GLOBAL VALUE CHAIN PARTICIPATION AND ITS
INFLUENCE ON LABOR
By
Pauline Loeff
Groningen, January 2014
Supervisor: Dr. D. H. M. Akkermans Co-assessor: Dr. R. K. J. Maseland University of GroningenFaculty of Economics and Business Master Thesis
p.loeff@student.rug.nl
Global value chain participation and its influence on labor
ABSTRACT
Until recently, participating in the global value chain was seen as a necessity for countries in order to be able to increase their level of development. While this still holds true for a large share of the
countries, a debate has been started on the importance of the manner of participation. This paper responds on this debate and investigated the influence of participation in the global value chain on employment and wages. A distinction was made between developed and developing countries and participating in
manufacturing or service industries. Furthermore, it has been investigated if the impact is different on low skilled or high skilled workers. Based on the results it was concluded that the benefits of global value chain participation are not straightforward or evenly spread. It has become clear that a higher participation in the global value chain decreases employment for low skilled workers while it increases employment for high skilled workers. It can be concluded that participation by a country in the global value chain needs to be done with caution.
TABLE OF CONTENT
1. INTRODUCTION ... 5
2. THEORY ... 7
2.1 Global value chains ... 7
2.2 Power distribution in the global value chain ... 9
2.2.1 Power distribution ... 9
2.2.2 Employment and wages ... 10
2.3 Residual and relational view on poverty ... 10
2.3.1 Residual and relational view on poverty ... 10
2.3.2 Manufacturing versus services ... 11
2.3.3 High skilled versus low skilled workers ... 12
2.4 Control variables ... 13
3. DATA AND METHODS ... 16
3.1 Indicators ... 18
3.1.1 GVC participation ... 18
3.1.2 Employment and wages ... 22
3.1.3 Control variables ... 23
3.1.4 Dummies and interaction variables ... 23
3.2 Final dataset ... 24 3.2.1 Final dataset ... 24 3.2.2 Dataset limitations ... 27 3.3 Tests ... 27 3.3.1 Regression tests ... 27 3.3.2 Assumptions tests ... 30 4. RESULTS ... 32
4.1 Data summary statistics ... 32
4.1.1 GVC participation and wages: a comparison between countries and regions ... 32
4.1.2 GVC participation over time ... 32
4.2 Regression results ... 33
4.2.2 Crisis year versus non crisis years ... 34
4.2.3 Regression results total dataset, developed versus developing countries (model 1a/b) ... 36
4.2.4 Regression results manufacturing industries versus service industries (model 2 and 3) ... 37
4.2.5 Regression results high skilled workers versus low skilled workers (model 4 and 5) ... 39
4.2.6 Summary ... 41
5. DISCUSSION ... 43
5.1 Contribution to current literature ... 43
5.2 Research limitations ... 43
6. CONCLUSION ... 45
7. IMPLICATIONS AND FUTURE RESEARCH ... 46
7.1 Implications ... 46
7.2 Future research ... 46
Acknowledgement ... 47
REFERENCES ... 48
APPENDICES ... 53
APPENDIX 1 – DEVELOPED VERSUS DEVELOPING COUNTRIES ... 53
APPENDIX 2 – CONCORDANCE TABLES MANUFACTURING AND SERVICE INDUSTRIES 54 APPENDIX 3 – COUNTRIES RANKING, GVC PARTICIPATION ... 57
APPENDIX 4 – GVC PARTICIPATION, OVER TIME ... 58
1. INTRODUCTION
Thomas Friedman (2005) believed the world to be flat, or in other words to be a level playing field. In his book ‘The world is flat’ Friedman discusses the consequences of a world that is becoming more and more interconnected. He guides the reader through different examples of a world that is changing and he highlights the positive and negative consequences of this. This book provides an extensive collaboration on how to act in a world that is changing, a quest of a bigger story of which this paper focuses on one particular issue, participation in the global value chain.
Much earlier than Thomas Friedman, Ricardo (1817) noted in his book ‘On the principles of political economy and taxation’, that even though some countries are more efficient in the production of the
majority of products (‘competitive advantage’), other countries can relatively be better at the production of some products (‘comparative advantage’). Therefore, he proposed that it would be beneficial for all countries if a country would specialize in the production of products in which they have a ‘comparative advantage’. As a consequence of this, firms started to look for cheaper production possibilities abroad. The concept of moving production abroad is referred to as outsourcing. A well-known example of this is the move of labor intensive production to China, which has the advantage of cheap labor. Liberalization of trade and investment restrictions has greatly influenced the process of outsourcing (United Nations
Conference on Trade and Development (UNCTAD), Organisation for Economic Co-operation and Development (OECD) and World Trade Organization (WTO), 2013). First, this increase in
internationalization was characterized by the outsourcing of production processes, while later, firms started to outsource specific tasks to other firms and countries and therefore, spreading production over the globe (Grossman and Rossi-Hansberg, 2008). This shows that globalization is not a phenomenon just recently introduced. What has been new, however, is the current, very rapid increase of globalization. Today the production of one product can involve a large number of firms and countries, with products having passed through several countries before being consumed. This has resulted in both a large flow of intermediate products over the globe and an increase in complex webs of linkages involving many firms and countries. The World Investment Report (2013) estimates that about 60% of global trade consists of trade in intermediate goods and services that are used at various stages of production (UNCTAD, 2013, p. 122).
others. When visiting garment manufacturing factories in Vietnam, it occurred to me that even though the garment factories in Vietnam produce products that provide a large profit for developed countries where the end products are sold, the laborers working in these factories work very hard at low wages and often harsh working conditions. While the Vietnamese factories are actively engaged in the manufacturing of a product that is profitable, they still seem to be the losers from globalization. Kaplinsky (2005) has focused on this subject. In his book ‘Globalization, Poverty and Inequality’, he states the following; ‘There is a widespread recognition that globalization may induce greater inequality. But the idea that it might cause greater poverty runs against much of current conventional wisdom. This, as we shall see, argues that inequality and poverty are caused not so much by the working of the global economy as by the failure to engage positively with globalization’ (Kaplinsky, 2005, p. 24/25). From this statement the question follows on what influence participating in the globalization process has on important indicators of poverty that are greatly influenced by globalization, namely, employment and wages. The research question in this paper will therefore be: ‘Does global value chain participation influence the level of employment and wages? And if so is this influence positive or negative?’
2. THEORY
Globalization has brought wealth, in the form of income, increased availability of better quality products and a higher variety of products, to many people over the world. The World Bank (2002) has supported the view that participating in globalization has strong positive effects and has pulled many countries out of poverty. The countries that remain poor, they argue, are the ones that are not taking part in the
globalization process. This view is also reflected in the title of the World Bank report of 2002,
‘Globalization, growth and poverty: building an inclusive world’ (World Bank, 2002). Also based on this view, there was a request for policies that promoted globalization and the inclusion of as many as possible in the process. The most important contribution to this was the ‘Washington Consensus’ that was
implemented in 1989. This refers to the joint initiative by among others the IMF, the World Bank and the US Treasury Department to promote trade and investment liberalization. Currently, still many of the world’s biggest organizations have the intention to promote free trade and the participation of everybody in this process. However, more recently it has been noted that the countries that remain poor are not only the ones that have not been taking part in the globalization process. Also some countries that have been actively participating in globalization have remained poor and inequality is still growing. Therefore, recently, the discussion on participation in globalization seems to be becoming more complex.
2.1 Global value chains
Global value chain Sector A Raw materials Intermediate Final product Consumption Intermediate
Figure 1: The global value chain.
2.2 Power distribution in the global value chain
2.2.1 Power distribution
Within the framework of a GVC lie complex relationships between the different participants. Competition plays an important role in these relationships. Competition arises because ‘barriers to entry’ decrease. Barriers to entry are factors that influence the difficulty of a participant to become a player in a market. For example, they can arise due to economies of scale or government regulations. However, they can also be influenced by players in the market. Strong buyers can artificially keep barriers to entry low for suppliers, in order to improve their bargaining position, by for example taking control over important assets of a certain production process. Because of this ownership of crucial assets suppliers become reliant on these buyers. This way, a few buyers gather the power in a certain chain. Therefore, competition between the suppliers increases (Kaplinsky, 1998). The competitive pressure within the chains has increased impressively with the entry of more and more countries in production activities. Consequently, there is also increased competition for the economic rents to be earned in the chain (Kaplinsky, 2000). The concept of power in the value chain is related to the concept of governance that Kaplinsky (2000) refers to. Kaplinsky (2000) describes governance as follows; ‘there are key actors in the chain who take responsibility for the inter-firm division of labor, and for the capacities of particular participants to upgrade their activities’ (Kaplinsky, 2000, p. 124). In the World Investment Report (2013) it is stated that this power generally lies with transnational corporations (TNCs) that coordinate the GVCs (UNCTAD, 2013). The UNCTAD, OECD and WTO estimate in another report that total trade involving TNCs accounts for 80% of global gross exports (UNCTAD, OECD and WTO, 2013). Different structures of governance and the consequently different power distributions within GVCs have an influence on the distribution of economic gains from trade in GVCs (UNCTAD, OECD and WTO, 2013).
2.2.2 Employment and wages
The power distribution concerns that have been highlighted in the preceding paragraphs put pressure on the distribution of rents that are earned in the value chains. Consequently, this can be felt by the workers in the chain. Whenever rents earned within a country go down, this will influence employment and wages. Therefore, this paper will focus on the consequences of the previously introduced development concerns on labor indicators, employment and wages.
2.3 Residual and relational view on poverty
2.3.1 Residual and relational view on poverty
Two views on poverty exist, namely the residual and relational view on poverty (Kaplinsky, 2005). The ‘residual view on poverty’ argues that participating in the globalization process is the answer to poverty (Kaplinsky, 2005) and is therefore crucial in the development process. If it can be proven that this is indeed the case, this has important implications for developing countries’ policies. In the World Investment Report the following is stated: ‘Overcoming obstacles to GVC participation can pay big dividends; developing economies with the fastest growing GVC participation have GDP per capita growth rates 2% above average’ (UNCTAD, OECD and WTO, 2013, p. 7).The World Investment Report adds to this: ‘GVC participation tends to lead to job creation in developing countries and to higher employment growth, even if GVC participation depends on imported contents in exports’ (UNCTAD, OECD and WTO 2013, p. 22). Based on this the following hypotheses will apply:
H1a: Developing countries with relatively high level of GVC participation will have a higher level of
employment than developing countries with relatively low GVC participation.
and
H1b: Developing countries with relatively high level of GVC participation will have higher wages than
developing countries with relatively low GVC participation.
Opposing the residual view on poverty Kaplinsky (2000) argues that: ‘The issue is not whether to participate in the global economy but how to do so in a manner which provides for sustainable and equitable income growth’ (p. 117). This means that solely participating might not be enough. Kaplinsky (2005) describes this as the ‘relational’ view of poverty. This view argues that it is the process of
Moreover, he adds: ‘When firms confine their competences to the simple assembly of imported materials, they become subject to increasing competition and hence to falling returns’ (Kaplinsky, 2000, p. 120). These arguments indicate that Kaplinsky implies that participating in the global value chain does not per definition lead to an increase in income. The hypothesis as stated above are not confirm the residual view on poverty and therefore a third hypothesis, for the case of wages, will be added. The following
hypothesis will be tested:
H1c: Developing countries with relatively high GVC participation will have lower wages than developing
countries with relatively low GVC participation. 2.3.2 Manufacturing versus services
Indeed, it has been noted that a number of countries that actively participate in the value chain have experienced declining returns since their insertion in the chain. Kaplinsky (2000) refers to this concept as ‘immiserising growth’: ‘this describes a situation where there is increasing economic activity (more output and more employment) but falling economic returns’ (p. 120). This is especially apparent for the least developed countries that are restricted to their non-profitable comparative advantages, low skilled labor. Kaplinsky (1998) provides an example of immiserising growth. This example refers to China that started to rapidly increase its export in labour intensive products. When this happened it happened at such a significant scale that this had an impact on all developing countries’ exports. Terms of trade as a consequence fell significantly. With more countries inserting themselves in labour intensive production, competition increased. This had negative consequences for export income in these countries (Kaplinsky, 1998). Singer (1950) and Prebisch (1950) also discussed this phenomenon. Singer (1950) focused his arguments on the production of primary products. He argued that countries that specialized in the production of primary products experienced trouble when the prices of primary products fall, while the prices of manufacturing products increase. Value is added along the production process in the chain and the product becomes more expensive. Due to the position of these countries in the chain these countries are forced to export products that capture a lower value than the products they import from steps further on in the chain (Singer, 1950). ‘The exchange between ‘rent-rich’ and ‘rent-poor’ products’ is referred to by Kaplinsky as barter terms of trade (Kaplinsky, 1998, p. 34). Initially, the explanation for this was found in the declining terms of trade in the commodity sector. A large share of developing countries was active in this sector. Therefore, the advice at that time was to move out of the commodity sector, into
manufacturing of industrial products, referred to as import substitution. However, lately, countries in the manufacturing sector have experienced the same declining terms of trade. This has been especially
Kaplinsky (1998) states the following about the example provided above: ‘to foster growth through labour-intensive exports in LDCs1 may well be associated with an increase in economic activity, but not necessarily with an increase in real income, not just for the working class but also for the nation’ (p. 9).
The discussion above indicates that it possibly makes a difference in what industry a country participates. The theory argues that participation in manufacturing is subjected to falling barter terms of trade.
However, this analysis cannot be applied in the same way to the service sector. This is because in the service sector the rule does not apply that services produced at the beginning of the chain are necessarily worth less than at the end of the chain. This is because the value of the services is in many cases
dependent on a place or a person and, therefore, immobile. This suggests that services are less influenced by these barter terms of trade, which is beneficial for the rents to be earned in the sector. This is expected to also be visible in the effect on employment and wages. Therefore, the following hypotheses will apply: H2a: countries with relative high GVC participation in service industries have higher levels of employment
than countries with relatively high participation in manufacturing industries.
And
H2b: countries with high GVC participation in service industries have higher wages than countries with
relatively high GVC participation in manufacturing industries. 2.3.3 High skilled versus low skilled workers
The influence of GVC participation on employment is expected to not be homogeneous across different level of skills for workers. Research conducted by the Groningen Growth and Development Center (GGDC) on EU27 countries shows that low skilled employment is affected differently by globalization than high skilled employment. Timmer, Los, Stehrer and de Vries (2013b) from the GGDC state the following: ‘we also find that there is a shift away from activities carried out by low-skilled workers towards those carried out by higher-skilled workers.’ (p. 652). This statement applies to the EU27 countries. Referring to a theory from Baldwin (2006) Timmer, Erumban, Los, Stehrer and de Vries (2013a) argue in another article the following on the subject: ‘Taken together these trends fit a broad story in which firms in mature economies relocate their unskilled-labor intensive production activities to lower-wage countries, while keeping strategic and high value-added functions concentrated in a few urban regions where the high-skilled workers and intangible capital they need are available’ (p 23). For this reason the following hypotheses will apply:
H3a: GVC participation will increase employment for high skilled workers in developed countries.
And
H3b: GVC participation will increase employment for low skilled workers in developing countries.
2.4 Control variables
Some other factors besides GVC participation possibly have an influence on employment and wages. When investigating the influence of GVC participation on employment and wages these factors need to be controlled for.
Firstly, domestic investments have an influence on employment within a country. The influence of domestic investments, also referred to as capital accumulation, was shown in the ‘Solow model’, or ‘model for exogenous growth’ (Acemoglu, 2009). This model shows that capital accumulation has a positive influence on economic growth. Therefore, it is expected that domestic investment will have a positive influence on employment. The following hypothesis applies:
H4: An increase in domestic investment will increase the level of employment.
Domestic investments, or capital accumulation, could come in the form of human capital accumulation. This could for example be investments in education that increase the level of skill of the workforce. This will consequently increase the efficiency of the workers and therefore is expected to also influence wages. Therefore, the following hypothesis applies:
H5: An increase in domestic investment will increase wages.
Secondly, the level of foreign direct investment (FDI) inflow, like domestic investment, has an influence on employment in a country. FDI is considered to have a positive effect on employment for a couple of reasons. This is through an increase of activities in the host country, but also indirectly through increased competition and the import of better technology and management techniques (Alter, 1994). However, the effect of FDI inflow on employment can also be negative. Bornschier and Chase-Dunn (1985) find that in peripheral2 countries FDI will in the long run have a negative effect on economic growth. This is
explained by the dependency theory that states that resources flow from peripheral countries to the more wealthy countries, thereby reinforcing the division between poor and rich countries. Therefore, it is expected that this negative effect will be most pronounced in developing countries. Another explanation for a negative relation between FDI and employment is that FDI is invested in capital rather than labor.
However it is expected that the negative effect implied by Bornschier and Chase-Dunn (1985) will not have an effect on employment in this research. This is because, except for a one year lag, the long run effect is not taken into consideration in this research. Therefore, the following hypothesis will apply:
H6: An increase in FDI will increase the level of employment.
FDI, like on employment, is also expected to have an influence on wages. The OECD indicates that FDI generally has a positive influence on wages in the countries in which they operate (Arnal and Hijzen, 2008). However, there are examples where this does not hold. An example of a situation where FDI negatively influences wages can be seen in export processing zones (EPZ), which are generally situated in developing countries. In these areas, that stimulate economic growth by attracting FDI, FDI has a very big impact on employment. Big issues of wages, working conditions and a division between high skilled and low skilled workers exist in these areas. Large shares of employment in these areas are a consequence of increased FDI. Nevertheless, because of the repression of unions, wages and labor conditions have been influenced negatively with the coming of FDI (ILO, 2011). Still, when considering the effects of FDI on wages outside of the export processing zones the OECD argues that the effect is positive in a large number of cases. This is due to spillover effects (Arnal et al., 2008). While export processing zones may have a relatively significant impact on a country, it is expected that the influence on export processing zones does not outweigh the overall wage effect for a country. Therefore, we expect FDI to have a positive influence on wages in developed countries and developing countries:
H7: In both developed countries and developing countries an increase in FDI inflow will increase
the level of wages.
Furthermore, the level of education is expected to have an influence on wages. Studies on different countries have been performed and for the largest part they show that education has a positive effect on wages. Studies from Sudan and Portugal show a positive relation between education and income (Vieira, 1999, Samia, 2011). It must be taken into account that this effect will only be visible in the long-run and under the assumption that the education is completed and there is no early drop-out. The following hypothesis applies:
H8a: An increase in the level of education will increase the level of wages.
H8a: An increase in the level of education will not lead to an increase in the level of wages in the
manufacturing sector.
Finally, in order to test the separate influence of GVC participation in manufacturing and service industries, as proposed in hypothesis 3, a distinction will be made between manufacturing and services employment and wages. For this hypothesis the same control variables will apply as in hypothesis one, however, employment in manufacturing additionally will be controlled for employment in services, and employment in services for employment in manufacturing.
H9a: An increase in employment in manufacturing will decrease employment in services. H9b: An increase in employment in services will decrease employment in manufacturing. GDP per capita growth, domestic investments, FDI and hours per work week will be expected to have a comparable effect on both manufacturing and services as with the pooled sample.
3. DATA AND METHODS
It has become increasingly difficult to measure the contribution of countries in global value chains. This is the result of an increasingly large number of countries performing small value added tasks. Traditionally, mistakes have been made by using gross export as an indicator to investigate the contribution of a country to trade. The measure of gross export to investigate trade is not correct because countries use a large amount of intermediate products imported from other countries in their production. Hence, a significant share of gross export is not actually contributed inside the country. Consequently, a lot of value added is ‘double counted’, as gross export in two countries (UNCTAD, 2013). The amount of value imported in the form of intermediate products to be used for production is referred to as ‘foreign value added’ (FVA), the amount of value that is actually added inside the country is referred to as ‘domestic value added’ (DVA)3. Only DVA adds to the GDP of a country (UNCTAD, OECD and WTO, 2013). An indicator that
demonstrates how this issue can cause measurement problems is the indicator of ‘revealed comparative advantage’ introduced by Balassa (1965). Revealed comparative advantage indicates the advantage that a country has in a certain sector by investigating trade flows. The measure is calculated using the following formula:
Revealed comparative advantage = (Eij/Eit) / (Enj/Ent) (Balassa, 1965, p. 106)
i = country index n = set of countries j = commodity index t = set of commodities
This formula measures the relative contribution of a country in a certain sector by measuring its export in this sector as a share of total export in this sector in the world. This way it indicates if a country has a comparative advantage in the particular sector. Gross export is used in this case as a measure of the contribution of a country in a certain sector. This method neglects the fact that part of this export could have been imported from abroad as intermediate products to be used for production, like textile for clothing or computer components. Therefore, the measure of gross exports overestimates the value that is added within the country. On the basis of this example it can be seen how gross export measures provide us with false information.
In some cases the difference between gross export and domestic value added is not very significant. This will be the case when only a small amount of foreign value added is used in production. However, and this is increasingly the case, in some cases the difference between gross export and domestic value added will be very significant. This will be when a product crosses borders multiple time and many different
participants contribute small amounts of value. Dedrick, Kreamer and Linden (2008), while investigating
how profits from innovations are distributed, find that when an Apple iPod leaves China it is valued at $144, while only $4 is actually added in China. With an increasing amount of intermediate products being traded, this problem becomes more severe. The total value that is double counted worldwide shows how severe this problem is and how much the problem has increased over time. Table 1 shows the total export value that was double counted for 1995, 2000, 2005 and 2008 for a total of 56 countries. The World Investment Report estimates that, in 2010, about $5 trillion of the $19 trillion world exports of goods and services is foreign value added and was therefore double counted (UNCTAD, OECD and WTO, 2013).
Year Amount of export double counted (USD million) as a % of gross export 1995 1.156.742 20.2 % 2000 1.639.106 23.3 % 2005 2.887.859 25.7 % 2008 4.494.964 26.4 %
Table 1: Amount of value double counted for 56 countries (FVA + reimported value4).
Source: Based on own calculations (with data from TiVA database)
Research is currently being conducted to improve methods to investigate global trade relations more accurately. Measurements and datasets are being improved by institutions like the Organization of Economic Co-operation and Development (OECD), World Trade Organization (WTO) and United Nations Conference on Trade and Development (UNCTAD). Additionally, different researchers like Kaplinsky (2000), Gereffi (2011), Dickens (2011), Hummels, Ishii and Yi (2001), and Koopman, Powers, Wang and Wei (2010) have devoted attention to the subject (UNCTAD, OECD and WTO, 2013,
UNCTAD, 2013). One of these initiatives, the TiVA database, will now be discussed in more detail in the next paragraph.
The OECD and the WTO recently joined in an initiative to set up a project to consider value added in trade by each country participating in the production processes over the world and provide a new database with more accurate data. This project, as mentioned above, is the ‘Trade in Value Added’ project (TiVA). The first set of data was released by the TiVA in January 2013 and a second set of data was released in May 2013 (www.oecd.org). The TiVA database provides an improved method of measuring trade statistics, compared to conventional gross trade flows. The database focuses on the origin of value added and deals with two problems. Firstly, it addresses the problem of double counting. The TiVA data set deals with this by decomposing gross export and providing data on five different components of gross
export that will be explained in detail below. Secondly, the TiVA dataset provides a decomposition of the data for different industries, which makes it possible to compare export values for different industries within a country. With this improvement it can be indicated in which industry a country has an advantage over another country (TiVA, 2013a) (OECD, WTO, 2013). With the release of the TiVA database and other initiatives it has just become possible to do global analysis on the subject of global supply chains and value added. The TiVA database provides a detailed decomposition of gross export for 56 countries, both OECD and non OECD, in a convenient database.
In order to test the hypotheses that are introduced in the theory section, this innovative way of looking at value added is important. The concern for double counting has raised the attention for the decomposition of gross export, as is done in the TiVA database. This decomposition provides us with an opportunity to measure the contribution of countries to trade in more detail and reveal how much a country contributes and consequently earns from participation in the global value chain. The indicator of GVC participation is calculated using this more thorough decomposition of gross export. The following section will provide a description of the indicators that will be used to test the hypotheses.
3.1 Indicators
3.1.1 GVC participation
GVC participation is included in the regression models as an independent variable. GVC participation is defined by the World Investment Report as an indicator of ‘the share of a country’s export that is part of a multi-stage trade process, by adding to the foreign value added used in a country’s own exports also the value added supplied to other countries’ export’ (UNCTAD, 2013, p. 126). Koopman, et al. (2010), have introduced the indicator of GVC participation. As a first step towards this measure they decompose gross export into different value-added components, by considering efforts from Hummels, et al. (2001) and several others. The decomposition of gross export consists of the following five components:
(1) Domestic value-added in direct final goods5 (DVAF)
(2) Domestic value-added in intermediates absorbed by direct importers6 (DVAI) (3) Indirect domestic value added exports to third countries7 (IDVA)
(4) Reflected domestic value added8 (RIM) (5) Foreign value added (FVA)
5 Domestic value-added embodied in exports of final goods and services absorbed by the direct importer
6 Domestic value-added embodied in exports of intermediates inputs used by the direct importer to produce its domestically needed products
These five components add up to gross export: (Koopmans et al., 2010)
Gross Export = DVAF (1) + DVAI (2) + IDVA (3) + RIM (4) + FVA (5) (1)
GVC participation in this case is defined by the sum of ‘indirect domestic value added exports to third countries’ (IDVA) and ‘foreign value added’ (FVA) (Koopman et al., 2010). By definition, ‘reflected domestic value added’ (RIM) is included in the measure for IDVA. RIM measures value that is exported back to the source country. IDVA measures value that is exported to ‘third countries’. The source country is also included in these ‘third countries’. IDVA and FVA, together, capture exclusively value that is part of the value chain.
GVC participation = IDVA9 + FVA (2)
Using the definition of GVC participation as provided by Koopman (2010), it will be possible to calculate GVC participation using data from the TiVA database. In the TiVA database an indicator is provided for IDVA and for FVA. The GVC participation indicator can then be calculated as the sum of IDVA and FVA. GVC participation will be calculated as a share of gross export, to control for size differences between the countries. The following formula will apply:
GVC participation (as a share of gross export)10 = IDVA/EX + FVA/EX (4) Based on the theory by Koopman (2010) and the data available from the TiVA database, GVC participation will be calculated using formula (4).
Figure 2 shows the components of GVC participation in a schematic example. In the figure IDVA is shown in two components, namely IDVA to third countries and value exported back to the source country (RIM). These components are respectively indicated with IDVA (1) and IDVA (2) (RIM). GVC
participation for country Y is in this case the summation of FVA Y, IDVA (1) Y and IDVA (2) Y (RIM).
9 Including RIM
Decomposition Gross Export Country Y Sector B Country X Sector A Country Z Sector C X: Raw materials Y: Intermediates Z: Intermediates Raw materials Final product Consumption Intermediate Final product Consumption Intermediate Raw material Final product Consumption Third countries FVA Y DVA X FVA Z DVA Y DVA Z IDVA (1) Y IDVA (2) Y (RIM) DVA Z = Exports
= Domestic product flows
Third countries
The calculations explained above are illustrated in a numerical example in this paragraph. The numbers are shown in Table 2. In 2005 Malaysia contributed about 58.46 cents to every dollar that it exported (DVA), the rest, 41.54 cents per dollar, were intermediate products imported from abroad (FVA). Of the DVA, 26.53 cents was meant for re-export (IDVA), while the other 31.93 cents was meant for
consumption within the importing country. Of this 26.53 cents, 26.06 cents were exported by direct importers to third countries (IDVA excluding RIM), including 0.47 cents that returned to Malaysia (RIM). In 2005, Norway, on the other hand, contributed 85.52 cents to every dollar exported (DVA), while 14.48 cents was imported from abroad (FVA). Of this DVA, 43.53 cents was meant for re-export, while the other 41.99 cents was meant for consumption within the importing country. Of this 43.53 cents, 43.29 cents were exported by direct importers to third countries (IDVA excluding RIM), including 0.24 cents that returned to Norway (RIM).
Indicator Malaysia Norway
Total export = 100 100
FVA 41.54 14.48
DVA = 58.46 85.52
Value not exported 31.93 41.99
IDVA = 26.53 43.53
IDVA (excluding RIM) 26.06 43.29 RIM 0.47 0.24 Table 2: numerical example; decomposition gross export
Source: Koopman et al., 2010, based on own calculations, with data from TiVA database.
It must be noted that the measure of GVC participation includes both forward and backward linkages. Forward linkages are measured by IDVA and backward linkages are measured by FVA. Consequently, if a country has a high GVC participation, this could be due to the country having a high value for forward linkages, a high value for backward linkages, or both. However, investigating forward and backward linkages is beyond the scope of this paper.
3.1.2 Employment and wages
In order to test the hypotheses, data on employment and wages are included as dependent variables. Descriptive statistics will be discussed below. The International Labour Organization (ILO) database provides one of the most extensive and up to date datasets on labor indicators, covering over 230 countries and territories (http://laborsta.ilo.org). The ILO is a specialized agency of the United Nations that aims to promote rights at work, encourage decent employment opportunities, enhance social protection and strengthen dialogue on work-related issues. The ILO obtains data from official submission and official websites of national authorities and is updated monthly (www.ilo.org).
For wages a standardized dataset, constructed by Oosterdorp (2013) will be used that is based on data from the ILO. Oosterdorp (2013) has standardized the wage dataset provided by the ILO on an hourly wages and monthly wages basis. The dataset covers data for 171 countries and 161 occupations in 49 industries, for several years. For this research data based on hourly wages, as opposed to monthly wages, will be used to control for differences in the number of hours counted as working hours per month between different countries. Furthermore, the wages will be measured in current US$, corrected for inflation. To correct the wages for purchasing power parity (PPP) a conversion ratio computed with constant 2005 GDP PPP data from the World Bank is used. The database does not provide an average value of wages for all occupations, so in order to compute total wages per country the average of all occupations will be taken, making the assumption that this range of occupations represents all jobs in the country. Due to lack of information on the number of workers per occupation this average will be unweighted. A limitation of this source is that data is not available for all countries for all occupations. The average wage for different countries might be based upon wages for different occupations.
Nonetheless, the data give a suitable average of wages for the purpose of this research.
Data on employment are available from the ILO database. Data for different occupations, for a large number of countries and for several years are provided. It must be noted however, that the ILO uses two approaches to collect their data. These are through labor force surveys and population census. The data for this research is collected randomly from the two methods. Consequently, the two different surveys might cover different samples from the population. Nevertheless, this is not expected to significantly influence the results. In order to correct for differences in size of labor force per country, employment will be taken as a share of the number of people participating in the labor force.
For hypothesis two again a distinction needs to be made between wages and employment in the
World Input-Output Database (WIOD) provides a database that can be used for this purpose including data on industry level (www.wiod.org). The WIOD is a project set up in 2012 that constructed a database on world input-output data and socio economic data. Data on high and low skilled labor compensation in percentages (as a share of total labor compensation) and data on hours worked by high and low skilled persons engaged in percentages (as a share of total hours) is provided through this database (Timmer, 2012). Skill level is in this database defined on the basis of the level of educational attainment of the worker. The 1997 International Standard Classification of Education (ISCED) classification is used to define low medium and high skilled labor. More information can be found on
http://www.uis.unesco.org/Education/Pages/international-standard-classification-of-education.aspx (Timmer,
2012, page 58).
3.1.3 Control variables
Data for the control variables will be collected primarily from the World Bank Database
(data.worldbank.org). Domestic investment will be measured as gross capital formation as a percentage of GDP. The World Bank also provides data on foreign direct investment net inflow measured as a
percentage of GDP. Data on education will be measured as gross primary school enrollment. This measurement indicates the share of total participation in primary education, regardless of age, out of the population of official primary school age. While a measure of school completion rate would have been superior, this measure could not be used due to lack of data availability. Also due to lack of data
availability it is not possible to consider a lag for education. These limitations make the variable less well suited for the purpose, however, it will give an indication of the effect of education on wages.
3.1.4 Dummies and interaction variables
Bank11 (www.data.worldbank.org). In this case countries classified as ‘high-income’ countries are considered developed countries and all other countries (‘low-income’, ‘low-middle income’ and ‘upper-middle income’) are considered developing countries. The final selection of countries is shown in
Appendix 1, Table 12. Due to limited data availability the number of developing countries in the dataset is significantly lower than the number of developed countries. This also provides another reason why the Blonigen and Wang method is useful to apply in this case. Due to data availability constraints it is not possible to separately investigate the developing country dataset in hypothesis 112. By pooling the data, as is done in the Blonigen and Wang method, this problem is eliminated.
3.2 Final dataset
3.2.1 Final dataset
The final dataset includes data for 54 countries, including 38 developed countries and 16 developing countries over 4 years, 1995, 2000, 2005 and 2008. Because of the cross-sectional and time-series characteristics of the data, a panel dataset will be constructed. Table 3, provides a descriptive statistics table for the dependent, independent, control and interaction variables. It can be noted that GVC participation appears to have an outlier (maximum GVC participation of 4993.526), however, this data point represents Hong Kong, which we could assume to be realistic as Hong Kong is a very small country that is possibly highly reliant on imports and exports. To summarize, five separate datasets will be constructed: total dataset, a dataset for manufacturing industries, a dataset for service industries, a dataset for high skilled workers and a dataset for low skilled workers.
The time period used in the data represent a time period in which most of the developments surrounding GVCs took place. It must be noted that the global financial crisis, starting in the year 2008, most likely has had a significant influence on the outcome for the year 2008. Hence, in the regression analysis, a dummy will be included for 2008 to investigate whether there is a significant difference, and if so, this will be considered when interpreting the results.It should be taken into account that a lag might occur between a change in GVC participation and a change in employment and wages. This lag is caused by the fact that some time passes before the results are visible. The data on employment and wages will therefore cover
11 The groups are classified as follows: low income, $1,035 or less; lower middle income, $1,036 - $4,085; upper
middle income, $4,086 - $12,615; and high income, $12,616 or more. Based on 2012 GNI per capita (www.data.worldbank.org)
12 The rule of thumb states that for every dependent variable 15 data points should be included. In this case 16
the years 1996, 2001, 2006 and 2008. Data for 2009 is not available. Therefore, a lag for the year 2008 cannot be considered.
There are some missing data points. Data on employment and wages for Saudi Arabia is missing. However, considering GVC participation this is still an interesting case, so it will not be excluded. Moreover, due to a lack of availability many countries miss wage data for certain years. This is the case for the total dataset, as well for the wage data in manufacturing and services industries. Though, the distribution of available data is approximately evenly distributed across developed and developing
N (countries times 4 years)
Mean Std. Dev. Min. Max. DUMMIES
Year 216 2.5 1.120631 1 4
Crisis year 216 .25 .4340185 0 1
DC ( dummy, 1= developing country, 0= developed country)
216 .7037037 .4576839 0 1 GVC PARTICIPATION GVC participation 216 49.37205 11.42355 20.97558 84.86675 GVC participation manufacturing 216 374.2102 124.2068 96.97893 795.1744 GVC participation services 216 317.9689 525.288 38.13176 4993.526 WAGES/EMPLOYMENT Employment (total) 202 .913946 .1117563 .5073685 1.576543 Wages (total) 96 9.882101 6.590718 .195978 28.13715 Employment (manufacturing) 195 .3419386 .1411266 .0335925 .7659265 Wages (manufacturing) 96 9.040093 6.368572 .1952991 29.87504
Wages (manufacturing) (log13) 96 .7947175 .448268 -.7092997 1.475309
Employment (services) 186 .5586184 .1674205 .0350296 1.209421
Employment (services) (square12) 186 .3399334 .1904788 .0012271 1.462699
Wages (services) 93 10.92856 6.877827 .1985964 26.83057
Employment (high skilled) 152 .2005273 .0972095 .0251607 .4827951
Employment (low skilled) 152 .3315406 .2411347 .0233258 .8512724
Employment (low skilled) (log12) 152 -.6177969 .3741247 -1.632163 -.0699314
CONTROLS Domestic investment 216 23.93634 5.518596 14.30565 44.04627 FDI inflow 211 5.483997 7.072118 -3.528611 48.62239 Education 196 102.5968 6.371265 85.98441 133.5181 INTERACTIONS GVC participation*DC 216 13.5688 22.15561 0 84.86675 Domestic investment*DC 216 7.405092 12.15617 0 44.04627 FDI inflow*DC 211 1.361879 4.153206 -2.75744 48.62239 Education*DC 196 29.62897 47.77102 0 133.5181 Table 3: Descriptive statistics
3.2.2 Dataset limitations
A couple of data issues need to be taken into account when interpreting the results from the regression analysis. Firstly, when considering the total dataset no distinction is made between participation in
different industries. As a result a country might have a high participate in one industry and a relatively low participation in another, which would result in an average participation. Since the data is taken as an unweighted average of all industries, there is no possibility to investigate numbers separately per industries. Secondly, the availability of data for wages is very limited. As discussed above, gaps in the dataset exist that can influence the outcome of the regression analysis. Finally, due to a lack of data the number of cases studied in the regressions is low, with 54 countries.
3.3 Tests
3.3.1 Regression tests
A couple of regression analyses will be done in order to test the hypotheses as discussed in the theory section. These tests will be discussed in more detail below. Table 4 summarizes the hypotheses on the main effects discussed in the theory section that will be tested by the regression models.
Hypothesis 1 H1a: (Developing) countries with relatively high GVC participation will have higher employment
numbers than developing countries with low GVC participation. And
H1b: (Developing) countries with relatively high GVC participation will have higher wages than
developing countries with low GVC participation.
Hypothesis 2 H2a: Countries with relatively high GVC participation in service industries have higher
employment than countries with relatively high GVC participation in manufacturing industries. And
H2b: Countries with relatively high GVC participation in service industries have higher wages
than in manufacturing industries.
Hypothesis 3 H3a: GVC participation will increase employment for high skilled workers in developed
countries. And
H3b: GVC participation will increase employment for low skilled workers in developing
countries.
To test H1a and b, the following relationship will be tested, taking into account a dummy for developing countries, the variables that need to be controlled for and interaction variables.
(1) Employmentit = α + β1 GVC participationit + β2 DC + β3 Domestic investmentit + β4 FDIit + β5 GVC*DCit + β6 Domestic investment*DCit + β7 FDI*DCit + e
And
(2) Wagesit = α + β1 GVC participationit + β2 DCit + β3 Domestic investmentit + β4 FDIit + β5 Educationit + β6 GVC*DCit + β7 domestic investment*DCit + β8 FDI*DCit + β9 Education*DCit + e
In this case ‘employment’ and ‘wages’ are the dependent variable and ‘GVC participation’ is the
independent or explanatory variable. ‘Domestic investments’, ‘FDI’ and ‘education’ are control variables. ‘GVC*DC’, ‘domestic investment*DC’, ‘FDI*DC’ and ‘education*DC’ are the interaction variables. The subscripts i and t indicate respectively the year and the country that the data point represents. This test will apply to a sample including both developed and developing countries. The dummy for developing
countries will indicate the difference in employment and wages between a developed and a developing country, whereas the interaction variables indicates whether there is a significant difference in the effect of the specific variable on employment and wages between developed and developing countries.
In order to test H2a and b, two separate datasets will provide data on employment and wages in
manufacturing industries and on employment and wages in service industries. The same control variables will apply as in H1. Additionally, employment in manufacturing will be controlled for employment in services, and vice versa employment in services will also be controlled for employment in manufacturing based on arguments provided in the theory section. The following relationships will be tested for
manufacturing industries:
(3) Employment manufacturingit = α + β1 GVC participation manufacturingit + β2 domestic investment + β3 FDIit + β4 employment services + e
And
(4) Wages manufacturingit = α + β1 GVC participation manufacturingit + β2 domestic investment + β3 educationit + β4 FDIit + e
(5) Employment servicesit = α + β1 GVC participation servicesit + β2 domestic investmentit + β3 FDIit + β4 employment manufacturingit + e
And
(6) Wages servicesit = α + β1 GVC participation servicesit + β2 domestic investmentit + β3 educationit + β4 FDIit + e
H3 will also be tested on two spate databases including data on employment and wages for high skilled workers and including data on employment and wages for low skilled workers. A dummy and interaction variables will be included to indicate the difference between the effect in developed countries and the effect in developing countries. The following hypothesis applies for high skilled workers:
(7) Employment high skilled workersit = α + β1 GVC participationit + β2 DC + β3 Domestic investment + β4 FDIit + β5 GVC participation*DC + β6 domestic investment*DC + β7 FDI*DC + e
And the following for low skilled workers:
(8) Employment low skilled workersit = α + β1 GVC participationit + β2 DC + β3 Domestic investment + β4 FDIit + β5 GVC participation*DC + β6 domestic investment*DC + β7 FDI*DC + e
All tests will also be tested separately including a dummy for the year 2008 in order to test if the global financial crisis has had an impact on the regression outcomes. Table 5 summarizes the regression models and the accompanying datasets that will be investigated in order to test the hypotheses.
OVERVIEW OF MODELS AND DATASETS (employment/wages)
Hypotheses Models Dataset
Hypothesis 1: Model 1 (a/b) Total dataset
(employment/wages)
Hypothesis 2: Model 2 (a/b) Manufacturing industries (employment/wages)
Model 3 (a/b) Services industries
(employment/wages)
Hypothesis 3: Model 4 High skilled workers (employment)
Model 5 Low skilled workers (employment)
3.3.2 Assumptions tests
The models will be tested on ‘balanced’, ‘short’ and ‘wide’ panel datasets. Every cross-sectional unit has the same number of observations and thus we can consider the panel balanced. The number of countries is bigger than the number of years included in the panel and the number of years is relatively small, therefore the panel is considered to be a short and wide panel.
Several assumptions need to be tested before proceeding with the regressions. These are the following: normal distribution of the dependent variable, multicollinearity, heteroskedasticity and serial
autocorrelation. Furthermore, a test needs to be performed to test the appropriateness of a pooled OLS regression. Finally, if OLS cannot be performed, a test needs to be run to decide whether the regression needs to be done using random effect estimations or fixed effect estimations. Table 6 summarizes the tests that will be performed in order to verify the assumptions corresponding to the regression analyses. Table 6: Summary assumption tests
Two tests need to be conducted to indicate what regression method, pooled OLS, fixed effect or random effect, is the most appropriate to use. In order to test for the appropriateness of the pooled OLS regression the Breusch Pagan Lagrange Multiplier test for random effects will be performed. Whenever the null hypothesis14 is rejected, there is significant difference in the variance between countries meaning that it is not possible to perform pooled OLS estimations. For this dataset we expect that OLS is not appropriate since it neglects the heterogeneity across countries, which most likely there is.Also, a test for
overidentification is performed to test for fixed or random effects estimation. This test produces a Sargan-Hansen statistic. Whenever the null hypothesis is rejected, random effect estimation cannot be applied and fixed effect should be performed. Both fixed and random effect estimation apply assumptions that might be inappropriate for the purpose of this regression. The random effect model assumes that the sample is chosen randomly, which is not the case. The countries not included in the dataset are countries for which
14 Var(u) = 0
EXPLANATION TESTS (STATA)
data was unavailable. Data might be unavailable for different reasons of which one is war. The fixed effect model, on the other hand, omits dummy variables that do not change value in the given time period from the regression. In this case the dummy for developing countries will be omitted in all cases, because countries do not change status within this time period. Therefore, the test results need to be subjected to further argumentation, which will be done in the result section.
In order to test for heteroskedasticity and serial autocorrelation, respectively a modified Wald test for group wise heteroskedasticity and a Wooldridge test for autocorrelation are performed. The modified Wald test tests the null hypothesis that data is homoscedastic against the alternative hypothesis that the data is heteroscedastic15. The Wooldridge test for autocorrelation in panel data tests the null hypothesis of no first-order autocorrelation against the alternative hypothesis of first-order autocorrelation (Drukker, 2003). Normal distribution of the dependent variable will be tested by means of a Skewness-Kurtosis test for normality, which tests the null hypothesis that a variable is normally distributed against the alternative hypothesis that the variable is not normally distributed. Furthermore, multicollinearity will be tested by investigating the ‘variance inflation factors’. ‘Variance inflation factors’ indicate the existence of
multicollinearity, by indicating the impact of collinearity on variance. Whenever this indicator has a value higher than four, it can be concluded that the data is subjected to multicollinearity. The results for these tests will be provided in the result section.
4. RESULTS
4.1 Data summary statistics
4.1.1 GVC participation and wages: a comparison between countries and regions
Appendix 3, Table 15, provides a table indicating the seven countries with the highest and the seven countries with the lowest GVC participation. It can be noted that most of the countries that rank highest in GVC participation are relatively small countries like Singapore, Luxembourg, Malaysia and Estonia. Just out of the ranking, on respectively an ninth and eleventh place are Belgium and Malta, confirming this pattern. On the other hand, countries that rank low on the GVC participation, are relatively large countries, most of which rely heavily on primary products. For example, Saudi Arabia, a country that relies heavily on primary products, ranks just out of the ranking on an eleventh place. These findings would logically result from the fact that small countries are relatively heavily reliant on imported value for production and also relatively heavily reliant on value earned from export. Additionally, large countries with relatively low GVC participation are countries that do not rely on import or export value.
Furthermore, Appendix 3, Figure 3, includes a graph showing the distribution of GVC participation across different regions. It can be seen that the EU has the highest GVC participation. The World Bank explains that this is due to the large amount of internal trade within the EU. The EU would score considerably lower if internal trade would be excluded from the measure (UNCTAD, 2013). Consequently, we can conclude that East EU and Asia have the highest GVC participation rate, while Latin America (America’s) has experienced the largest increase in GVC participation.
Finally, Appendix 3, Table 16, shows a ranking of countries by wages. Not surprisingly, from the ranking it can be seen that the countries with the highest wages are the most developed countries, while from the ten countries with the lowest wages only three are classified as developed countries.
4.1.2 GVC participation over time
Appendix 4 includes several graphs showing GVC participation over 4 year for different sets of data. The first graph shows GVC participation for all countries for 1995, 2000, 2005 and 2008. From the graph it can be noted that GVC participation has been increasing between 1995 and 2008 by about 34 percent. This increase has slowed down over the years however. The second graph shows a comparison between developed and developing countries16. It can be observed that GVC participation has increased more
16 Developed and developing countries are defined according to the definitions by the World Bank as explained in
rapidly in developing countries than in developed countries, with developing countries having almost reached equal levels of participation as developed countries by 2008.
In the final four graphs a separation is made between participation in manufacturing industries and in service industries. The first two graphs represent manufacturing industries, while the last two graphs represent service industries. Again the amount of participation has increased over time, but, while the rate of growth between 1995 and 2008 has been slowing down in manufacturing industries, it has been rising steadily for service industries. In the two right graphs a separation is made between developed and developing countries. It can be noted that developed countries have had higher GVC participation in both manufacturing and service industries. In manufacturing industries, however, developing countries have been catching up with developed countries, while in service industries developing countries are still far behind the level of GVC participation in developed countries. This is consistent with the theory that developing countries are advised to participate in manufacturing industries first, before starting to specialize in services industries.
4.2 Regression results
4.2.1 Regression assumption tests
Before running the regressions a couple of assumptions need to be tested and the appropriate regression method needs to be chosen as described in the methodology section. The detailed results for the tests discussed in this paragraph are provided in Appendix 5. Table 17 in Appendix 5 shows an overview of the tests and Tables 18 through 20 show the results for the different models. Firstly, the Breusch Pagan Lagrange multiplier is estimated. In all cases the null hypothesis is rejected, indicating that pooled OLS regression is not appropriate. Therefore, the overidentification test will be tested next. The
overidentification test for fixed or random effects shows whether a fixed effect regression or a random effect regression is more appropriate. The results of this test vary between the models. In five cases the null hypothesis is rejected and fixed effects should, according to the test, be applied. In all other cases the outcomes of the overidentification test suggest that a random effect regression is more appropriate. However, as discussed in the methodology section random effect cannot be applied because the
assumption of a random sample is violated and fixed effect will wrongfully omit the dummy variables and the interaction variables in the case that they are multiplied by zero. Due to these contradictory results the regressions are tested using all three methods and then compared. Only in two cases, models 3a and 3b, do fixed effect and random effect give comparable results. In all other cases the results differ between the three different models17. Since the fixed effect regression cannot be applied when interaction terms are
included in the regression, random effect regression will be applied to models 1a/b, 4 and 5. In models 2a/b and 3a/b the results from the overidentification tests will be followed and fixed effect will be applied. The existence of heteroskedasticity is tested by performing a Wald test. All the models are subjected to heteroskedasticity. When running the regression this will be controlled for using a robust option. This option includes robust standard errors that are valid for both heteroscedastic and homoscedastic errors. Furthermore, a Wooldridge test is applied to test for serial autocorrelation. This needs to be done because it could be that data for countries in the same region are not independent. If this is true, the data needs to be clustered. Indeed, with the exception of model 2a, all the models are subjected to serial autocorrelation. Therefore, these models will be controlled for serial autocorrelation using the cluster option. This option clusters the time-series observations on individuals.The data distribution of the dependent variables is also tested on normality by a skewness-kurtosis test. With the exception of wages in the total sample, none of the dependent variables are normally distributed. Therefore, in some cases, transformations will be applied to make the dependent variable more normally distributed. In the case of low skilled workers employment the data will be logged and in the case of service employment the data will be squared. These transformations made the data normally distributed in the first case and a slightly more normally
distributed in the second case. The results from the Skewness-kurtosis test before and after the transformation can be seen in the table in appendix 5. Manufacturing employment and high skilled workers employment are approximately normally distributed so therefore, will not be transformed and service wages and total employment and manufacturing wages cannot be transformed to make them normally distributed. Finally, variance inflation factors are investigated to test for multicollinearity. None of the variance inflation factors, however, are above three and most are not above 2.5, indicating that none of the models suffers from multicollinearity. Only the dummy variable and interaction variable have large variance inflation factors. However, this is expected and does not need to be controlled for.
4.2.2 Crisis year versus non crisis years
By including a dummy variable for the crisis year (2008) into the regression it can be investigated whether this year has a significantly different outcome in the regression than the other years. The dummy
Table 7: regression result for the dummy variable for crisis years NOTE: robust standard errors are in parentheses
NOTE: significant results are indicated by * (* p<0.1, ** p<0.05, *** p<0.01)
From the table it can be seen that half of the coefficients for the dummy for crisis year are significant at a 5% or 1% level. Therefore, it could be concluded that the crisis year has had a significant impact on these regression. It must be noted that all except one regression for employment are significant, while the regressions for wages are all not significant. This could indicate that the crisis has had a significant impact on employment, however, not so much on wages. Furthermore, the crisis seems to have had opposite effects on high skilled and low skilled workers. While the coefficient is positive for high skilled worker employment, the coefficient is negative for low skilled workers employment. Unexpectedly, many of the significant coefficients are positive. According to the expectations the crisis would have a negative effect on employment, due to a stagnation of economic activity. It must be noted that while the crisis officially started at the end of 2007, its impact only became visible over the next year. Therefore, the impact on GVC participation, wages and employment measured for 2008 is likely to be limited. Furthermore, the significant and in most of the cases positive results could be a reflection of an overall upwards trend in this time period, therefore, there is no well-founded reason to exclude 2008 from the regression. Moreover, by excluding 2008, important information is possibly missed.
Dummy (1= year 2008) coefficient
Model 1a (Total employment) .0089211* (.0046055) Model 1b (Total wages) -.2255449
(.7613052)
Model 2a (Manufacturing employment) -.0035595 (.008349)
Model 2b (Manufacturing wages) .0237024 (.0366293)
Model 3a (Services employment) .0371832*** (.0049561) Model 3b (Services wages) .6975282
(.4454688)
Model 4 (High-skilled worker employment) .037738*** (.0052433) Model 5 (Low-skilled worker employment) -.0660387***
4.2.3 Regression results total dataset, developed versus developing countries (model 1a/b)
Model 1a/b tests the relationship between GVC participation and employment and wages for the total dataset. A dummy and interaction variables are included to compare the results between developed and developing countries. Table 8 shows the regression results for model 1a/b. Whenever applicable the expected signs are indicated as a plus or minus in the column behind the coefficient. Furthermore, the robust standard errors are given in parentheses below the coefficient. For both models the Wald Chi2 is significant at a 1% level, indicating that the model fits the data well. However, the R2 for the employment model (1a) is very low, whereas the R2 for wages is high, indicating that the model of wages fits the data well, while the model of employment does not. Therefore, the interpretation of this model needs to be considered with caution. Most of the control variables have the expected sign, however, many are not significant. Education has a different than expected sign. However, this coefficient is not significant. The interaction variables for the control variables are negative for employment and positive for wages, indicating that employment in developing countries is less positively influenced that in developed countries, however, wages in developing countries are more positively influenced than in developed countries. However, all coefficients are not significant.
GVC participation is expected, according to the hypothesis on the residual view of poverty introduced in the theory section, to increase employment in both developed and developing countries. While in
developed countries the effect of GVC participation on employment is positive, the same effect is negative for developing countries18. However, both effects are not significant. GVC participation is also expected to increase wages in both developed and developing countries. Indeed, this effect is in both cases positive19, however, again not significant. When considering the hypothesis on the residual view on poverty, which states that GVC participation is an effective manner to reduce poverty, the results suggest that indeed in terms of wages GVC participation could possibly reduce poverty, however, in terms of employment in the case of developing countries the opposite effect seems to be visible. In developing countries GVC participation seems to decrease employment, instead of increase as expected by the hypothesis. However, from the results it also seems that the effect of GVC participation on both
employment or wages was not significant. From this we can concluded that the regression results suggest that the relation between GVC participation and poverty reduction in terms of increased wages and employment does not seem to be clear-cut as is suggested by the hypothesis. Therefore, in the following models more specific regression will be run based on hypotheses 2 and 3 as described earlier.
