labour exceeds that of capital. 2.2 Earlier empirical work and trends in offshoring and wages Several empirical research papers have been devoted to studying the effects of offshoring on wages. Feenstra and Hanson (1996)[4] state that most studies on the effects of trade on wage inequality focus too much on macro level data and 11

13 assume that import competition effects are the same for every industry, which neglects the within-industry effects of trade as well as changes in the skills composition of industries. Therefore Feenstra and Hanson introduce outsourcing as a process through which the effects of trade on wage inequality can be measured. To that end they construct outsourcing indicators from import data and import purchase data. Feenstra and Hanson then measure the effect of outsourcing on labour demand for skilled labour. They do this for two time periods: and For the period Feenstra and Hanson found a positive and significant effect of outsouring on the labour demand for skilled workers, meaning that the relative demand for unskilled workers dropped as did their relative wage. For no significant effects were found. Studies on the effects of trade and offshoring on wage inequality in Europe are rare. One of such studies is a study from Anderton and Brenton (1999)[1] about the effects of outsourcing on wage inequality in the United Kingdom. Like Feenstra and Hanson, they use disaggregated import data to contruct offshoring variables. They furthermore split imports according to their origin. They identify imports coming from low-wage countries, and imports coming from high-wage countries. For both origins they check the effects of outsourcing on the wage share of less-skilled workers in total wages. Anderton and Brenton found this share to be falling in the 1970s and 1980s, especially if outsourcing came from low-skilled countries. They also found that industries employing low-skilled workers were more affected than other industries. Also worth mentioning is the collection of papers from Ingo Geishecker, who investigated the effect of offshoring on wages and wage inequality in Germany. Geishecker (2008)[8] used micro-level wage data from a household survey, skills data, employment data and constructed indicators on offshoring to determine how offshoring affects wages. Geishecker finds that offshoring reduces the wages of unskilled workers, while raising them for skilled workers. This confirms the main hypothesis formulated in this thesis, namely that offshoring leads to lower wages for unskilled workers and to wage divergence between unskilled and skilled workers. Geishecker also conducted research into the question whether labour market institutions influenced the wage outcomes of offshoring (Geishecker, 2010)[9]. He found that labour market institutions did not have a substantial effect on the wage outcomes of offshoring. The Geishecker papers also showed that offshoring from Germany to Eastern Europe took off after the opening of the Iron Curtain in the early 1990s. Later on offshoring to other countries (most notably the BRIC countries) would take off as well. The studies above are rare examples of studies on the effects of offshoring on wage inequality in Europe, and studies that are Europe-wide are even rarer (if not non-existent). The regressions used by Anderton and Brenton and Geishecker directly estimate the effects of offshoring on wages or wage shares. The regressions discussed and run in the coming chapters will estimate effects of offshoring on wage inequality for 10 European countries. This will be done by directly estimating the effects of offshoring on wages and using the methods Geishecker used in his studies. The theories and empirical work that were discussed in this chapter form the bedrock of these regressions. 12

14 Chapter 3 Model setup and data 3.1 Model setup In chapter 2, we reviewed some theories about offshoring, took a look at some earlier pieces of empirical research on offshoring and wages, and used the latter to establish trends in offshoring and wages. With this knowledge we can develop an econometric model for testing the hypothesis described in the introduction. The test amounts to checking whether wage inequality rises or falls als a result of offshoring Basics of the model At first Ordinary Least Squares (OLS) regressions will be used to estimate the effects of offshoring on wages, because no specific specifications are needed to obtain meaningful estimates. There is however endogeneity between wages and offshoring, which requires a Two-stage Least Squares (2SLS) regression, involving instrumental variables. The next subsection will deal with endogeneity and the possibilities of a 2SLS regression. It should be noted that two different types of regression will be conducted: single destination regressions and separate destination regressions. In single destination regressions all offshoring to destinations outside of Western Europe is lumped together into one offshoring indicator. Single destination regressions are used to explore the general effects of offshoring on wages. In separate destination regression offshoring is split into three geographic areas thought to entail different patterns of offshoring: Central and Eastern European countries (CEC); Brazil, Russia, India and China, together called the BRIC countries (BRIC); and the rest of the world (RotW). Offshoring is attributed to each of the three destinations with the help of imports statistics and inputoutput tables (for details see the section on data). Separate destination regressions are used in order to check whether the effects of offshoring on wages differ between offshoring destinations. This may be helpful in revealing the causes of (differences in) the wage effects. Geishecker also used separate destination regressions in his 2006 paper (Geishecker, 2006)[7]. Furthermore these types of regression are subdivided into subtypes as two definitions of offshoring will be used: the narrow definition, which measures 13

15 offshoring of an industry as far as it destined for the same industry abroad (taking only imports from the same industry into account); and the wide definition, which measures offshoring no matter the industry abroad for which it is destined (taking all imports into account, including those from other industries). Offshoring regressors (and associated interaction terms and instrumental variables) using the narrow definition use the abbreviation OUTS (from outsourcing) combined with the Narrow suffix (i.e. OUTSNarrow), while those using a wide definition use the suffix Wide (i.e. OUTSWide). When running a separate destination regression further suffixes are added to denote destinations. For example narrow definition offshoring to Central and Eastern European countries (CEC) is denoted by OUTSNarrowCEC. In order to test our hypothesis the econometric model should include real hourly wages as a dependent variable. The (sets of) regressors included in the model measure different effects of offshoring on wages. The first of such effects is the direct effect of offshoring on wage levels, measured through constructed offshoring indicators (for details concerning the construction of these indicators see the section on data). Positive values of the coefficient(s) of this (these) indicator(s) denote an increase in the wage level, whereas negative values denote a decrease in the wage level. In order to prevent bias in the estimates of the effect of offshoring on the wages of people with different skills levels, the direct effect of skills and education (no matter the level of offshoring, on wages) has to be measured. For that purpose a set of three regressors will be introduced, representing the share of employees in a particular industry that is either medium- or high-skilled, or that has not reported its educational attainment (non-respondents). The regressors are therefore called MediumSkilled, HighSkilled and NoAnswer. By taking these shares the skill effects are automatically corrected for the size of industries (which differs greatly across industries). The share of employees that is low-skilled serves as the baseline, meaning that the regressors measure the differences between the wage levels of the medium-skilled, high-skilled and non-respondents respectively, and the wage level of the low-skilled. Positive values of the coefficients of these regressors indicate that the wages for the skills group involved is higher than those of the low-skilled. For example if the coefficient of MediumSkilled reads 6, then an increase by one percent point in the share of employees that is medium-skilled results in an increase in the hourly wage of 6 euros when compared to the wages of the share of low-skilled workers. The regressors measuring skill effects are always the same, regardless of the regression type and offshoring definition. The effects of offshoring on wage equality are obtained by estimating the effect of the skills level and educational attainment on wages given there is a specific level of offshoring. This can be done by multiplying, for each observation, each the offshoring indicators by each of the skill regressors. The number of interaction terms resulting is obtained by multiplying the number of destinations (one or three depending on the regression type) with the number of skill regressors (three: MediumSkilled, HighSkilled and NoAnswer). The number of interaction thus varies between three (in a single destination regression) and nine (in a separate destination regression). The interaction terms measure the differences between the change of the wage level of the medium-skilled, highskilled and non-respondents respectively, and the change of the wage level of the low-skilled. Low skills again serve as the baseline. As such the interaction 14

16 terms measure the changes in wage inequality between workers of different skill levels and educational attainment. Positive values of the coefficients of the interaction terms indicate an increase in wage inequality, whereas negative values indicate a decrease in wage inequality. For example if the coefficient OUTSNarrowCEC MediumSkilled reads 5, it means that narrow defintion offshoring to Central and Eastern European countries causes the change in the wage level of the medium-skilled to be 5 euros more positive then the change in the wage level of the low-skilled, which means that wage inequality has increased. Another example: if the coefficient OUTSWideBRIC HighSkilled reads -4, it means that wide defintion offshoring to the BRIC countries causes the change in the wage level of the medium-skilled to be 4 euros less positive than the change in the wage level of the low-skilled, which means that wage inequality has decreased. Not all changes in wages are caused by offshoring. Other factors also influence wage setting. They include country-specific effects (such as labour market institutions), industry-specific effects (such as product market conditions) and business cycle and trend effects (such as recessions). In order to capture these effects dummy variables for each of these effects have to be introduced: country dummies for each of the countries, in order to capture countryspecific effects; industry dummies for each of the industries, in order to capture industry-specific effects; and year dummies for each year, in order to capture time-specific effects. Effects arising from these regressors can be seen as effects that arise in the country/industry/year involved on top of the other effects just mentioned. Positive coefficients indicate that a specific effect contributes positively to wages, whereas negative coefficients contribute negatively towards wages, when measured against the baseline dummy. Of each set of dummies one will serve as a baseline dummy, and will thus be dropped from the regression. Note that all data mentioned in this paragraph are gathered on the industry level. The sources and use of the data mentioned will be explained in more detail in the next paragraph on data, but the types of variables will be summarised in tables below Endogeneity The previous subsection briefly touched upon the existence of endogeneity in the sample. It is probable that there wages are not only influenced by offshoring, but that offshoring is also influences by wages. Labour costs may feature heavily in firm decision about whether to offshore production or not. This endogeneity may cause the coefficients to wrongly estimate the effects of offshoring on wages. In order to correct for endogeneity 2SLS regressions were run with lagged offshoring indicators, but these regressions were identical to the OLS regressions reported in chapter 4 (probably due to the lagged indicators being very much correlated with the indicators themselves). Therefore the 2SLS regressions will not be reported in chapter 4. Only the instruments themselves will be briefly discussed, for reasons of transparency. As only OLS regressions will be run, it should be noted that the conclusions drawn in this thesis are conditional on the assumption that offshoring itself is exogenous (i.e. not influenced by wages). 15

17 Variables used in regression on wages; narrow offshoring definition Regressors by type (effect measured) Dependent Offshoring indicators Skills data Interaction terms (wage inequal- (direct effect on (skill effect ity) wages) on wages) Real hourly Separate destinations MediumSkilled Separate destinations wage OUTSNarrowCEC HighSkilled OUTSNarrowCEC Medium- Skilled OUTSNarrowBRIC NoAnswer OUTSNarrowCEC HighSkilled OUTSNarrowRotW OUTSNarrowCEC NoAnswer Single destination OUTSNarrowBRIC Medium- Skilled OUTSNarrow OUTSNarrowBRIC HighSkilled OUTSNarrowBRIC NoAnswer OUTSNarrowRotW Medium- Skilled OUTSNarrowRotW HighSkilled OUTSNarrowRotW NoAnswer Single destination OUTSNarrow MediumSkilled OUTSNarrow HighSkilled OUTSNarrow NoAnswer Industry dummies (industry-specific effects Country dummies (countryspecific effects) Year dummies (timespecific effects) dat d15 d1999 (base) dbe d17 d2000 ddk d20 d2001 dfr d21 d2002 dde d22 d2003 die d24 d2004 dnl (base) d25 d2005 des d26 d2006 dse d27 d2007 duk d28 d29 d31 d32 d33 d34 d35 d36 (base) Table 3.1: Variables used in regressions: narrow offshoring definition 16

18 Variables used in regression on wages; wide offshoring definition Regressors by type (effect measured) Dependent Offshoring indicators Skills data Interaction terms (wage inequal- (direct effect on (skill effect ity) wages) on wages) Real hourly Separate destinations MediumSkilled Separate destinations wage OUTSWideCEC HighSkilled OUTSWideCEC MediumSkilled OUTSWideBRIC NoAnswer OUTSWideCEC HighSkilled OUTSWideRotW OUTSWideCEC NoAnswer Single destination OUTSWideBRIC MediumSkilled OUTSWide OUTSWideBRIC HighSkilled OUTSWideBRIC NoAnswer OUTSWideRotW MediumSkilled OUTSWideRotW HighSkilled OUTSWideRotW NoAnswer Single destination OUTSWide MediumSkilled OUTSWide HighSkilled OUTSWide NoAnswer Industry dummies (industry-specific effects Country dummies (countryspecific effects) Year dummies (timespecific effects) dat d15 d1999 (base) dbe d17 d2000 ddk d20 d2001 dfr d21 d2002 dde d22 d2003 die d24 d2004 dnl (base) d25 d2005 des d26 d2006 dse d27 d2007 duk d28 d29 d31 d32 d33 d34 d35 d36 (base) Table 3.2: Variables used in regressions: wide offshoring definition 17

19 Instrumental variables used in 2SLS regressions by offshoring definition Narrow definition Wide definition Separate destinations Separate destinations OUTSNarrowCECLag OUTSWideCECLag OUTSNarrowBRICLag OUTSWideBRICLag OUTSNarrowRotWLag OUTSWideRotWLag OUTSNarrowCECLag MediumSkilled OUTSWideCECLag MediumSkilled OUTSNarrowCECLag HighSkilled OUTSWideCECLag HighSkilled OUTSNarrowCECLag NoAnswer OUTSWideCECLag NoAnswer OUTSNarrowBRICLag MediumSkilled OUTSWideBRICLag MediumSkilled OUTSNarrowBRICLag HighSkilled OUTSWideBRICLag HighSkilled OUTSNarrowBRICLag NoAnswer OUTSWideBRICLag NoAnswer OUTSNarrowRotWLag MediumSkilled OUTSWideRotWLag MediumSkilled OUTSNarrowRotWLag HighSkilled OUTSWideRotWLag HighSkilled OUTSNarrowRotWLag NoAnswer OUTSWideRotWLag NoAnswer textitsingle destination textitsingle destination OUTSNarrowLag OUTSWideLag OUTSNarrowLag MediumSkilled OUTSWideLag MediumSkilled OUTSNarrowLag HighSkilled OUTSWideLag HighSkilled OUTSNarrowLag NoAnswer OUTSWideLag NoAnswer Table 3.3: Instrumental variables used in 2SLS regressions Regression equations Now that all the characteristics and variables of the model have been explained it is appropriate to look at the resulting regression equations in order to obtain a good overview of the regressions. In the appendix (Appendix A) the regression equations are presented in two sections: section A.1 showing equations applicable to single destination regressions and section A.2 showing equations applicable to separate destination regressions. Both sections show two regression equations: one applicable to regressions in which only country and year dummies are used and one applicable to regression in which all dummies (country, industry and year) are used. All equations (both single and separate destination) estimate the hourly wage wage cit. The first term in each of the equations in the appendix is the intercept (β 0 ), which can be interpreted as the hourly wage that would prevail if none of the effects measured in the regressions would apply. From then onwards, the appearance of the equations differs between the types. Let me discuss the single destination regression equations first, starting with the one featuring only country and year dummies. In this equation the coefficient β 1 represents the direct effect of offshoring on the wage, as exerted by the offshoring indicator. the coefficients β 2 to β 4 measure the direct effects of skill and educational attainment on the wage. The coefficients β 5 to β 7 accompany the interaction terms and measure the effects of offshoring on the wages of particular skills groups, given there is offshoring. β 5 measures the change in wages of medium-skilled workers compared to those of low-skilled workers, β 6 measures the change in wages of high-skilled workers compared to those of low-skilled workers and β 7 measures the change in wages of workers not reporting their skills compared to those of low-skilled workers. As such the co- 18

20 efficients β 5 to β 7 measure changes in wage inequality as a result of offshoring. The remaining coefficients are associated with dummies. The coefficients β 8 to β 16 capture country-specific effects, while β 17 to β 24 capture time-specific effects associated with the year of that particular beta. The regression equation applicable to a single destination regression with country, industry and year dummies (the single destination all dummies equation) starts out the same as before: β 0 is the intercept, β 1 represents the direct effects of offshoring on the wages, β 2 to β 4 the direct skill effects on wages, β 5 to β 7 the effects of offshoring on wage inequality, and β 8 to β 16 country-specific effects. Industry-specific effects will now be captured by the coefficients β 17 to β 32, which are associated with industry dummies named after their ID. In the all dummies equation the time-specific effects are represented by the coefficients β 33 to β 40. The separate destination regression equations have a buildup simular to the single destination regression equations, but they are considerably longer due to the fact that there are three times as many offshoring indicators and interaction terms (because there are three destinations instead of one). First the country and year dummy equation is discussed. β 0 is the intercept as before. β 1 to β 3 represent the direct effect of offshoring on wages (β 1 for offshoring to Central and Eastern Europe, β 2 for offshoring to the BRIC countries and β 3 for offshoring to the rest of the world. β 4 to β 6 represent direct skill effects. β 7 to β 9 represent the effects of offshoring to Central and Eastern Europe on wage inequality. β 10 to β 12 represent the effects of offshoring to the BRIC countries on wage inequality, whereas β 13 to β 15 represent the effects of offshoring to the rest of the world on wage inequality. The country-specific effects are represented by β 16 to β 24. Time-specific effects are captured by β 25 to β 32. The separate destination all dummies equation is quite similar to the previous one. β 0 is the intercept. β 1 to β 15 represent the effects of interest for this thesis (direct offshoring effects for three destinations, direct skill effects, effects on wage inequality for three destinations) exactly as in the previous equation I presented. β 16 to β 24 capture country-specific effects as in the previous equation. β 25 to β 40 measure industry-specific effects, whereas time-specific effects are captured by β 41 to β 48. One thing should be noted. There is always one dummy serving as the base dummy, which causes the number of coefficients for each set of dummies to be one less than the number of dummies in the dataset. The base country dummy is dnl (Netherlands), the base industry dummy is d36 (Recycling), whereas the base year dummy is d1999 (1999, the first year featuring in the regressions). The base dummies do not feature in the regression equations shown in appendix A. In the regression equations from the appendix all offshoring indicators and interaction terms are introduced in their narrow definition. When wide definition regressions are run narrow definition indicators should be replaced with wide definition indicators. So the reader may read Narrow as Wide when wide definition regressions are being discussed. 3.2 Dataset The dataset consists of industry level data from 1995 to 2008, available for 11 countries (Austria, Belgium, Denmark, France, Germany, Ireland, Luxem- 19

21 Industries featuring in regressions Industry ID Industry description 15 Manufacture of Food Products and Beverages 16 Manufacture of Tobacco Products 17 Manufacture of Textiles 18 Manufacture of Wearing Apparel; Dressing and Dyeing of Fur 19 Tanning and Dressing of Leather; Manufacture of Luggage, Handbags, Saddlery, Harness and Footwear 20 Manufacture of Wood and of Products of Wood and Cork, except Furniture 21 Manufacture of Paper and Paper Products 22 Publishing, Printing and Reproduction of Recorded Media 23 Manufacture of Coke, Refined Petroleum Products and Nuclear Fuel 24 Manufacture of Chemicals and Chemical Products 25 Manufacture of Rubber and Plastics Products 26 Manufacture of Other Non-Metallic Mineral Products 27 Manufacture of Basic Metals 28 Manufacture of Fabricated Metal Products, except Machinery and Equipment 29 Manufacture of Machinery and Equipment NEC 30 Manufacture of Office Accounting and Computing Machinery 31 Manufacture of Electrical Machinery and Apparatus NEC 32 Manufacture of Radio Television and Communication Equipment and Apparatus 33 Manufacture of Medical Precision and Optical Instruments, Watches and Clocks 34 Manufacture of Motor Vehicles Trailers and Semi-Trailers 35 Manufacture of other Transport Equipment 36 Manufacture of Furniture; Manufacturing NEC Table 3.4: Industries featuring in regressions bourg, Netherlands, Spain, Sweden and United Kingdom). The industry sectors featuring in the regression are all manufacturing sectors and can be found in table 3.4. The industries are from the NACE 1.1 Rev. classification, used by the European Union. Data from the Organisation for Economic Cooperation and Development (OECD) and the International Labour Organisation (ILO) follow the ISIC 3 classification, which is compatible with NACE 1.1 Rev. I decided to drop some countries and industries from the dataset, because the share of missing data from these countries and industries was higher than 30%. This was the case for Luxembourg, as well as data from industries with the IDs 16, 18, 19, 23 and 30, which are mostly small industries that do not provide reliable data. Most countries have data from 1996 to 2007, while only some have data from 1995 to For 1998 most countries had no reliable data on skills. Prior to 1999 Eurozone countries still had separate currencies with the euro s predecessor the ECU being an accounting unit. This complicated the conversion of wages from national currencies to euros/ecu. Even though I was able to make reliable conversions from national currencies into ECU for pre-1999 cases, this complication, together with 1998 not providing skills data and some countries having no 1995 data, was a reason to drop all observations from before Only a few countries have data all the way till 2008, but most have data uptil and including 2007, so I dropped observations from Ireland and Netherlands only have data uptil and including 2005, but this was 20