November 2012

Offshoring and the Elasticity of Labour Demand Neil Foster♦ Johannes Poeschl# Robert Stehrer∗

Abstract This paper examines the impact of offshoring on labour demand and labour demand elasticities for a sample of 40 countries over the period 1995-2009 using the recently compiled World Input-Output Database (WIOD). Estimating conditional and unconditional labour demand models we find that both narrow and broad offshoring have impacted positively upon labour demand, an effect driven by the scale effect of offshoring. Despite this observed benefit of offshoring, we also show that offshoring has tended to increase labour demand elasticities, which can increase the vulnerability and reduce the bargaining power of workers.

Keywords: Labour Demand, Offshoring JEL Classification: F16

♦

Neil Foster, Vienna Institute for International Economic Studies (wiiw), Rahlgasse 3, A1060, Vienna, Austria. Email: [email protected] # Johannes Poeschl, Vienna Institute for International Economic Studies (wiiw), Rahlgasse 3, A1060, Vienna, Austria. Email: [email protected] ∗ Robert Stehrer, Vienna Institute for International Economic Studies (wiiw), Rahlgasse 3, A1060, Vienna, Austria. Email: [email protected]

1

1. Introduction The ongoing process of globalisation has seen the increasing frequency and extent of international outsourcing – or offshoring – of production, involving the contracting out of activities that were previously performed within a production unit to foreign subcontractors. While such offshoring is expected to bring about gains for consumers and producers there is a concern that the costs of offshoring fall disproportionately on workers, and in the developed world at least on low-skilled workers in particular. While a large empirical literature on the impact of offshoring on labour markets exists, the majority of this work has been concerned with offshoring’s impact on the wage or cost shares of low- and high-skilled workers, that is, on the skill composition of labour demand. This literature tends to support the view that offshoring has been one of the factors lowering the demand for low-skilled workers in developed countries, but that it has not been the major cause of this shift in relative labour demand.

In this paper we move away from the impact of offshoring on relative labour demands and consider its impact on employment more generally, and its impact on the elasticity of labour demand in particular. From a theoretical perspective there are two main direct effects of offshoring on employment. The first is a ‘technology’ or ‘substitution’ effect that reflects the destruction of jobs that occurs when firms relocate part of their production activities overseas. The second is a ‘scale’ effect that captures the creation of jobs following the expansion in industry output that may arise as a result of the productivity gains from offshoring. A third indirect ‘substitution effect’ may also be relevant, in which offshoring affects domestic subcontracting relationships, thus leading to a negative impact on employment in other domestic sectors (Cappariello, 2010). Such an effect would imply that there would be negative employment effects on an industry of offshoring in other domestic industries. An alternative view (Arndt, 1997) would suggest that the positive productivity effect of offshoring may lead to increased demand for intermediate goods from the domestic economy also, which may offset 2

this indirect ‘substitution effect’. In the analysis that follows we concentrate on the direct effects, leaving the possibility of spillover effects from offshoring on other domestic industries to future work.

Rodrik (1997) argues that labour demand elasticities are an important channel through which an increase in international trade – and offshoring – can affect labour markets. In particular, Rodrik suggests that greater product market competition, due to a decline in trade protection and the entry of less developed countries into the manufacturing sector, should make labour demand more elastic. Similarly, Senses (2010) argues that offshoring, by allowing for the substitution of foreign for domestic labour, is also likely to flatten the labour demand curve. Hijzen and Swaim (2010) identify a number of reasons why this issue is particularly relevant. They argue that to the extent that offshoring increases the labour elasticity of demand it may help explain why workers may feel increasingly insecure, since the wage and employment effects of a shock will be amplified by the higher elasticity of demand. In addition, a higher elasticity of demand will tend to reduce worker’s bargaining power and may limit the scope for risk-sharing arrangements between workers and firms.

A small number of recent studies examine empirically the impact of globalisation on the level of employment using industry-level data, examples including Slaughter (2001), Bruno et al (2004), Molnar et al (2007) and Hijzen and Swaim (2010). Slaughter (2001) initially estimates wage elasticities using data for the US and in a second stage relates these estimates to a large number of economic integration measures – including offshoring measures. Slaughter (2001) finds some evidence that labour demand has become more elastic as integration increased. Bruno et al (2004) concentrate on measures of import penetration for seven OECD countries and find that in the majority of cases there is no significant relationship between import penetration and labour demand. Molnar et al (2007) do something similar to Bruno et al (2004) but use measures 3

of outward FDI rather than import penetration. Their results indicate that labour demand elasticities have increased in response to FDI in manufacturing industries, but declined in services industries. Hijzen and Swaim (2010) concentrate explicitly on offshoring when considering the impact of globalisation on labour demand, examining the relationship between offshoring and industry employment using data on 17 high-income OECD countries for 1995 and 2000. They distinguish between a narrow (i.e. intra-industry) and broad (i.e. inter-industry) measure of offshoring, often finding that the narrow measure impacts negatively upon labour demand, while the broad offshoring measure tends to have no significant impact. They argue that this makes intuitive sense since intra-industry offshoring is more likely to substitute for domestic value added previously performed in that industry.1 When interacting the offshoring measures with wages they find that short-term changes in offshoring have no significant impact on labour demand elasticities, though cross-sectional differences in offshoring do appear to be positively related to differences in the elasticity of demand.

More recently a literature has developed addressing these kinds of issues using micro-level data (i.e. either individual-level or plant-level data), though the majority concentrate on the wage effects of offshoring. An advantage of this approach is that individual and plant-level heterogeneity can be taken account of in the analysis, though data availability and consistency mean that such studies are forced to concentrate on a single country in their analysis. Geichecker and Görg (2008) combine a household panel dataset with industry level information from inputoutput tables to examine the impact of offshoring on wages in Germany for the period 19912000. They find that offshoring has a significant impact upon the wages of low-skilled workers in Germany, with a 1 percent increase in offshoring reducing the wages of low-skilled workers by 1.5 percent. High-skilled workers are found to benefit from offshoring however, with a 1 percent 1

Hijzen and Swaim (2007) find that while intra-industry offshoring reduces the labour-intensity of production, it does not impact on the overall level of employment. In their analysis, inter-industry offshoring does not affect labour-intensity but has a positive effect on overall sectoral employment. Results reported by OECD (2007) find that offshoring lowers the conditional and unconditional demand for labour in OECD countries.

4

increase in offshoring increasing their wages by 2.6 percent. Ebenstein et al (2009) combine information from the US Current Population Survey (CPS) with industry level data on trade and offshoring for the period 1982-2002. They find that within the same industry the impact of import competition or offshoring has a zero or slightly positive impact on wages. What they also show however is that there has been significant employment reallocation across sectors in response to offshoring. Considering the impact of offshoring on wages within occupations they find that offshoring with low wage countries has had a significant and negative impact on wages of workers performing routine tasks. They further show that wage losses are particularly large for workers that move from manufacturing to services industries, and especially for those workers that switch occupations. Liu and Trefler (2008) also use CPS data to examine the impact of offshoring of services to China and India on earnings. They further consider the impact of such offshoring on occupation and industry switching and the risk of unemployment. They find small negative effects or insignificant effects of services offshoring to China and India on switching, earnings and a generally insignificant coefficient on the share of labour-force weeks spent unemployed. Senses (2010) is most related to the hypothesis tested in the current paper and uses US plant level data for the period 1972-2011 to consider the impact of offshoring on the elasticity of demand. He finds that demand elasticities for production workers increase in industries that experience an increase in offshoring over time.

In this paper, we use the recently compiled World Input Output Database (WIOD) to examine whether offshoring impacts upon the elasticity of labour demand in a large sample of countries. The current paper makes a number of contributions to the literature. Firstly, we use the recently compiled World Input-Output Database (WIOD) which reports international Supply and Use tables and international Input-Ouput tables for each year between 1995 and 2009 for 40 countries (plus a rest of the world). This allows us to consider a larger number of countries than has been possible in earlier studies. We further estimate labour demand equations for both total 5

employment and for employment by skill level (i.e. low, medium and high educated labour) separately, allowing us to examine the impact of offshoring on labour demand elasticities for different skill-types. Our results indicate that both narrow and broad offshoring have impacted positively upon labour demand, an effect driven by the scale effect of offshoring. Despite this observed benefit of offshoring, we also show that offshoring has tended to increase labour demand elasticities, which can increase the vulnerability and reduce the bargaining power of workers.

The remainder of the paper is set out as follows: Section 2 describes the econometric approach that we adopt; Section 3 provides information on the data used in the analysis and reports some initial descriptive statistics; Section 4 reports the main results from the analysis; and Section 5 concludes.

2. Methodology The empirical approach that we adopt to consider the impact of offshoring on employment is very similar to the approaches adopted in the above mentioned studies. This involves estimating two models of labour demand – the conditional and unconditional labour-demand models. The major difference in our analysis is that in addition to estimating the model for total employment we also estimate the model for employment by education level.

In the conditional model, the profit-maximising level of labour demand is determined by

minimising the costs of production conditional on output, i.e. industry
’s production costs are a function of factor prices and output. The conditional model of labour demand thus allows one

to assess the technology effect of offshoring by keeping output constant. In a conditional demand function we expect that if offshoring increases productivity, then this will have a negative effect on the demand for labour since fewer inputs are needed to produce the same 6

amount of output. In the unconditional model it is assumed that firms maximize profits, by choosing the optimal mix of input quantities and the level of output for given input and output prices. In the case of labour demand, this corresponds to adjusting hiring so that the marginal value product of labour equals the wage. The unconditional model thus allows one to analyse the total effect of offshoring on labour demand. Hijzen and Swaim (2007) argue therefore that differences in results between the two models thus gives a measure of the scale effect associated of offshoring.

The conditional labour demand equation can be written as:

ln    ∑  ln     ln    ln   ∑  ln 

(1)

where  is industry-level labour demand,  is the nominal price of variable factors (the average

wage and the price of materials),  is the capital stock,  is gross output, and  are demand

shifters, in particular our indicators of offshoring. In our analysis we consider a measure of narrow offshoring as well as a broad measure of offshoring. We further estimate this model for the three different types of labour (low-, medium- and high-skilled), in which case the dependent

variable is industry-level labour demand for a particular labour type and the wage variable is the average wage of that type of labour.

The unconditional (or capital-constrained) labour demand model is given by: ln    ∑  ln     ln   ∑  ln 

(2)

Following Hijzen and Swaim (2010) the output price is excluded from the unconditional model since in imperfectly competitive industries it is considered endogenous as it will be a decreasing function of output. By substituting out the quantity of output this equation allows for scale effects (Hijzen and Swaim, 2007). The net effect of offshoring will then depend upon whether

7

the scale effects are large enough to outweigh the substitution and productivity effects. Once again, this equation will be estimated for total labour and for the three different labour types.

Adding a random disturbance term to the above equations allows us to estimate these models. In the regression analysis that follows we adopt the fairly standard approach of differencing the data to account for time-invariant fixed effects. In particular, we use long (i.e. five-year) differences since these have been shown to be less sensitive to measurement error than either first differences or fixed effects (Griliches and Hausman, 1986). We further include year dummies to capture any time specific heterogeneity, such that our final estimating equations are:

∆ ln    ∑  ∆ ln     ∆ln    ∆ln   ∑  ∆ln  

  

and

(1A)

∆ln    ∑  ∆ln     ∆ln   ∑  ∆ln     

where ∆ indicates the (five-year) difference of a variable.

(2A)

The approach described above involves including offshoring as a demand shifter (i.e. a

component of ), which allows us to examine whether offshoring impacts upon labour demand

but doesn’t allow us to examine its impact upon labour demand elasticities. In the literature a number of approaches have been adopted to address this latter question. Slaughter (2001) undertakes a two-stage procedure obtaining estimates of the wage elasticities in the first stage and relating them to trade and offshoring measures in the second, while Hijzen and Swaim (2010) interact the offshoring measures with the wage variable. In our analysis, we adopt two approaches. Firstly, we follow Hijzen and Swaim (2010) and simply include the interaction of our offshoring measure with the wage variables. Secondly, we employ threshold regression models

8

which allow the coefficient on the wage variables to vary discretely depending upon the value of the offshoring measure. This has a number of advantages over other methods of splitting the data. In particular, it allows the data to determine both the number of splits in the data and their positioning, thereby allowing for a richer pattern of interactions between wages and offshoring.

3. Data and Descriptive Statistics The basic data source for our analysis is the recently completed World-Input-Output-Database (WIOD), which reports data on socio-economic accounts, input-output tables and bilateral trade data across 35 industries and 40 countries over the period 1995-2009.2 These data result from an effort to bring together information from national accounts statistics, supply and use tables, data on trade in goods and services and corresponding data on factors of production (capital and labour by educational attainment categories). The starting point for the WIOD data are national supply and use tables (SUTs) which have been collected, harmonized and standardized for 40 countries (the 27 EU countries, Australia, Brazil, Canada, China, India, Indonesia, Japan, Korea, Mexico, Russia, Taiwan, Turkey and the US) over the period 1995-2009. These tables contain information on the supply and use of 59 products in 35 industries together with information on final use (consumption, investment) by product, value added and gross output by industry. These tables have been benchmarked to time series of national accounts data on value added and gross output to allow for consistency over time and across countries. These tables provide information on supply and use of products by industry for each country. Using detailed trade data the use tables are then split up into domestic and imported sourcing components, with the latter further split by country of origin. Data on goods trade were collected from the UN COMTRADE database at the HS 6-digit level. These detailed bilateral trade data allow one to differentiate imports by use categories (intermediates, consumption and investment goods) by applying a modified categorisation based on broad end-use categories at the product classification. Bilateral 2

See www.wiod.org

9

trade in services data were collected from various sources. Services trade data are only available from Balance of Payments (BoP) statistics providing information at a detailed level only in BoP categories. Using a correspondence these data were merged to the product level data provided in the supply and use tables. The differentiation into use categories of services imports was based on information from existing import use or import input-output tables. Combining this information from the bilateral trade data by product and use categories with the supply and use tables resulted in a set of 40 international use tables for each year. This set of international supply and use tables was then transformed into an international input-output table using standard procedures. A rest-of-world was also estimated using available statistics from the UN and included in this table to account for world trade and production. This results in a world inputoutput database for 41 countries (including the rest-of-world) and 35 industries. Additional data allow for the splitting up of value added into capital and labour income and the latter into low, medium and high educated workers. These data are available both in factor income and physical input terms. The data on low, medium and high educated workers are constructed based upon the 1997 International Standard Classification of Education (ISCED), with ISCED levels 1 and 2 being classified as low educated workers, ISCED levels 3 and 4 being classified as medium educated and ISCED levels 5 and 6 being classified as high educated workers.

In our analysis we make some small departures from the WIOD, and in particular we drop some industries from the analysis. While the offshoring measures defined below are calculated using intermediate inputs from all 35 industries we include only 33 industries in the regression analysis below.3 The industries that are dropped are the service industry P (i.e. private households) and industry 23 (i.e. Coke, Refined Petroleum and Nuclear Fuel) from our analysis. For a number of countries this latter industry shows very low levels of value-added, which often leads to very large values for the offshoring measures. To avoid these outliers affecting our results we drop 3

The 35 industries are listed in Table A1 of the Appendix.

10

this industry from the analysis.4 We also only consider data up to and including 2007, which thus avoids the crisis period from influencing our results.

When measuring offshoring the majority of existing studies focus on some measure of trade in intermediates, though as Hijzen and Swaim (2007) note this ignores the offshoring of assembly activities. In our analysis we use data from international input-output tables, which allow one to measure the intermediate input purchases by each industry and country from each industry and country. In terms of the measures of offshoring Feenstra and Hanson (1999) distinguish between narrow and broad offshoring, where the former considers imported intermediates in a given industry from the same industry only, while the latter considers imported intermediates from all industries. Feenstra and Hanson (1999) prefer the narrow definition as it is thought to be closer to the essence of fragmentation, which necessarily takes place within the industry.5 In our analysis we will consider measures of both narrow and broad offshoring. Following Hijzen

and Swaim (2007) a measure of narrow offshoring (or intra-industry offshoring) for industry
, ! , , can be calculated as:

! , 

"#$%,& '%,&

(3)

where ( , refers to imported intermediate purchases from industry ) 
by industry
in country *, and + refers to value-added.6 Similarly, we define broad offshoring (or inter-industry , offshoring) for industry
, , , as: , , 

∑-#$. "#,& '%,&

(4)

4 As it turns out including this industry (and the excluded private households) doesn’t affect our results qualitatively. These results are available upon request. 5 Hijzen et al (2005) note that this distinction is not without problems, most notably due to the way industries are defined in the data. They consider the example of two industries in which outsourcing is important, namely ‘motor vehicles and parts’ and ‘textiles’, noting that while ‘motor vehicles and parts’ is a single industry in the UK IO table, ‘textiles’ consists of up to ten industries. 6 For purposes of exposition time subscripts, /, are omitted.

11

Figures 1 and 2 report developments in narrow and broad offshoring over the period 1995-2009

by summing up across countries the values of imported intermediates from industry ) 
(i.e. the narrow measure) and from all industries (i.e. the broad measure). When reporting these figures we make a distinction between the development levels of countries based on whether they were classified as high-income (developed countries) or not (developing countries) according to the 1995 World Development Report.7 We do this because we may expect that the effects of offshoring will differ for countries at different levels of development, something that we allow for in the regression analysis below. Figures 1 and 2 indicate that the patterns of both narrow and broad offshoring tended to be fairly similar, showing rapid increases in offshoring until the start of the crisis in 2008 when large drops were observed. Growth rates were particular rapid in the mid-2000s, particularly in the run-up to the crisis in 2008. The figures also show that both narrow and broad offshoring are dominated by the set of developed countries, with the share of developing countries being relatively low. The gap between the two sets of countries is narrowing however. In 1995 the ratio of both narrow and broad offshoring in developed to offshoring in developing countries was above four, whereas in 2007 the ratios had fallen to just above two.

7

The group of countries classified as developed are: Australia, Austria, Belgium, Canada, Cyprus, Denmark, Germany, Finland, France, Ireland, Italy, Japan, Luxembourg, Netherlands, Spain, Sweden, Taiwan, the UK and the USA.

12

Narrow Offshoring (Million US$)

Figure 1: Developments in Narrow Offshoring 3000000 2500000 2000000 1500000 1000000 500000 0

Year Developed Countries

Developing Countries

Broad Offshoring (Million US$)

Figure 2: Developments in Broad Offshoring 8000000 7000000 6000000 5000000 4000000 3000000 2000000 1000000 0

Year Developed Countries

Developing Countries

Figures 3 and 4 report the ratios of narrow and broad offshoring (i.e. equations (3) and (4)) to value added for each country in 1995 and the change between 1995 and 2007, while figures 5 and 6 report the same information but by industry. Figures 3 and 4 indicate that both the narrow and broad measures tend to be relatively large for small open economies (such as Luxembourg, 13

Belgium and Ireland) as well as for a number of Central and East European countries (Estonia, Slovenia, Slovakia, Czech Republic and Hungary). The USA has some of the smallest values of the offshoring measures along with other large developed (France, Germany, Japan) and large developing (India, Brazil, Russia) countries. The value of the offshoring measures tended to increase in most countries between 1995 and 2007, with particularly large increases in the narrow measure observed in Slovakia, the Czech Republic, Hungary and Poland. Similar results are found for the broad measure with Bulgaria and Germany also having large growth rates.

Figures 5 and 6 indicate that both narrow and broad offshoring are heavily concentrated in a small number of industries. These industries include transport equipment (34t35), electrical and optical equipment (30t33), leather and footwear (19), basic metals and fabricated metals (27t28), textiles and textile products (17t18) and chemicals and chemical products (24). There is very little evidence of offshoring in most services sectors, the major exceptions being water (61) and air transport (62). The growth of narrow offshoring over the period 1995-2007 has also tended to be in those industries where narrow offshoring was already relatively intensive (in particular industries 34t35, 30t33, 24, 27t28), as well as machinery nec (29), manufacturing nec, recycling (36t37) and financial intermediation (J). Broad offshoring grew relatively rapidly across a large number of sectors with the growth rates in air transport (62) and electricity, gas and water supplies (E) being particularly pronounced.

14

-0.2 Luxembourg Malta Ireland Belgium Taiwan Slovenia Estonia Slovakia Czech Republic Hungary Netherlands Portugal Korea Canada Sweden Finland Mexico Lithuania Bulgaria UK Latvia China Cyprus Indonesia Austria Denmark Romania Italy France Germany Spain Poland Greece Turkey Russia Australia USA India Japan Brazil

-0.2

Luxembourg Malta Estonia Ireland Belgium Czech Republic Slovakia Hungary Slovenia Lithuania Netherlands Taiwan Latvia Bulgaria Sweden Cyprus Portugal Denmark Finland Romania Canada Austria Korea Mexico UK China Poland Italy Spain Indonesia France Australia Greece Germany Russia Turkey India USA Brazil Japan

Figure 3: Narrow Offshoring by Country 1

0.8

0.6

0.4

0.2

0

Narrow Offshoring in 1995

Broad Offshoring in 1995 Change in Narrow Offshoring 1995-2007

Figure 4: Broad Offshoring by Country

1.4

1.2

1

0.8

0.6

0.4

0.2

0

Change in Broad Offshoring 1995-2007

15

Figure 5: Narrow Offshoring by Industry 0.6 0.5 0.4 0.3 0.2 0.1

-0.1

30t33 34t35 27t28 24 17t18 19 21t22 61 20 29 15t16 25 C 26 36t37 62 AtB 63 71t74 J O 60 E 64 51 F H L 70 50 52 M N

0

Narrow Offshoring in 1995

Change in Narrow Offshoring 1995-2007

Figure 6: Broad Offshoring by Industry 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 34t35 30t33 19 27t28 17t18 24 25 61 29 36t37 15t16 20 21t22 26 62 F E 50 C 63 AtB H 60 N O 64 L 51 71t74 J 52 M 70

0

Broad Offshoring in 1995

Change in Broad Offshoring 1995-2007

Our measure of labour demand is based upon hours worked, and is again available from the (socio-economic accounts of the) WIOD. In the analysis below we use either total hours worked 16

or total hours worked by each of the three skill types. Further variables that we require are measures of the capital stock and gross output, and measures of average wages and the prices of output and intermediate inputs. All of these data are available directly from the WIOD database or can be easily calculated. 8 Table 1 reports information on the average growth rate of employment over the period 1996-2007. The table indicates that while total employment has grown in the vast majority of countries, employment of low-skilled workers has tended to decline (in 27 countries) or not grown by as much as employment of other skill-types (in 11 countries). This is the case for both developed and developing countries with employment falling by 1.51 percent on average for all countries, with larger declines (1.62%) found for developed countries than for developing countries (1.40%). Employment of medium- and high-skilled workers has shown positive growth in the majority of countries however. Negative growth rates of high-skilled employment are never observed, while negative growth rates of medium-skilled employment are found in six countries only. The growth in employment has been higher for high-skilled workers (3.94%) than for medium-skilled workers (2.00%) when considering all countries. Distinguishing between developed and developing countries we observe that the growth of high-skilled employment has been slightly higher for developed countries (3.97% versus 3.91%), while the growth of medium-skilled employment has been higher for developing countries (2.12% versus 1.86%).

8

Indices for the prices of output and intermediate inputs are available directly from the WIOD database, while average wages (for the total labour force and the labour force by skill level) are calculated using data on employment, total compensation to workers, and the share of total compensation to workers by skill-level.

17

Table 1: Average Growth of Employment by Country (in %) Country

Total Employment

Low Educated

Medium Educated

High Educated

Australia

1.79

0.05

2.81

4.27

Austria

0.83

-1.12

0.67

4.24

Belgium

1.15

-3.10

2.80

3.37

Bulgaria

0.21

-0.35

2.05

1.63

Brazil

2.03

-0.65

6.04

5.30

Canada

1.72

-4.29

1.48

3.66

China

1.72

0.74

2.71

8.83

Cyprus

1.76

-0.50

3.04

2.35

Czech Republic

-0.16

-3.36

-0.23

2.09

Germany

-0.14

-0.51

-0.59

1.24

Denmark

1.06

2.04

-0.57

3.11

Spain

3.01

0.05

6.51

6.43

Estonia

0.32

0.48

0.18

0.50

Finland

1.26

-2.28

2.14

2.42

France

0.51

-2.69

1.06

3.48

United Kingdom

0.70

-3.40

1.30

3.78

Greece

1.14

-2.06

3.15

4.54

Hungary

0.20

-3.19

0.36

2.71

Indonesia

0.97

0.23

2.28

6.36

India

1.61

0.57

3.18

5.73

Ireland

3.68

0.35

3.66

7.69

Italy

0.92

-1.79

2.89

5.36

Japan

-0.72

-6.87

-0.64

2.08

Korea

0.38

-6.62

-0.45

4.06

Lithuania

0.92

-0.96

0.52

2.31

Luxembourg

3.94

0.64

6.00

6.36

Latvia

6.80

6.79

6.81

6.81

Mexico

2.63

0.15

5.41

2.20

Malta

0.81

-0.02

2.50

2.83

Netherlands

1.16

-0.92

0.62

4.65

Poland

0.20

-4.89

-0.11

5.50

Portugal

1.15

0.23

3.41

4.31

Romania

0.33

-0.30

2.79

2.96

Russia

0.17

-3.79

0.35

1.79

Slovakia

-0.14

-6.59

0.01

1.88

Slovenia

0.16

-3.46

0.11

4.07

Sweden

0.64

-3.48

0.54

4.31

Turkey

-0.63

-2.46

3.51

5.60

Taiwan

0.12

-3.10

1.06

4.29

USA

1.05

0.07

0.55

2.39

All Countries

1.13

-1.51

2.00

3.94

Developed Countries

1.29

-1.62

1.86

3.97

Developing Countries

0.99

-1.40

2.12

3.91

18

Before proceeding to the regression analysis we begin by reporting information on developments in estimated labour demand elasticities. To do this we estimate the conditional labour demand equation given by equation (1A) for each year in our dataset. Given that we are estimating the model in fifth differences and given that we only use data for the period 1995-2007 in the analysis below we can report estimated elasticities for the period 2000-2007 only. While this is a relatively short period of time it will provide some information on the development of demand elasticities over time. We report in Table 2 estimated elasticities (in absolute values) for total employment and for employment by skill level. We further report results separately for the developed and developing country sub-samples to examine whether elasticities have developed different in these two groups of countries. The table indicates that the elasticity of labour demand showed a tendency to increase during the 2000s, both for total employment and the different skill-types. Splitting the sample into developed and developing countries we find that elasticities tended to increase to a greater extent in developed countries, with the percentage change between 2000 and 2007 being 58 percent in the case of developed countries and 30 percent in the case of developing countries. By skill-level the results in Table 2 indicate that elasticities increased particularly strongly for low-skilled employment in developed countries (an increase of 122 percent when comparing 2007 with 2000), while in the case of developing countries elasticities increased more strongly for medium- and high-skilled employment. Overall, the table indicates that elasticities of labour demand did show a tendency to increase during the 2000s, but that there was a great deal of heterogeneity across skill-types and across countries in the extent of these changes. In the regression analysis below this heterogeneity is allowed for by splitting up our sample into a developed and developing country sample, and by considering the three different labour skill types.

19

Table 2: Estimated Elasticities from the Conditional Labour Demand Model Total Employment

Low Educated

Medium Educated

High Educated

ALL COUNTRIES 2000 2001 2002 2003 2004 2005 2006 2007

0.386 0.425 0.459 0.385 0.436 0.432 0.498 0.483

0.329 0.402 0.439 0.382 0.394 0.391 0.418 0.375

0.411 0.439 0.473 0.406 0.401 0.399 0.498 0.065

0.432 0.430 0.466 0.330 0.449 0.411 0.477 0.477

DEVELOPED COUNTRIES 2000 2001 2002 2003 2004 2005 2006 2007

0.420 0.385 0.540 0.523 0.595 0.684 0.693 0.663

0.196 0.349 0.477 0.131 0.457 0.499 0.476 0.437

0.454 0.402 0.595 0.660 0.506 0.510 0.585 0.528

0.567 0.339 0.532 0.454 0.642 0.633 0.637 0.733

DEVELOPING COUNTRIES 2000 0.396 0.367 0.423 0.410 2001 0.448 0.426 0.464 0.454 2002 0.480 0.460 0.493 0.479 2003 0.403 0.454 0.401 0.331 2004 0.448 0.405 0.408 0.460 2005 0.444 0.393 0.412 0.428 2006 0.516 0.427 0.514 0.505 2007 0.516 0.397 0.053 0.509 Notes: The table reports the estimated elasticity of labour demand for each year between 2000 and 2007 for all countries, for developed countries only and for developing countries only. Estimated elasticities are based on the conditional labour demand model (Equation 1A). Elasticities by skill level are estimates of the elasticity of labour demand by skill-level with respect to wages of that skill-level.

4. Results The results section is split into three parts. In the first sub-section we report results from estimating the conditional and unconditional labour demand models when including the two measures of offshoring linearly. These results thus allow us to address whether offshoring impacts upon labour demand in our sample of countries. In the second sub-section we report results when including interactions between offshoring and the wage variables and when estimating the threshold regression model, both of which allow us to examine whether offshoring has impacted upon the elasticity of labour demand. In the third sub-section we report results from an instrumental variables approach which helps us deal with potential endogeneity

20

issues that may arise due to the lack of measures on skill-biased technological change (SBTC). In all of the sub-sections we report results for the full sample of countries and separately for the sample of developed and developing countries. This approach allows for heterogeneity in the impact of offshoring across these two subsamples therefore. To avoid the crisis impacting upon our results we also only consider the years 1995-2007 in the regression analysis. As mentioned above, the regressions are also estimated using long differences, which removes sector-country fixed effects and helps deal with measurement error.

4.1. Offshoring and Labour Demand Tables 3 and 4 report results from the conditional and unconditional labour demand equations using the narrow and broad definitions of offshoring, with Table 3 reporting results when using the narrow definition and Table 4 those when using the broad definition. The first four columns of Table 3 (and 4) reports results from the conditional model for total employment and the three different skill levels, while the latter four report similar results from the unconditional model. These tables and all others below report three sets of results: the first panel reports results for the full sample of countries; the second panel reports results for the subsample of developed countries; and the third panel reports results for the subsample of developing countries.

Concentrating initially on results in Table 3 we observe that the wage elasticities are found to be negative and significant for all labour types and for both the full sample of countries and the two subsamples of developed and developing countries. The coefficients tend to be larger for highskilled workers than for the other two skill types for all countries and for the two country subsamples. Elasticities also tend to be larger for low-skilled than for medium-skilled labour in the full sample of countries and for developing countries, with the reverse being the case in the case of developed countries. Coefficients on the price of intermediates are found to be consistently positive and significant, suggesting that intermediates and labour are substitutes, as 21

are the coefficients on the measure of gross output in the conditional model. Coefficients on the capital stock are only found to be consistently significant in the unconditional model, where they are always positive and significant. In the conditional model coefficients are usually insignificant, but become negative and significant when considering developed countries only.

Turning to the coefficients on the narrow offshoring measure we observe in the conditional model that the coefficients on the offshoring measure are positive and significant in the case of low- and high-skilled labour when looking at the full sample of countries and are insignificant for total employment and medium-skilled labour. In the unconditional model the coefficients are positive and significant across the different skill levels, with elasticities being relatively low (0.93%) for medium-skilled labour and relatively high for high-skilled labour (2.84%). In the developed country subsample we observe an insignificant coefficient on narrow offshoring for total employment (and for low- and high-skilled labour), with a significant negative effect found for medium-skilled labour. Coefficients are found to be consistently positive and significant in the unconditional model however, with the effect being relatively small for medium-skilled labour and relatively large for low-skilled labour. In the developing country sample we again observe generally insignificant coefficients on offshoring in the conditional model, with the effect being positive and significant for high-skilled labour. In the unconditional model, coefficients tend to be positive and significant, being again relatively low for medium-skilled labour and relatively high for high-skilled labour. In general, coefficients on the offshoring measure tend to be smaller in the developing country sample than in the developed country sample.

Results when considering the broader measure of offshoring (Table 4) show similarities and differences to those found when using the narrow measure. Wage elasticities and coefficients on intermediate prices, gross output and the capital stock are largely similar to those in Table 3, as 22

are the coefficients on offshoring when considering all countries. In the case of developed countries however, we find in the conditional model a significantly negative coefficient on offshoring, which appears to be driven by negative impacts of offshoring on the demand for medium- and high-skilled labour (with no significant effect found for low-skilled labour). In the unconditional model coefficients are again found to be consistently positive and significant, being relatively large for low-skilled and high-skilled labour. The coefficients on offshoring in the developing country subsamples are similar in the conditional and unconditional models and indicate that there is a significantly positive coefficient on offshoring, which is driven by a positive and significant coefficient on labour demand for high-skilled labour (with insignificant coefficients found for low- and medium-skilled labour).

23

Table 3: Labour Demand and Narrow Offshoring (1) 01_3

ALL COUNTRIES -0.397*** ∆ (0.0168) 0.357*** ∆66 (0.0195) 0.00258 ∆7 (0.00578) 0.371*** ∆8( (0.0106) 0.00563 ∆! (0.00399)

(2) 01_4

(3) 01_4

(4) 01_54

(5) 01_3

(6) 01_4

(7) 01_4

(8) 01_54

-0.354*** (0.0167) 0.359*** (0.0201) 0.00371 (0.00669) 0.369*** (0.0127) 0.0102** (0.00513)

-0.291*** (0.0450) 0.293*** (0.0410) 0.000837 (0.00637) 0.331*** (0.0152) -0.000463 (0.00469)

-0.407*** (0.0174) 0.338*** (0.0197) 0.00439 (0.00740) 0.373*** (0.0127) 0.0182*** (0.00499)

-0.332*** (0.0170) 0.244*** (0.0188) 0.0918*** (0.00663)

-0.295*** (0.0169) 0.251*** (0.0193) 0.0925*** (0.00725)

-0.249*** (0.0380) 0.206*** (0.0339) 0.0814*** (0.00719)

-0.346*** (0.0175) 0.230*** (0.0192) 0.0936*** (0.00799)

0.0164*** (0.00445)

0.0212*** (0.00547)

0.00926* (0.00494)

0.0289*** (0.00535)

Observations R-squared F-Test

9,854 0.227 107.7***

9,854 0.254 78.55***

9,854 0.247 105.6***

9,854 0.199 47.13***

9,854 0.124 45.62***

9,854 0.142 22.97***

9,854 0.146 48.88***

-0.364*** (0.0377) 0.139*** (0.0278) -0.0288*** (0.00783) 0.544*** (0.0207) 0.0107 (0.00765)

-0.507*** (0.0287) 0.149*** (0.0230) -0.0367*** (0.00788) 0.550*** (0.0171) -0.0102** (0.00457)

-0.556*** (0.0262) 0.0914*** (0.0293) -0.0541*** (0.00966) 0.578*** (0.0195) 0.00304 (0.00598)

-0.272*** (0.0323) 0.265*** (0.0302) 0.0879*** (0.00861)

-0.174*** (0.0383) 0.283*** (0.0382) 0.0889*** (0.00996)

-0.282*** (0.0340) 0.284*** (0.0332) 0.0835*** (0.00935)

-0.368*** (0.0319) 0.248*** (0.0349) 0.0686*** (0.0102)

0.0306*** (0.00494)

0.0426*** (0.00841)

0.0211*** (0.00555)

0.0344*** (0.00696)

4,795 0.241 81.53***

4,795 0.372 119.3***

4,795 0.288 117.2***

4,795 0.173 34.49***

4,795 0.082 22.56***

4,795 0.138 34.26***

4,795 0.106 29.66***

DEVELOPING COUNTRIES -0.415*** ∆ (0.0201) 0.360*** ∆66 (0.0236) 0.00656 ∆7 (0.00823) 0.313*** ∆8( (0.0147) 0.00386 ∆! (0.00536)

-0.386*** (0.0199) 0.361*** (0.0240) 0.00998 (0.00940) 0.310*** (0.0172) 0.00623 (0.00649)

-0.283*** (0.0509) 0.271*** (0.0468) 0.00432 (0.00882) 0.252*** (0.0205) -0.00170 (0.00626)

-0.410*** (0.0211) 0.333*** (0.0240) 0.0193* (0.0104) 0.277*** (0.0175) 0.0206*** (0.00644)

-0.366*** (0.0202) 0.246*** (0.0214) 0.0773*** (0.00880)

-0.340*** (0.0199) 0.252*** (0.0217) 0.0801*** (0.00956)

-0.254*** (0.0450) 0.188*** (0.0388) 0.0617*** (0.00946)

-0.367*** (0.0208) 0.234*** (0.0218) 0.0812*** (0.0107)

0.00988* (0.00578)

0.0124* (0.00679)

0.00326 (0.00650)

0.0258*** (0.00675)

Observations R-squared F-Test

5,059 0.234 55.39***

5,059 5,059 5,059 0.214 0.264 0.234 21.43*** 68.16*** 35.38*** Robust standard errors in parentheses *** p