The China Shock: Examining the effects of Chinese import competition on the labor share of developed countries

(1)

The China Shock: Examining the effects of Chinese import competition on the

labor share of developed countries

Master Thesis

Master International Economics and Business

Faculty of Economics and Business

University of Groningen

Marieke van Beek

S2768577

m.g.j.van.beek@student.rug.nl

Supervisor: Prof. Dr. R.C. Inklaar

Co-assessor: Dr. M.A. Papakonstantinou

Final version

(2)

2 Abstract

This study aims to investigate the effect of globalization on the labor share of national output. In order to examine the effect of globalization, a new measure of import exposure is composed for a set 15 high-developed countries on an industry level between 1995 and 2007. This measure reflects the change in the level of Chinese import competition for domestic industries in high-developed countries after China joined the World Trade Organization (WTO) in 2001. The results indicate that there is no evidence of an effect between the change of import exposure and the change of the labor share as a result of the China shock. This implies that this relationship cannot be generalized for a wider set of high-developed countries included in this analysis. In addition, after performing a sensitivity analysis and several extensions, the effect remains insignificant. However, this study still shows that most increased exposure is observed in labor-intensive industries, which confirms that offshoring of activities from the developed world over the last two decades happened mainly for labor-intensive manufacturing production.

(3)

3 Acknowledgements

(4)

4 Contents

1. Introduction ... 6

2. Literature review ... 9

2.1 The decline of the labor share ... 9

2.2 Globalization and the labor share ... 11

2.2.1 Theoretical foundations ... 12

2.2.2 Empirical studies ... 13

2.3 Hypotheses ... 16

3. Methodology and data ... 16

3.1 Model specification ... 17 3.2 Data ... 18 3.2.1 Labor share ... 19 3.2.2 Import exposure ... 20 4. Empirical analysis ... 22 4.1 First steps ... 22 4.2 Descriptive statistics ... 23 4.3 Results ... 24 4.4 Sensitivity analysis ... 26

4.5 Extension empirical analysis ... 27

4.6 Comparing the results to preceding work ... 30

4.7 Other remarks ... 35

5. Summary and conclusion ... 37

References ... 41

Appendix A ... 45

Appendix B ... 47

(5)

5 List of tables

Table 1: Summary statistics ... 24

Table 2: Ordinary Least Square (OLS) regression (full sample) ... 25

Table 3: Ordinary Least Square (OLS) Regression (excluding industries without exposure) . 26 Table 4: Ordinary Least Square (OLS) regression (only manufacturing) ... 27

Table 5: Ordinary Least Square (OLS) regression (t0) ... 28

Table 6: Ordinary Least Square (OLS) regression with data from India and Indonesia ... 29

Table 7: Ordinary Least Square (OLS) regression (percentage change)... 31

Table 8: Ordinary Least Square (OLS) regression (only United States) ... 32

Table 9: Instrumental variable (2SLS) regression (only for the United States) ... 34

Table 10: Overview of industries included in this analysis ... 45

Table 11: Overview of manufacturing industries included in this analysis ... 46

Table 12: Overview of countries included in this analysis ... 46

(6)

6 1. Introduction

The question whether globalization has led to increased income inequality received a lot of attention by politicians and societies throughout the world, especially after the Brexit vote, the election of president Trump, and the rise of populism in Europe. More recently, the debate about globalization again has intensified after president Trump announced to introduce protection measures to offset the large US trade deficit with China (Amadeo, 2018). By definition, globalization deals with the importance of international transactions in the world’s economic activity and includes international trade in goods, financial assets and migration flows (Helpman, 2016). Over the recent years, it has been argued by several scholars that emerging Asian economies together with the top one percent of the global income distribution significantly gained from globalization, while the middle income class of rich nations experienced almost no growth in income (Milanovic, 2016; Alvaredo et al., 2017; Maskin, 2015). In more detail, the opening up of emerging economies led to a new wave of globalization that is typified by increased international trade in tasks and activities. This resulted in offshoring of activities all over the world and countries increasingly participating in global value chains (GVC’s) (Milberg and Winkler, 2010). Developing countries made use of their cost advantage in producing labor-intensive manufacturing goods. Subsequently, this caused a huge increase in imports from the developing world without offsetting an increase in demand for exports of developed countries. This trade imbalance together with limited labor mobility led to heterogeneous consequences for different groups of the income distribution (Autor et al., 2016).

(7)

7

reflect a too limited view on the current changes caused by international integration in the economic environment (Andersen and Herbertsson, 2003). In addition, it is difficult to control for endogeneity since openness depends on a wide range of other facts that might have an influence (Llloyd and Maclaren, 2002). An alternative measure of capturing globalization is introduced by Elsby et al. (2013) who use import exposure as a proxy and investigate whether the change in exposure can help explain the current changes in the labor share. This measure allows them to see how much additional value added would have been generated if US industries produced all imported goods domestically. They find a negative relationship between their measures of import exposure and the labor share. However, Elsby et al. (2013) only consider import exposure of industries in the United States. This makes it debatable whether the international trend of the declining labor share can be attributed to the effect of globalization. Furthermore, the authors consider all goods the US imports and this makes it difficult to argue that the changes in imports are a result of offshoring of labor-intensive production to the developing world. Finally, the authors admit themselves that their estimation is subject to several measurement issues which they do not further specify. As a result, the article of Elsby et al. (2013) fits the current definition of globalization in terms of increased trade in activities and offshoring, but the above mentioned issues indicate that there exists room for improvement.

In order to make the effect of globalization more distinct and to deal with the above mentioned shortcomings and endogeneity issues, a different approach is introduced. Several scholars have examined the effect of increased import competition on labor market outcomes by considering the rise of China as a shock to the US economy which allows for dealing with endogeneity (Autor et al., 2013; Balsvik et al., 2015; Bloom et al., 2016). They use the China shock as explanatory factor for changes in trade, production structures and thereby also the level of import exposure for domestic US industries.

(8)

8

if they would have produced these goods themselves together with the level of competition with Chinese industries. Chinese imports are used since this country is considered to be the ‘factory of the world’ and holds a comparative advantage in producing labor-intensive products. Furthermore, to reflect the change in import exposure, China’s membership to the World Trade Organization in 2001 is considered as a shock. This allows for a comparison of the level of imports before and after their membership and to deal with endogeneity. This measure is composed by using the World Input-Output Database Release 2013 (WIOD) which allows for capturing both imports of intermediate inputs and final goods. Moreover, the WIOD is also used for calculating the labor share and captures the share of labor compensation divided by value added. This analysis includes fifteen high-income countries and their respective 35 industries and investigates the total time period of 1995-2007.

After having performed the empirical analysis, the findings of this study show that there is no evidence for an effect between the change in import exposure and the labor share for a larger set of high-income countries. The results remain the same after performing a sensitivity analysis and several empirical extensions. This could be explained, amongst other factors, by heterogeneity between countries and industries, as already shown in their level of exposure. However, the data on import exposure does show that, on average, the industries (textiles, leather and electrical equipment) that experience the largest increase in exposure are labor-intensive manufacturing. This is in line with what one expects of trade with China and is also confirmed by other studies. Nevertheless, at this point, there is no effect found between import exposure and the labor share in this study and can therefore not be generalized for a wider set of developed countries.

(9)

9

exposure on labor market outcomes as is done in the early work of Borjas and Ramey (1995) and subsequently in Autor et al. (2013) and (2016).

The rest of the paper proceeds as follows: first, it provides a comprehensive overview of the relevant literature that deals with the labor share and globalization. Subsequently, the paper discusses in depth the methodology and data used for its empirical analysis. The next section performs its analysis and presents the results of the relationship between globalization and the labor share. Subsequently, this section includes a sensitivity analysis together with several empirical extensions. Finally, the paper concludes and provides direction for future research.

2. Literature review

2.1 The decline of the labor share

Since Kaldor (1961) presented one of his famous “stylized facts” of growth, the constancy of the labor share has been an important assumption in macroeconomic studies. It implies that over time, the share of capital and labor to total income are both driven by constant parameters. This fact fostered the widespread assumption in several macro-economic models that use a Cobb-Douglas production function that also assumes constancy of factor shares (Elsby et al. 2013; Harrison, 2005). However, this stylized fact is currently being questioned after observing a decreasing trend in the labor share (Dorn et al., 2017; Doan and Wan, 2017; Elsby et al., 2013; Kristal, 2010). The labor share represents “the ratio of total labor compensation over national output and includes wages, salaries and employment benefits” (Doan and Wan, 2017, p.4). This definition implies that over time, there has been a decrease in the share of output going to labor and results in Kaldor’s fact no longer able to explain the current trend. This worldwide trend has therefore left many economists puzzled and several of them have proposed potential factors that can explain this current trend.

(10)

10

expensive, driving firms to substitute away from labor towards capital. These authors conclude that the decrease in the relative price of investments can explain around half of the decline of the global labor share. However, this high elasticity of substitution is not confirmed by others and most of the studies suggest that it is smaller than one (Dorn et al., 2017). Furthermore, in line with investigating the effect of technological change, Reijnders et al. (2016) examine the composition of output of final products in value added, at different stages of the production process, at home or abroad. They use a vertically integrated production process also referred to as a global value chain (GVC) in which countries from all over the world engage and perform tasks and activities. The authors conclude that there exists a strong bias of technological change in favor of capital for developed countries which explains the decreasing share of labor in GVCs for these countries. Additionally, next to the capital bias, these authors highlight that technological change is biased in favor of workers with a college education. Subsequently, they find that these biases and reallocation of tasks had a negative effect on employment for low-educated workers in advanced nations. Finally, Aghion et al. (2017) propose a task-based view and consider the role of artificial intelligence (AI) as a new force of increased automation influencing factor payments. They argue that as a result of improvements in AI, a shift takes place from tasks performed by labor to capital, which subsequently results in a decrease of the labor share.

(11)

11

Finally, globalization has been proposed as an important factor. Several studies argue that increased international trade and investments over the past three decades has led to a decline in the labor share (Rodrik, 1997; Guscina, 2006; Stockhammer, 2017; Doan & Wan, 2017). As a result of opening up, countries become specialized, which subsequently caused production to take place all over the world. Specifically, Elsby et al. (2013) argue that globalization has led to vertical specialization of countries, which subsequently generated increased offshoring of labor- intensive activities from advanced to developing nations while capital-intensive activities stayed behind. The departure of labor-intensive activities subsequently can help explain the current downward trend in the labor share for the United States.

2.2 Globalization and the labor share

(12)

12

with the labor market frictions help explain why labor market outcomes for workers in the low-middle income class within developed countries worsened and how this has fostered increased income inequality.

The changes in the economic environment, the increased level of income inequality and the alignment in trends all together provide interest to further investigate the relationship between globalization and the labor share. As aforementioned, several studies have examined this relationship before and have used different approaches to capture this relation. Before considering these empirical studies, it is first important to provide a theoretical foundation that can help explain how globalization influences the labor share.

2.2.1 Theoretical foundations

Firstly, Heckscher Ohlin’s theory of international trade and the Stolper Samuelson theorem, predict that countries that engage in international trade specialize in the factor input, either labor or capital, for which they hold a comparative advantage. Countries will export this good and due to increased demand, prices will increase leading to increased returns for this factor input. The opposite happens to the factor input that is not relatively abundant and will experience decreasing returns due to specialization. In theory, it is assumed that developed countries will export goods that are capital-intensive while developing countries specialize in labor-intensive goods. This implies increasing returns for capital in developed nations, which subsequently squeezes the share of GDP going to labor while the adverse should hold for developing countries. However, trade theory cannot explain why developing countries also experienced a decline in their labor share while the opposite should have occurred (Stockhammer, 2017).

(13)

13

demand for high-skilled workers in developing nations. A comparable analysis can be made with regards to the labor share. Irrespectively of the skill level, for developed nations, labor-intensive activities have been offshored to places that hold a cost advantage. Consequently, domestic labor is now being substituted for foreign labor, which subsequently leads to a decline in domestic labor demand together with a decrease in real wage and the respective share of GDP going to labor (Abdih and Danninger, 2017). This is also found by Reijnders et al. (2016), namely that for most developed countries offshoring is associated with increased returns to capital and high-skilled labor.

Finally, in reaction to Hecksher-Ohlin, an alternative theoretical approach has been introduced namely the bargaining framework (Rodrik, 1997). The benefits from trade are now determined through the bargaining position of factor inputs instead of the relative prices. The factor that holds larger bargaining power will have higher returns and this power is determined by the fixed costs of offshoring together with the return to factor inputs abroad and at home (Harrison, 2005). In addition, mobility of factor inputs plays a large role in determining the rents of labor and capital. Rodrik (1997) states that the factor that is more mobile will reap most benefits from reallocation. It is expected that since capital is more mobile, it can relocate to places with higher returns which provides capital with a strong bargaining position. At the same time, the reductions in barriers make it easier to substitute domestic labor, which consequently decreases the bargaining position of workers not only in developed, but also developing countries. This can subsequently help explaining the declining labor share for both advanced and developing nations since capital has substantially benefitted more from globalization.

2.2.2 Empirical studies

(14)

14

estimate the amount of workers in this sector (Gindling and Newhouse, 2014). Finally, there exists increased overlap between labor and capital income, especially at the top of the income distribution (Alvaredo et al., 2013). In fact, high income earners have more investment opportunities to accumulate wealth and high business executives are increasingly paid in capital. As a result of these measurement issues, several adaptations have been introduced which mainly focus on the mixed self-employment income. Several studies split the mixed income between capital and labor based on shares for income that were not mixed (Gollin, 2002; Feenstra et al., 2015). Moreover, other studies use alternative measures of the labor share such as the corporate labor share or payroll share in order to avoid these potential measurement problems (Karabarbounis and Neiman. 2013; Elsby et al., 2013).

To capture globalization, a wide range of different approaches exists. Harrison (2005) is one of the first that considers globalization as an important explanatory variable of the declining labor share and measures globalization by four different ways: trade shares, exchange rates, movement in foreign investments, and capital controls. Yanikkaya (2003) discusses the differences between measurements of trade volumes and trade restrictions and their subsequent different effect on measures of growth. Moreover, he provides an overview of different measures of globalization and highlights the commonly used approach to capture this is by considering the share of exports and imports of total GDP. This approach is often also referred to as openness and is a commonly used by other scholars for both advanced and developing countries (Rodrik, 1997; Guscina, 2006; Stockhammer, 2017; Doan & Wan, 2017).

(15)

15

Herbertsson, 2003). These concerns make it debatable whether openness is a satisfying indicator of the current wave of globalization.

An alternative measure of capturing globalization is introduced by Elsby et al. (2013) who use import exposure as a proxy. They highlight that the United States has experienced a large increase in imports over the past decades. This can be explained by vertical specialization, which refers to countries specializing in activities instead of final products (Hummels et al., 2001). For countries like the United States, this implies specialization in capital-intensive activities while the relatively costly labor-intensive goods are offshored to countries with cheaper labor costs (Elsby et al., 2013). In order to identify the effect of globalization, the authors use a measure of import exposure, which allows them to see how much value added would have been generated if industries in the United States would produce all imported goods domestically.

The effect of increased import competition for the US as a result of China imports has been examined before by several authors (Autor et al., 2013; Balsvik et al., 2015; Bloom et al., 2016). In more detail, Bloom et al. (2016) investigate the effect of increased import competition on measures of technological change while Autor et al. (2013) and Balsvik et al. (2015) look at the effect on labor market outcomes. These authors treat import competition as the value of annual US imports coming from China and consider this as a shock to the economy of the United States in order to deal with endogeneity. This is because China has been able to make use of its comparative advantage in labor-intensive activities (manufacturing goods) since it opened up to trade, which has led to the country becoming the ‘factory of the world’. Autor et al. (2013) and Balsvik et al. (2015) find that Chinese imports have increased and fostered worsened labor market outcomes for US manufacturing industries directly competing with China. In addition, Elsby et al. (2013) find that, by using data of the Bureau of Labor Statistics (BLS), import exposure increased for almost all industries over time and that the ones that were exposed also experienced the biggest decline in the labor share. Overall, the authors conclude that their results indicate that import exposure of US industries for a large part explain the decline in the US payroll share over the past twenty five years.

(16)

16

of Elsby et al. (2013) what is exactly driving the changes in import exposure since they consider all goods the US is importing which leads to an endogeneity problem. In contrast, Autor et al. (2013) use increased import competition with China as a shock and works as an explanatory factor for the changes in production structures of developed nations. Finally, the authors highlight that the regression of their proxy of globalization on the labor share is subject to several issues which they do not further specify.

2.3 Hypotheses

The literature review stresses the potentially important role of globalization explaining in the current decline in the labor share. Moreover, it shows that several authors have already examined this relation before and almost all have found a negative effect of globalization on the declining labor share. However, these previous studies are subject to several shortcomings that mainly deal with measurement issues and often consider only one or a small sample of countries. Therefore, this research aims to further investigate the relationship between globalization and the labor share. The main goals are to deal with these measurement issues and to use an adequate measure of globalization that captures the recent changes in the global economy and production structures as a result of increased trade in intermediates and activities. In addition, there exists also ambition to consider a wider set of countries in order to see whether this relationship can be generalized for developed nations.

Based on the recent literature, it is expected that the share of GDP going to labor has declined over the past decades. In addition, since globalization and the trade in activities have intensified especially after China becoming part of the World Trade Organization (WTO) in 2001, it is assumed that the import exposure, in terms of increased imports from China, has raised over the years. Finally, based on the offshoring theory and the empirical studies, it is expected that globalization has a negative effect on the labor share. This is because labor-intensive activities have been offshored to developing world, which resulted in a decline in the labor share of developed countries.

3. Methodology and data

(17)

17

offshoring of activities and tasks all over the world. They assess by using annual input-output matrices the percentage increase in value added needed to satisfy US final demand when the US would produce all imported goods domestically. This shows the additional value added the US could have gained if it would not import and instead produce domestically.

However, the issue with this approach is that it considers all goods the United States imports from all over the world and includes a broad range of factors could have influenced the changes in imports. This makes it more difficult to identify what is driving the changes in imports and whether it is a result of offshoring of manufacturing production to the developing world. In addition, it could also be the case that others factors like unobserved demand shocks also played a role which leads to a subsequent endogeneity issue. Therefore, in order to make the effect of globalization more distinct, the China shock approach is used by Autor et al. (2013), Acemoglu et al. (2016) and Bloom et al. (2016). These authors argue that, because China has experienced rapid increases in productivity growth and its membership to the WTO have contributed to China making us of its comparative advantage and specializing in labor-intensive manufacturing good. This has led to the distinction of China becoming the ‘factory of the world’ and resulted in a major increase in import competition for US industries with China.

Including the China shock helps to explain the change in production structures of developed countries as a result of increased offshoring of activities to countries that hold a comparative advantage producing labor-intensive goods. In addition, this approach deals with an endogeneity problem since one can state that the changes in imports are a result of China’s productivity gains and their membership to the WTO. This empirical investigation considers the change in both the labor share and import exposure in the periods prior and after China joined the WTO in 2001. Furthermore, Autor et al. (2013) show that other high-income countries also experienced a high level of import competition with China. These countries are a suitable sample of countries in order to investigate whether the findings of Elsby et al. (2013) for the US can be generalized for a wider set of countries. These are high-income countries and include Australia, Denmark, Finland, Germany, Japan, New Zealand, Spain and Switzerland.

3.1 Model specification

The baseline model of this research is the following:

(18)

18

For which the Δ labor shareij represents the change of the total labor compensation over value added in country i for sector j, Δ import exposure represents the change of Chinese imports from industry j divided by gross output of domestic industry j in developed country i, µ_𝑖 and µ_𝑗 include country and industry dummies, 𝜀_𝑖𝑗 is the error term including white noise.

This baseline model is an Ordinary Least Square (OLS) regression model and includes 15 countries that consist out of 35 industries each with observations for the time period of 1995-2007. In addition, country and industries dummies are included in order to control for unobserved heterogeneity and individual characteristics of countries and industries. For example, there could be countries and industries that are more likely to be exposed to import competition or certain industries that are more likely to be offshored already have a lower labor share going to GDP (Doan and Wan, 2017).

3.2 Data

The first main data source is the World Input-Output Database (WIOD) 2013 Release. The WIOD includes a series of several databases for 27 EU countries and 13 other major countries for a period of 1995-2011, which leads to a total of 40 countries and for every country 35 industries. Subsequently, the WIOD includes World Input-Output Tables (WIOT), Socio Economic Accounts (SEA) and Environmental Accounts. The WIOD allows for investigating the consequences of increased fragmentation and changes in production structures that have occurred over the past decades since it considers both intermediate and final goods for a time period of 1995-2011.

(19)

19

of the WIOD, several additional high income countries have been added to the analysis in order to create a larger sample of countries (see Appendix A for further specification).

3.2.1 Labor share

The labor share represents the total labor compensation over value added as expressed in the equation 2 below:

𝐿𝑎𝑏𝑜𝑟 𝑠ℎ𝑎𝑟𝑒𝑖𝑗𝑡 =

𝐿𝑎𝑏𝑜𝑟 𝑐𝑜𝑚𝑝𝑒𝑛𝑠𝑎𝑡𝑖𝑜𝑛 𝑖𝑗𝑡

𝑉𝑎𝑙𝑢𝑒 𝑎𝑑𝑑𝑒𝑑𝑖𝑗𝑡 (2)

In this case, the labor compensation is derived from the Socio Economic Accounts (SEA) of the WIOD Release 2013 and is defined by labor compensation in millions of national currency (LAB). The SEA provides the labor compensation per country per industry for the necessary years 1995-2007. In order to deal with labor compensation of the self-employed, the SEA assumes, for advanced nations, that the compensation per hour is equal to the compensation of employees. The data source relies on additional information per country in order to derive at the numbers of the self-employed workers. Subsequently, this income of the self-employed is added to the labor compensation. Next, to arrive at the labor shares it is necessary to divide the labor compensation by value added, which is also provided by the SEA and is expressed in millions of national currency. The reason to divide by value added instead of gross output is because gross output is subsequently influenced by a broad range of other factors. These steps are performed for the total of 15 industries and their respective 35 industries for the considered time period of 1995-2007.

Subsequently, in order to derive the change of the labor share in the period of 1995-2007 several additional steps have to be made. As aforementioned, the ‘China shock’ idea by Autor et al. (2013) is included in this analysis in order to make the effect of globalization more distinct. An average labor share, for all countries and their industries, is composed for the first six years that China was not a member of the WTO (1995-2000) and for the post period including seven respective years (2001-2007). To obtain the change of the labor share since China has been a member of the WTO, these averages are subtracted and displayed in equation 3 below: ∆ Labor shareij = [ 1 6∑ 𝐿𝑎𝑏𝑜𝑟 𝑠ℎ𝑎𝑟𝑒𝑖𝑗𝑡 2000 1995 ] − [ 1 7∑ 𝐿𝑎𝑏𝑜𝑟 𝑠ℎ𝑎𝑟𝑒𝑖𝑗𝑡 2007 2001 ] (3)

(20)

20

allows for dealing with endogeneity problems that often turn out to be an issue in other studies that investigate the relationship between globalization and the labor share. As a result, the above introduced approach uses the China shock as a way to make changes in the labor share attributable to China joining the WTO in 2001.

3.2.2 Import exposure

The main explanatory variable in this analysis is import exposure, which is a proxy of globalization and captures the change in import competition for developed domestic industries as a result of the China shock. The main reason to consider imports from China is because it has a comparative advantage in labor-intensive goods and allows for dealing with endogeneity. Finally, it also provides information which domestic industries experience most competition with Chinese industries and whether this fits the expectations of offshoring theory namely that China produces goods that are labor-intensive as a result of its cost advantage.

The WIOT provides data on how much developed countries are importing from Chinese industries. In order to compose level of exposure, it is assumed all goods, including both intermediates and final products, that are currently imported by developed countries will be produced by their domestic industries. Hypothetically, this would lead to an increase in gross output of industries. At the same time, the level of intermediates would remain the same since China is no longer producing these goods, but instead industries in developed nations produce these imports domestically. As a result, the level of value added for domestic industries would have increased, which implies that the additional production would have been produced with additional labor and capital.

(21)

21

Image 1: Amount of imports by high-income countries from a selection of Chinese industries in 1995. Source:

WIOT (2013) and author's calculations

Next, to provide insight on the level of exposure, the total level of imports from a Chinese industry, defined as IM, is divided by gross output (GO) of an industry in a developed country and is displayed the equation 4 below:

𝐼𝑚𝑝𝑜𝑟𝑡 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖𝑗𝑡 = 𝐼𝑀_𝑖𝑗𝑡 𝐺𝑂𝑖𝑗𝑡 (4)

(22)

22

Finally, after having decomposed the measure of import exposure, in the next step, the change of the import exposure is computed. This computation will follow the same approach as the change of the labor share in the above mentioned section, namely by composing an average of import exposure in the period before China joined the WTO and the period after. In order to obtain the change of import exposure, the averages of the two periods are subtracted which is displayed in equation 5 below:

∆𝐼𝑚𝑝𝑜𝑟𝑡 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖𝑗 = [ 1 6 ∑ 𝐼𝑚𝑝𝑜𝑟𝑡 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖𝑗𝑡 2000 1995 ] − [ 1 7∑ 𝐼𝑚𝑝𝑜𝑟𝑡 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖𝑗𝑡 2007 2001 ] (5)

This equation represents the observed change in import exposure for country i and industry j between the period prior to China joining the WTO and the period afterwards. This approach allows for dealing with endogeneity problems that often appear to be an issue in other studies that investigate the relationship between globalization and the labor share. As a result, the above introduced approach uses the China shock as a way to make changes in import exposure attributable to China joining the WTO in 2001.

4. Empirical analysis 4.1 First steps

(23)

23

Before continuing with the rest of this empirical analysis, it is worthwhile to already take a first glance at the data in order to see whether there is indeed support for a decrease in the labor share and an increase in import exposure in the considered time period. Despite the fact that this analysis aims to investigate the change of both variables, composed by a country average per year, Figure 1 and 2 below do provide several insights.

Figure 1: Labor share 1995-2007. Source: WIOD (2013) and author's calculations

Figure 2: Chinese import exposure 1995-2007. Source: WIOD (2013) and author's calculations

Figure 1 shows that over time, on average, the labor share had some fluctuations namely a slight increase around 1998 and again in 2002, but has been decreasing since. This might also have an effect on the change of the labor share, namely that for this period it has not always been decreasing. The trend for the import exposure, in Figure 2, is more decisive namely that on average, it has substantially increased especially since 2001. This provides support for a positive change in Chinese imports as a result of the China shock.

4.2 Descriptive statistics 0,63 0,635 0,64 0,645 0,65 0,655 0,66 0,665 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 L a bo r sha re ( la b div ided by VA)

Labor share 1995-2007 (average of 15 countries)

0 0,005 0,01 0,015 0,02 0,025 0,03 0,035 0,04 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 IE = I m po rt s div ided by g ro ss o up ut

(24)

24

Table 1: Summary statistics

In general, after analysing Table 1 above, a balanced sample is displayed with a total of 474 observations for every variable, which includes the 14 high developed countries and their respective 34 industries (no private households). In more detail, the ΔLS has a mean of -0.0158 implying that the change of the labor share as a result of the China shock is negative. The Irish agriculture industry experienced the largest positive change in the labor share of 0.0771, while the water transport industry in Belgium has the largest negative change in the labor share of -0.413. Furthermore, the ΔIE has a mean of 0.0109, which represents a positive change of import exposure. This implies that import exposure has increased as a result of the China shock. The Australian electrical and optical equipment industry has the largest change in terms of exposure of 0.413 while the Danish supporting and auxiliary transport industry experienced the largest negative change in import exposure of -0.0733.

4.3 Results

Note: This table displays the total number of observations, mean, standard deviation together with the minimum and maximum values of with the two main variables: the change of the labor share (labor compensation divided by value added) and the change of the import exposure (imports from China divided by gross output) derived from the WIOD Release 2013.

The dataset includes 14 developed countries and their respective 34 industries for a period of 13 years.

(1) (2) (3) (4) (5)

Variables N Mean Sd Min max

ΔLS 474 -0.0158 0.0770 -0.443 0.318

(25)

25

Table 2: Ordinary Least Square (OLS) regression (full sample)

.

Table 2 above displays the results of the primary Ordinary Least Square regression. This table shows that the effect of the change of import exposure as a result of the China shock is insignificant for the three regressions. This implies that there is no evidence of an effect between the change of import exposure and the change in the labor share. In other words, the hypothesis of a positive change in import exposure leading to a negative change in the labor share is not supported by this regression.

In more detail, the first column displays the effect of the change of import exposure on the change of the labor share without including any country and industry dummies. In this case, the change of import exposure is insignificant which means that there is no evidence for an effect. The explanatory power of this regression as indicated by R2 is rather low, namely that only 0.9 percent of the variance of the change in the labor share can be explained by the change of import exposure. In the next column, country dummies are included which control for differences between countries in this sample. The effect of import exposure remains insignificant, but the explanatory power increases which indicates that country differences influence the explanatory power of this model. Finally, in the last column both country and industry dummies are included which still leads to an insignificant effect of import exposure. Industry dummies control for variation between industries within a certain country and by including both the dummies it shows that the explanatory power of the model increases.

(1) (2) (3)

Variables OLS OLS OLS

ΔIE 0.198 0.202 0.219

(0.136) (0.124) (0.193)

Observations 474 474 474

R-squared 0.009 0.079 0.255

Country dummies Yes Yes

Industry dummies Yes

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(26)

26

Overall, this first empirical analysis shows that there is no significance relationship between the labor share and import exposure. This implies that there is no evidence of an effect between the change of import exposure on the change of the labor share as a result of the China shock.

4.4 Sensitivity analysis

In this section several adjustments have been made in order to see whether several data points are influencing the outcomes of the regression. First of all, it appears that there are relatively many industries included in the analysis that do not import any goods from China or only have started to import in the last couple years. Therefore, in order to see whether these industries have an influence on the results, all industries that did not experience any import exposure during the two periods have been excluded from the analysis. As can be observed in Table 3 below, there is still an insignificant relationship between the change of the labor share and the change of import exposure as a result of the China shock. However, when including country dummies the effect becomes significant at 10 percent level. This means that there has been the change in import exposure had a positive effect on the change of the labor share which is in contrast with what is expected based on offshoring theory. Subsequently, when including industry dummies and thereby also controlling for differences between industries, the effect again disappears. This repeatedly implies that there is no proof of an effect even when only including industries that experienced a certain amount of import exposure.

Table 3: Ordinary Least Square (OLS) Regression (excluding industries without exposure)

(1) (2) (3)

ΔIE 0.208 0.217* 0.228

(0.138) (0.124) (0.193)

R-squared 0.011 0.083 0.265

Note: This table displays three regressions without observations industries that did not experience any type of import exposure for the two periods and all include robust standard errors. The first regression does not include any dummies while

(27)

27

Next, since manufacturing industries are most likely to be influenced as a result of import exposure, based on the International Standard Industrial Classification of all Economic Activities Revision 3.0 (ISIC) and the WIOD (2013), the 14 manufacturing industries are separately examined (see Appendix A for a full overview). Table 4 below shows the results that indicate that there is still no evidence for an effect between the change of import exposure and the change of the labor share. This implies that the sensitivity analysis did not change the outcomes of the primary regression.

Table 4: Ordinary Least Square (OLS) regression (only manufacturing)

4.5 Extension empirical analysis

In this part of the paper several extensions are made to the empirical analysis. Firstly, the model aims to make the change in import exposure a result of increased imports. However, since the amount of imports is divided by gross output, it also be driven by changes in gross output. In order for the import exposure to increase, either the level of imports needs to increase or the level of gross output needs to decrease. For instance, when industries experience more competition, they also produce less which leads to a decline in gross output. However, since the aim is to make the change attributable to increased imports and because

(1) (2) (3)

ΔIE 0.205 0.204 0.200

(0.147) (0.131) (0.201)

R-squared 0.026 0.168 0.262

Note: This table displays three regressions without observations with only manufacturing industries and all include robust standard errors. The first regression does not include any dummies while the second includes country dummies and the final

(28)

28

gross output could again be influenced by other factors over time, the level of imports is divided by gross output in 1995 (t0). This is displayed in equation 6 below:

𝐼𝑚𝑝𝑜𝑟𝑡 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖𝑗𝑡= 𝐼𝑀𝑖𝑗𝑡 𝐺𝑂𝑖𝑗𝑡0 (6)

However, as can be observed in Table 5, the results do not change with this new measure. There remains no evidence for an effect for import exposure on the labor share.

Table 5: Ordinary Least Square (OLS) regression (t0)

Furthermore, it could be the case that China is not the only country that is competing with domestic industries in developed countries. The economic environment is rapidly changing and over time China may no longer be the only one that holds a comparative advantage in labor-intensive manufacturing production. This is mainly because Chinese wages have been increasing during the same time as the new wave of globalization (Yang et al., 2010). This could have led to a slow shift of production of labor-intensive goods towards other developing countries with relatively lower labor costs. For that reason, it is interesting to consider a wider set of developing countries that are likely to have a cost-advantage in producing manufacturing goods and thereby could be become a competitor for domestic industries of developed countries. Several countries in Asia are expected to take over a large part of the

(1) (2) (3)

ΔIE 0.0958 0.134 -0.0141

(0.102) (0.109) (0.168)

R-squared 0.002 0.072 0.249

(29)

29

labor-intensive manufacturing and include Malaysia, India, Vietnam, Indonesia and Thailand (Giffi et al., 2016; Aizenman et al., 2017).

Based on the availability of the WIOD and the above mentioned literature, Indonesia and India have been added to the analysis. The idea is the same for the primary regression, but now considers how much domestic industries in developed countries are importing from China, Indonesia and India. As a result, IM in equation 7 below includes the total level of imports per industry from these three respective countries.

𝐼𝑚𝑝𝑜𝑟𝑡 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑖𝑗𝑡 = 𝐼𝑀𝑖𝑗𝑡 𝐺𝑂𝑖𝑗𝑡 (7)

For this analysis, the US textile industry no longer turned out to be an outlier and is again included in the regression. The results are shown in Table 6 below and again show that there is no significant effect between the change of import exposure and the labor share. In other words, this implies that adding Indonesia and India did not change the outcomes of this analysis. In addition, as in the sensitivity analysis, this analysis also has been performed for only manufacturing industries, but once again there was no evidence for an effect (see appendix C).

Table 6: Ordinary Least Square (OLS) regression with data from India and Indonesia

(1) (2) (3)

ΔIE 0.163 0.163 0.245

(0.121) (0.108) (0.178)

R-squared 0.008 0.077 0.257

(30)

30 4.6 Comparing the results to preceding work

The results found in this study are in contrast with previous studies that find a negative effect of the labor share as a result of globalization (Doan and Wan, 2017; Harrison, 2006; Stockhammer, 2017). However, these studies use openness as a proxy of globalization and since the motivation behind this study is based on the work of Elsby et al. (2013) it is more valuable to compare the results with this article. These authors are one of the first ones that proxy globalization with their own measure of import exposure and whether this could explain the current decrease of the US labor share. This study composed a similar measure of import exposure, but the results show that this study cannot find proof for an effect between import exposure and the labor share. This is in contrast to the work of Elsby et al. (2013) who find a significant and negative relationship between the change in import exposure and the change of their measure of labor share. Their results indicate that increases in import exposure of US industries for a large part explain the decline in the US payroll share over the past twenty five years. In addition, the authors interpret their results as evidence for the fact that the decline in the US labor share has been largely driven by increases offshoring of activities to places all over the world.

(31)

31

Table 7: Ordinary Least Square (OLS) regression (percentage change)

The results by using percentage change for this analysis do not change. Thus, there continues to be no proof of an effect of import exposure on the labor share, even when including both country and industry dummies.

Moreover, Elsby et al. (2013) use for their measure of import exposure the total amount of goods the US is importing, while this study only considers the imports from China. This is in line with the approach introduced by Autor et al. (2013) and is selected with the idea to make the change in the import exposure attributable to the China shock which for this study implies the WTO membership of China in 2001. The measure of Elsby et al. (2013) considers all the goods the US is importing from the whole world. This makes it more difficult to state what is driving the changes in import exposure and thereby leads to an endogeneity problem.

Finally, another important difference with the analysis of Elsby et al. (2013) is that they only examine the effect of import exposure for the United States while this study considers 14 high-income countries and their respective 34 industries instead. In order to see whether the same effect as in Elsby et al. (2013) can be found for the US, Table 8 below shows the regression for only including US industries (using WIOD data). The US electrical and manufacturing industries are subsequently excluded in this analysis because they could be considered as outliers.

(1) (2) (3)

ΔIE(%) -0.000159 -0.00021 -0.00013

(0.000216)

R-squared 0.002 0.041 0.282

(32)

32

Table 8: Ordinary Least Square (OLS) regression (only United States)

As a result, the effect turns out to be insignificant, which is in contrast to Elsby et al. (2013) who do find a positive effect. This contrasting result is also observed in the data, namely that the industry in Elsby et al. (2013) being most exposed also experienced a large decline in their payroll share. However, for this analysis, the US machinery industry experienced the largest increase in import exposure, but at the same time, did not have a large decline in the labor share. In addition, the results of this analysis also provide a large standard error which might help explain why there is no evidence found. Moreover, it does not include any industry dummies that have been used in the previous analysis which makes it difficult to control for differences between industries. Lastly, the sample is rather small since only a total of 30 industries are used, whereas Elsby et al. (2013) have a total of 59 industries for the United States.

Although, this study is similar to the work of Elsby et al. (2013), for a large part it is also based on the work of Autor et al. (2013). They investigate the effect of increased imports from China on US labor market outcomes and aim to make the change in imports a result of a within China component. In more detail, the growth of Chinese imports is a result of rising productivity, increased investments and membership to the WTO thereby lowering trade barriers. However, these authors argue that the Chinese imports could also be correlated with increased unobserved industry demand, which leads to a subsequent endogeneity problem. In order to deal with this, the authors use an instrumental variables strategy. They instrument the the growth in Chinese imports for the US with the growth of Chinese imports from other

(33)

33

developed countries. In a following paper, Autor et al. (2016) show how the other developed countries also experienced Chinese import growth during the same period as the US. This provides support that there is an association between import competition for the US and other developed countries.

As a result of the similarities between these studies and since this study includes a total of 14 other developed countries (next to the US), the instrumental variable strategy could also be applied to in order to deal with the subsequent endogeneity problem. In general, this strategy allows for dealing with endogeneity, as a result of reversed causality or omitted variables, and to require consistent estimates of the included parameters (Altonji, 2005). In short, instrumental variable estimation is used when an explanatory variable is expected to be endogenous which in this case would be US import exposure as a result of unobserved industry demand shocks. In order to deal with this and to identify the supply driven component, the measure for import exposure for the US,∆𝐼𝐸 𝑈𝑆𝑖𝑗, is instrumented with a non-US exposure measure ∆𝐼𝐸 𝑛𝑜𝑛𝑈𝑆𝑖𝑗.This variable is based on data from the WIOD (2013) and represents the change in import exposure per industry, by an average per industry of 14 other high-income countries, which have been used throughout preceding parts of the analysis. The same restrictions hold for the OLS regressions, but the data for Great Britain for import exposure is again included in the analysis. Great Britain was excluded because of its labor share, but the data on import exposure could still be used for the construction of the instrument

(34)

34

Table 9: Instrumental variable (2SLS) regression (only for the United States)

Subsequently, despite the insignificant effect, in the post-estimation phase, several tests have been performed in order to see whether the instrument has been relevant. First, the test of endogeneity showed that US import exposure could be treated as exogenous. This implies that endogeneity is not a problem in this estimation and could mean that OLS estimation would already be sufficient since the variable is not endogenous, but this depends on the combination of other factors. It is therefore tested whether the exposure per industry for developed countries (not US) is a good instrument. The results provide a significant F-statistic which indicates that the instruments have significant explanatory power. However, it should be noted that according to Hall et al. (1996) a significant F-statistic is not sufficient. For that reason, others such as Stock et al. (2002) have suggested that the F-statistic should exceed 10 in order to be reliable. The F-statistic for this analysis is only 5.3, which despite the significant effect, makes it debatable whether the strength of this instrument is sufficient. In addition, the partial correlation between the change of the import exposure for the US and the instrument is around 0.4, which is not a strong correlation. Finally, there were no over- identifying restrictions since the number of instruments is equal to the number of right-hand expected endogenous variables.

Altogether, the results show that there is still no evidence of an effect of import exposure on the labor share and that the instrument used in the analysis turns out to be weak. This means that there could still be an endogeneity problem in this analysis and the instrumental variable technique could still be useful in order to deal it. These outcomes can be explained by

(1) Variables IV 2SLS regression ΔIE 2.104 (3.159) Observations 30 R-squared 0.05

(35)

35

differences between this study and the work of Autor et al. (2013). First, they use a different measure for import exposure namely the change of China imports divided by workers as their explanatory variable and their main dependent variable is the change in manufacturing employment. In addition, they use a longer time period namely already starting in 1990 which might have had an influence on the change of their variables. Subsequently, they consider a total of 397 manufacturing industries while this analysis only considers 30 industries. The authors also provide a wide range of control variables in their analysis which this study did not succeed to do. Finally, the authors admit themselves that even though they provide a good instrumental variable estimation, there could be subsequent problems in their analysis. For example, they highlight that demand shocks may be correlated in high-income countries and thereby provide several robustness checks which fall beyond the scope of this research. In general, already from the OLS estimation, it became clear that there was no evidence of an effect between import exposure and the labor share. Likewise, the same results were found for a separate regression for the US even after using instrumental variables. The above mentioned dissimilarities between the studies of Elsby et al. (2013), Autor et al. (2013), and this study can, amongst other factors, help explain the differences in results. In addition, as already highlighted, the labor share data appears to differ from Elsby et al. (2013) which might also have had an influence on the results. However, since the idea behind these studies is approximately the same, it is still valuable to make the comparison. Despite the insignificant effect, this study still provides a critical view on existing work. It makes one think about how import exposure and the labor share should be measured and that results should be interpreted with care since Elsby et al. (2013) admit themselves that their empirical strategy is subject to several caveats.

4.7 Other remarks

(36)

36

changes in both labor share and import exposure for many country industries relatively small. Interestingly enough, the data for import exposure actually shows what one would expect based on offshoring theory. Figure 3 below shows for a selection of industries, with a country average per industry, how as a result of the China shock both the labor share and import exposure changed.

Figure 3: Change in labor share and import exposure for a selection of industries

(37)

37 5. Summary and conclusion

To summarize, the aim of this research was to investigate the relationship between globalization and the labor share. Over the past three decades, there has been supporting evidence that the share of national output going to labor has declined. Several authors have introduced several factors that help explain this current decrease and one of them is globalization. The effect of globalization received substantial attention as a result of increased income inequality and the current changes in the economic environment of increased trade in activities and offshoring of production to places all over the world. Thereby, offshoring theory and vertical specialization highlight that developed nations have offshored labor-intensive activities to places that hold a cost-advantage. As a result, domestic labor is now being substituted for foreign labor, which subsequently leads to a decline in the domestic labor demand and a respective share of GDP going to labor. However, in literature, there has been quite some discussion about how globalization should be captured. Therefore, this study proposes a new measure of import exposure. This measure captures the change in import competition for developed domestic industries as a result of the China shock, namely China joining the World Trade Organization in 2001. As a result of its membership, China has been able to make use of its comparative advantage of producing labor-intensive goods. In more detail, this measure displays the level of imports from China divided by current gross output for a particular industry and measures how this has changed as a consequence of the China shock. This measure of import exposure is used as a proxy for globalization and this study investigates whether the change of the import exposure can explain the change of the labor share by using a sample of 15 high-developed countries and their respective 35 industries. The results of this study indicate that there is no evidence for an effect between increased import exposure and the labor share. The primary regression provides an insignificant effect even when controlling for unobserved heterogeneity between countries and industries. In addition, the sensitivity analysis and the several included extensions did not change the results. Finally, the results remained the same after having performed a separate regression for the United States and the subsequent instrumental variable strategy.

(38)

38

in the labor share over the considered time period. The data does show that the leather, textile, and electrical equipment industries are most exposed to Chinese import competition, which is in line with what is expected from trade with China. This provides support for the hypothesis introduced in this study namely that developed nations have increasingly offshored activities to places that hold a cost-advantage. In addition, because of current development in the popular media about the US and China having a large trade imbalance, it is worthwhile to further consider the effect of import exposure and the labor share going to GDP. However, at this moment, there is no support that increased import exposure led to a decline of the labor share.

This research is subject to several limitations that may also have influenced the results found in this study. First of all, this study examines a limited time period of 13 years due to data availability. It only considers a total of 35 industries because of data availability, while several comparable studies include substantially more (manufacturing) industries. For the primary regression, the data only includes China as a competitor for developed countries while other developing countries might also have a cost-advantage. In particular, over the past few years, China’s labor costs have been increasing which could have caused a shift of labor-intensive manufacturing towards other developing countries. For that reason, India and Indonesia are added in the extended empirical analysis. However, since the China shock only occurred in 2001, it is debatable whether the shift towards these countries is already observable in the data.

(39)

39

example, correlation between demand shocks in high-income countries. This would require several additional robustness checks which fall beyond the scope of this research.

The abovementioned limitations together with the relevance of this topic provide direction to further investigate the relationship between globalization and the labor share. An option for future research is to focus on collecting data for a longer time period in order to see a stronger effect of the China shock. This longer time period is also helpful when including the other developing countries that have started to produce labor-intensive manufacturing goods. At this point, India and Indonesia are included and in order to obtain a full picture of this shift, data on other developing countries like Vietnam, Thailand and Malaysia would be required. In general, the two measures used in this study could be improved: for the labor share, self-employed income needs to be dealt with in a way that both income from capital and labor are classified in a more suitable way. In addition, it should be further investigated which other mechanisms drive the labor share in order to see why the labor share, in this study, for several industries is different from what is expected. For the composition of import exposure, future research should focus on a more adequate measure of import exposure and should deal with the endogeneity problems. Even though this study already provided an instrumental variable strategy, future research should focus on finding a stronger instrument for import exposure to deal with endogeneity. This could be done by considering a wider set of manufacturing industries together with a longer time period. Besides, as highlighted in the literature, subsequent endogeneity problems exist again, after having performed a successful instrumental variable strategy. Future research should also try to take these into account by providing subsequent robustness checks.

(40)

40

exposed to increased globalization. For example, this could be done by only considering manufacturing industries, but this would require a broader range of manufacturing industries since the number of manufacturing industries in this dataset is rather limited. Subsequently, in order to further generalize this pattern for high-developed countries, future research should then focus on how to deal with heterogeneity between countries and industries and their respective level of exposure, or should consider countries and industries that are more similar.

(41)

41 References

Abdih, M. Y., & Danninger, M. S. (2017). What Explains the Decline of the US Labor Share

of Income? An Analysis of State and Industry Level Data (No.167). International Monetary

Fund Working Paper, Washington D.C.

Acemoglu, D., Autor, D., Dorn, D., Hanson, G. H., & Price, B. (2016). Import competition and the great US employment sag of the 2000s. Journal of Labor Economics, 34, 141-198. Aghion, P., Jones, B. F., & Jones, C. I. (2017). Artificial Intelligence and Economic Growth (No.23928). National Bureau of Economic Research Working Paper, Cambridge.

Aizenman, J., Jinjarak, Y., Ngo, N., & Noy, I. (2017). Vocational Education, Manufacturing,

and Income Distribution: International Evidence and Case Studies (No.23950). National

Bureau of Economic Research Working Paper, Cambridge.

Altonji, J. G., Elder, T. E., & Taber, C. R. (2005). An evaluation of instrumental variable strategies for estimating the effects of catholic schooling. Journal of Human Resources, 40(4), 791-821.

Alvaredo, F., A. B. Atkinson, T. Piketty, E. Saez (2013), The top 1 percent in international and historical perspective. Journal of Economic Perspectives, 27(3), 3–21.

Alvaredo, F., Chancel, L., Piketty, T., Saez, E., & Zucman, G. (2017). World Inequality

Report 2018. The World Inequality Lab, Paris School of Economics, Paris.

Amadeo, K. (2018, May 30th ). U.S. Trade deficit with China and why it is so high. The

Balance. Retrieved on 9th of June, 2018 from

https://www.thebalance.com/u-s-china-trade-deficit-causes-effects-and-solutions-3306277 Amiti, M., & Freund, C. (2010). The anatomy of China's export growth. In China's growing

role in world trade (pp. 35-56). University of Chicago Press, Chicago.

Andersen, T., & Herbertsson, T. (2003). Measuring globalization (No.817). Institute for Labor Economics (IZA) Discussion Paper, Bonn.