Industrialization in Asia and Latin America
Exploring the importance of global value chain integration
MSc Thesis International Economics and Business (IE&B) Sophie Pille (S2502771)
Supervisor: dr. A. Minasyan Co-assessor: dr. G.J. de Vries
2 ABSTRACT: This paper is the first to explore the relation between integration into global value chains and industrialization to explain the divergent industrialization paths of Asia and Latin America. Conducting panel analysis for 62 countries between 1995 and 2011, it finds suggestive evidence for the heterogeneous effects of global value chain participation on industrialization in Asia and Latin America. In Asia, many backward linkages in manufacturing global value chains are associated with larger manufacturing sectors. However, this relation is not statistically different for Latin America. The results do not indicate that global value chain integration induces industrialization.
3
Table of Contents
1. Introduction ... 4
2. Literature Review ... 7
Structural change, industrialization and the centrality of the manufacturing sector ... 7
Drivers of deindustrialization in developed countries ... 8
Drivers of deindustrialization in developing countries ... 9
A comparative advantage in manufacturing and industrialization: suggestive evidence . 10 Global value chains and the manufacturing sector ... 11
Measures of global value chain participation ... 13
Hypothesis ... 14
3. Data and descriptive statistics ... 15
3.1 Data and data sources ... 15
Measuring industrialization ... 15
Measuring vertical specialization ... 16
Control variables ... 17
3.2 Descriptive statistics ... 17
3.3 Descriptive evidence ... 21
4. Methodology ... 23
4.1 The share of manufacturing employment in total employment ... 24
4.2 The absolute change in the share of manufacturing employment ... 26
5. Results ... 27
5.1 Estimation and results of the first model ... 27
5.2 Estimation and results of the second model ... 31
5.3 Robustness check ... 33
6. Concluding remarks ... 35
7. Bibliography ... 37
8. Appendices ... 42
Appendix A – Data sources manufacturing employment ... 42
4
1. Introduction
The share of manufacturing employment in advanced economies is on the decline, which means that these countries are deindustrializing. Given that the manufacturing sector has made an important contribution to the development paths of most of these countries, this has received considerable attention (Sachs & Satz, 1994; Rowthorn & Ramaswamy, 1999; Bernard et al., 2017). More recently, studies find that deindustrialization is not only a pattern observed in advanced economies. The share of the manufacturing sector in developing countries has declined too, at an early stage of development. The exception is Asia. On average, it is the only region that has increasing manufacturing shares (Rodrik; 2016). This has resulted in the concentration of manufacturing in Asian countries (Haraguchi et al., 2017). In contrast, Latin American countries do, on average, no longer benefit of increasing manufacturing shares (Rodrik; 2016). Although studies hypothesize about the reasons for the differences between the two regions, no study attempted to address these empirically. Therefore, this paper contributes to existing literature by exploring the consequences of global value chain participation in explaining (de)industrialization in Asia and Latin America.
The acceleration of globalization since the 1990s has changed the production of goods and their trade considerably. The decline in trade costs and the recent decline in communication costs have made it possible to relocate a specific production activity to another country, instead of a whole production process. Therefore, countries no longer have to build a complete domestic supply chain to produce one final good. Instead, they can specialize in the activities that they perform most efficiently and join a global value chain (Baldwin, 2011). A global value chain comprises all the activities performed in different countries to produce one final good (Gereffi & Fernandez-Stark, 2016). The consequence is that countries often produce a specific part of a final product, as exemplified by the production of the Ipod and Notebook PC’s (Dedrick et al., 2010).
5 policy is considered to have had a negative effect on different countries in Latin America (Balassa, 1978; Rodrik; 2005; McMillan et al., 2014).
The fragmentation of production and the accompanying price competition may have reinforced the comparative advantage that Asia has gained in manufacturing. With access to world markets, the possibility of economies of scale is tremendous. Moreover, Asian countries benefit from their proximity to each other. This results in the relocation of activities from one Asian country to another, driven by low labour costs (Gereffi, 1999; Ng & Yeats, 1999).
The exploration of the drivers of industrialization and deindustrialization is important due to the distinct characteristics of the manufacturing sector. These characteristics make the sector essential for development. First, the sector is central in generating productivity, even leading to unconditional convergence with developed countries productivity levels (Rodrik, 2013). Second, it has the ability to absorb large amounts of low-skilled labour. Third, goods of the sector are highly tradable. This prevents the problems of small home markets and of the presence of domestic consumers with low purchasing power. Fourth, the sector is known for its many backward and forward linkages with other segments of the economy (Kaldor, 1967; Page, 2012; Rodrik, 2016).
In general, manufacturing as a share of total employment follows an inverted U-shape over the course of development, considering three broad sectors in the economy: agriculture, manufacturing and services. First, a country moves out of agriculture into the more productive manufacturing sector. As a country grows richer, the manufacturing sector declines and the services sector increases. The change in the share of these sectors is determined by the income elasticity of demand for manufactures, productivity growth in the manufacturing sector and a fall in the relative prices of manufacturing goods (Rowthorn & Ramaswamy, 1999).
6 fragmentation was made possible, which provides an additional reason to explore their influence on industrialization (Rodrik, 2016).
To capture the consequences of globalization with regard to the fragmentation of production, this study employs the vertical specialization measure of Hummels et al. (2001). Vertical specialization captures a country’s involvement in intermediate goods trade. More specifically, it indicates how much foreign value added shows up in a country’s exports of intermediate and final goods. It is a measure of backward linkages in global value chains and shows to what extent a country sources inputs from abroad.
It is expected that vertical specialization in manufacturing global value chains has affected the manufacturing sectors in Asia and Latin America in different ways. Therefore, the following research question is central to this paper:
Does vertical specialization within manufacturing global value chains have a heterogeneous effect on industrialization in Asia and Latin America?
To analyse the effects of vertical specialization on industrialization in Asia and Latin America, this paper conducts a panel analysis of 62 countries over the period 1995-2011. First, it finds that for Asian countries a higher rate of vertical specialization is associated with a larger share of the manufacturing sector in the economy. Second, it does not find this relation between vertical specialization and the level of industrialization for Latin America. Third, the results do not indicate that an increase in vertical specialization induces industrialization.
7
2. Literature Review
The literature review provides an overview of the most important literature with regard to industrialization and an overview of the literature related to global value chains. First, it elaborates on structural change, industrialization and the centrality of the manufacturing sector in development. Second, it considers the drivers of industrialization in advanced economies. Third, it assesses how the drivers of industrialization in developing countries are different. Fourth, it discusses the relation between global value chains and industrialization.
Structural change, industrialization and the centrality of the manufacturing sector
Structural change is an important concept when discussing industrialization. Models of structural change use two sectors to explain economic development: the traditional sector and the modern sector. The first is the agricultural sector and the latter can be seen as the manufacturing sector. Put simply, development takes place when labour moves from the less productive traditional sector to the more productive modern sector. Economy-wide productivity increases due to the shift of labour to the more productive sector, as the latter is characterized by capital-intensity and the possibility of large economies of scale (Lewis, 1954).
8 The importance of manufacturing in economic development has been under discussion due to the declining share of manufacturing and the increasing share of services in both developing and developed countries. Examining the period 1995-2005, Szirmai and Verspagen (2015) find that manufacturing is still very important in generating economic growth. However, the sector’s contribution to economic growth has decreased since 1990 and the positive effect of manufacturing is dependent on a sufficient amount of human capital. Su and Yao (2016) examine the role of manufacturing in establishing growth of the services sector. Their findings show that the manufacturing sector is indeed the driver of the services sector. This does not work the other way around. Marconi et al. (2016) find that higher output of the manufacturing sector leads to higher economic growth rates and productivity growth. This effect is larger for low and lower middle-income countries.
These empirical studies show that manufacturing is still very crucial in generating economic growth for developing countries. Therefore, the observation of premature deindustrialization in developing countries is worrisome (Rodrik, 2016). Capturing the drivers of deindustrialization is of vital importance to adjust policy advice. The next section discusses the drivers of deindustrialization in developed countries.
Drivers of deindustrialization in developed countries
Deindustrialization can be explained by both internal and external factors to an economy. I name three external and one internal factor(s). First, when a country becomes richer its demand switches increasingly from goods to services (Clark, 1957). Second, technological change spurs productivity growth in the manufacturing sector, reducing the need for manufacturing workers. This does not always result in a decrease in manufacturing output. However, it decreases the price of manufacturing goods. Third, due to the price elasticity of manufacturing goods, this price decline does not necessarily increase the demand for these goods. The latter reinforces the decreasing need for manufacturing workers (Rowithorn & Ramaswamy, 1999). Fourth, the external factor affecting the manufacturing in advanced economies is trade with developing countries (Sachs and Satz, 1994; Rowithorn & Ramaswamy, 1999).
9 industries that use more ICT, demand more high-skilled peopled at the expense of middle-skilled people. Considering trade, Foster-McGregor et al. (2013) find that the jobs of medium-skilled workers disappear most due to offshoring. In addition, Blinder and Krueger (2013) find that production jobs are the most likely to be offshored. Other studies find explanations in the rise of China, due to the import competition from this country (Autor et al., 2013).
Drivers of deindustrialization in developing countries
The literature suggests different explanations for the decline in the manufacturing sector in developing countries. Therefore, the influence of trade shocks, technology and inequality are discussed in this section.
First, trade shocks are considered important in explaining deindustrialization in developing countries. The first trade shock, the liberalization of trade, differently affected countries, depending on the policies they had pursued. On the one hand, countries that had implemented import substitution strategies were negatively affected. When exposed to global competition, firms that had received continuous state support were unable to compete on the world market. On the other hand, countries that had built up large industries under export-led growth were ready to compete on the world market. Countries that had pursued import substitution policies only, essentially Latin American countries, became net importers of manufacturing goods. In contrast, countries that combined both polices, mainly East Asian countries, became net exporters of manufacturing goods. The second trade shock, a decline in prices of manufacturing goods in developed countries due to technological change, influences the prices of developing countries as well. Developing countries that are technological more advanced are better capable of remaining cost-effective: they see an increase in productivity and subsequently in output (Rodrik, 2016).
10 The study by Haraguchi et al. (2017) contradicts the argument that the market for manufacturing goods has declined and that this results in deindustrialization in developing countries. This study compares developing and developed countries with regard to their average shares and total shares of manufacturing. It considers manufacturing value added at both current and constant prices and manufacturing employment. Both at current and constant prices, the average country share of manufacturing value added of developing countries is indeed on the decline. This is the case for manufacturing employment as well. However, when considering the total share of developing countries in the world aggregate of manufacturing value added, no declining trend is observable. Therefore, the explanation for deindustrialization in other developing countries is most likely to be found in the primacy of Asia in the production of manufacturing goods.
A comparative advantage in manufacturing and industrialization: suggestive evidence
In a further attempt to explain the different deindustrialization patterns, Rodrik (2016) splits his sample of countries into two groups. The first group consists of countries that import more manufacturing goods than they export. Manufacturing goods in the exports of these countries account for less than 75% of gross exports. The second group consists of countries that export more manufacturing goods than they import. Manufacturing goods in the exports of these countries account for more than 75% of gross exports. He finds that countries of the second group do not have declining shares of manufacturing employment and output, indicating that these countries have a comparative advantage in manufacturing (Rodrik, 2016).
11 All these studies suggest that one of the drivers behind the differences in deindustrialization patterns is the comparative advantage of a country in manufacturing. Hence, they take into account the amount of manufacturing exports in total exports. However, considering comparative advantages with regard to manufacturing exports is less relevant due to countries’ large involvement in intermediate goods trade. They import intermediates, add value and exports these goods again. Exports takes the value of both the imported intermediates and the added value of a country. They overestimate the value that a country adds to a product. Therefore, not a country’s exports, but a country’s involvement in manufacturing global value chains must be assessed to study the true comparative advantage of a countries in manufacturing (Timmer et al., 2013).
Stöllinger (2016) does this, but for European Union (EU) countries. He takes the change in manufacturing employment as a rough indicator of structural change and the global value chain participation by Koopman et al. (2014), and vertical specialization by Hummels et al. (2001) as drivers of this process. The main goal is to find out whether integration into global value chains has a different effect on structural change in Central European countries (seen as the manufacturing core) than in other EU member states. The findings indicate that this is indeed the case. Global value chain integration has a positive influence on the change in the share of manufacturing value added in the manufacturing core. In other EU countries integration has had a negative influence on structural change.
Global value chains and the manufacturing sector
The literature on the consequences of global value chains for deindustrialization focuses most on developed countries. Moreover, these studies consider the concept of offshoring instead of the direct link between global value chain participation and industrialization. However, Kummritz (2016) makes a contribution in this regard. He studies the consequences of global value chain participation on productivity growth and domestic value added for 63 countries. In contrast, the literature on global value chains and manufacturing jobs remains rather thin and non-conclusive. Moreover, it lacks empirical evidence. To provide a better understanding of the underlying mechanisms, this section briefly discusses the changing meaning of comparative advantage, offshoring and the different activities along the global value chain.
12 of classical trade theory is on the production of final goods. Countries specialize according to productivity differences (Ricardo) or factor abundance. The latter implies that a country specializes in producing the good that makes intensive use of the factor of production that is abundant in the country (Heckscher-Ohlin). However, due to a decline in trade costs and communication costs this no longer holds. Countries rather specialize in performing certain activities instead of whole production processes, which influences the structure of the economy (Baldwin, 2011).
As offshoring enables the creation of global value chains in which intermediate goods are traded, it is briefly discussed here. Offshoring entails “the contracting out of activities that were previously performed within a production unit to foreign subcontractors” (Foster-McGregor et al., 2013). The difference in the costs of the factors of production drives this process (Kohler, 2004; Hanson, Mataloni and Slaughter, 2005). Therefore, it is assumable that offshoring affects the labour market of both the offshoring country and the offshoring destination. An important implication is that all countries can specialize in the activity they perform most efficiently. This activity is a country’s “core capability” (Bernard et al., 2017).
13 Whether participation in global value chains creates manufacturing jobs or induces industrialization is difficult to assess. An increase in productivity due to the adoption of new technologies improves the productivity of the firms, resulting in less labour intensive processes. However, an increase in productivity is also very beneficial in attracting production activities. Furthermore, foreign supplies could replace domestic suppliers, which results in less employment opportunities. Therefore, it is dependent on the comparative advantage of the country, the type of sector and on the scale at which production takes place whether it actually results in more jobs (Farole, 2016; Taglioni & Winkler, 2016). For example, China exploits its comparative advantage in the labour-intensive apparel, leather and textile sectors. The scale at which production takes place in these sectors is large, offering many employment opportunities (Gereffi, 1999; Autor et al., 2013).
Although the literature has only just started to assess the effects of global value chain integration on domestic outcomes, studies on the measures of this phenomenon are more extensive. The next section elaborates on the different measures available and provides the justification of the chosen measure in this study.
Measures of global value chain participation
International product fragmentation demands new approaches to the analysis of trade and a country’s contribution to exports. Analysing gross exports is rather limited as it does not rightly capture the comparative advantage of a country. This has demanded increasing use of input-output analysis to arrive at decent measures of trade. This analysis provides the possibility to trace back the contribution of a particular sector to other domestic sectors or foreign sectors (Leontief, 1953; Timmer et al., 2015). In this way, measures of “value added in trade” and “trade in value added” can be calculated to assess a country’s global value chain integration (Stehrer, 2012).
14 sources its inputs from abroad. Exports in this case consider both intermediate and final goods. In addition, Hummels et al. (2001) also introduce a measure of forward linkages (VS1). It captures a country’s domestic value added in another country’s exports as share of its own exports. Therefore, the measure indicates the supply side of vertical specialization.
Koopman et al. (2014) emphasize the importance of this last measure of vertical specialization, as it takes into account the possibility that goods cross borders several times. The authors show the complicated nature of trade by breaking up gross exports in nine different components that could be contributed to either foreign or domestic value added. In the same study, a GVC participation rate is proposed that captures both forward and backward participation in GVCs: it sums up the foreign value added in exports (VS) and the domestic value added in another country’s export (VS1).
Considering trade in value added, Johnson and Noguera (2012) focus on the value added of bilateral trade, introducing the VAX ratio. The VAX ratio also calculates the domestic content in exports. They add to the existing literature regarding vertical specialization the consideration of intermediate and final goods that are not absorbed by final demand abroad as assumed by Hummels et al. (2001). The authors take into account the possibility that these goods could return home to satisfy domestic final demand.
This analysis employs the measure of vertical specialization (VS) by Hummels et al. (2001) in determining a country’s involvement in global value chains. As forward linkages are mostly associated with country’s involvement in higher value added activities, backward linkages seem to be more appropriate when discussing deindustrialization with regard to manufacturing jobs (Kummritz, 2016).
Hypothesis
15
H1: Vertical specialization within manufacturing global value chains has a
heterogeneous effect on industrialization in Asia and Latin America.
This paper is the first to seek an explanation in global value chain participation to address the different industrialization paths of Asia and Latin America.
3. Data and descriptive statistics
This chapter discusses the data and descriptive statistics of this paper. First, it discusses the dependent variable and the explanatory variable of interest. Second, it presents the control variables. Third, it reviews the descriptive statistics of the sample. Fourth, descriptive evidence shows the differences between Asia and Latin America with regard to industrialization and global value chain participation.
3.1 Data and data sources Measuring industrialization
The dependent variable of this analysis is industrialization. It follows other studies in quantifying industrialization. Therefore, it employs the manufacturing employment share of total employment as a measurement of industrialization (Rowthorn & Ramaswamy; 1999; Kang & Lee, 2011; Felipe et al., 2014; Rodrik, 2016; Stöllinger, 2016; Haraguchi, 2017).
Manufacturing employment as a share of total employment is composed from different sources. Data for 29 countries is collected from the 10-sector database of the Groningen Growth and Development Centre (Timmer, de Vries and de Vries, 2015). Data for 21 countries is collected from the OECD’s Structural Analysis database (OECD, STAN database) and data for 11 countries is collected from the ILOSTAT database (ILO, ILOSTAT database). Due to the lack of data on manufacturing employment in other databases for Cambodia, this is taken from the Asian Development Bank that gathers data from national accounts (SDBS, Asian Development Bank).
16 employment at a higher level of aggregation, which is the section level (United Nations Statistical Division, 2002). However, for some countries, the measurement of manufacturing employment is according to ISIC revision 4. This is due to the application of the new revision by the ILO since 2007. Therefore, from 2007 onwards, data on manufacturing employment for Bulgaria, Latvia, Lithuania and Malta is according to ISIC Revision 4 in which the manufacturing sector constitutes sector D. Comparing ISIC Revision 3 and ISIC Revision 4, the changes with regard to the manufacturing sector are small as well. More divisions are added and two divisions are removed from the section of manufacturing and placed under other sections.1 On a country level, it is assumed that, this does not make a large difference in determining the amount of people in the aggregate manufacturing sector (United Nations Statistical Division, 2008).
Measuring vertical specialization
The explanatory variable of interest is the vertical specialization rate. This study employs the first specialization measure as introduced by Hummels et al. (2001): VS. The following equation (1) clarifies the concept:
𝑉𝑆𝑐,𝑖 =
𝑖𝑚𝑝𝑜𝑟𝑡𝑒𝑑 𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒𝑠
𝑔𝑟𝑜𝑠𝑠 𝑜𝑢𝑡𝑝𝑢𝑡 ∗ 𝑔𝑟𝑜𝑠𝑠 𝑒𝑥𝑝𝑜𝑟𝑡𝑠 (1)
For this analysis specifically, VS indicates the imported intermediates by the manufacturing sector i in country c. This value is divided by the output of the manufacturing sector. After this, it is multiplied by the gross exports of the manufacturing sector. Therefore, it is a measure of foreign value added in exports and it indicates the backward linkages in global value chains. Data for this measure is derived from the Trade in Value Added (TiVA) database. It is available for 63 countries over the period 1995-2011. It is a joined project of the WTO and the OECD (2018). This initiative includes more Asian and Latin American countries than other initiatives, such as the World Input Output Database (WIOD) (Timmer et al., 2015).
1 Division 22 (Publishing, printing and reproduction of recorded media) and division 37 (Recycling) are removed
17
Control variables
The overview of all the control variables and their data source is provided here. As this analysis uses two different models, not all control variables are used in both. The method section elaborates on this further.
The first control variable is GDP per capita expressed in 2011 US dollars, which also enters the regression in quadratic form. Including GDP per capita in quadratic form as well is necessary to account for the stage of development of a country. The quadratic form provides the inverted U-shape discussed before. It controls for the observation that manufacturing as a share of total employment and as a share of GDP first increases and then decreases again when a country develops. Data of GDP per capita is from the Maddison Project Database of the Groningen Growth and Development Centre (Bolt et al., 2018). The second control variable is population in absolute numbers to control for demographic trends that may influence the composition of the labour force. Data is also from the Maddison Project Database of the Groningen Growth and Development Centre (Bolt et al., 2018).
3.2 Descriptive statistics
This section discusses the descriptive statistics to provide a better picture of the sample. The analysis in this paper uses two different models. Table 1 provides the descriptive statistics of model 1 and Table 2 provides the descriptive statistics of model 2. FVAX indicates the vertical specialization measure of backward linkages: the foreign value added in gross exports.
Table 1: Descriptive statistics of the first model
Observations Mean Std. Dev. Min. Max.
manemp 950 0.16 0.05 0.02 0.31
Log GDP per capita 1,054 9.78 0.82 7.05 11.31
Log GDP per capita – squared 1,054 96.31 15.55 49.6 127.99 Log population 1,054 9.73 1.71 5.59 14.11 Log population – squared 1,054 97.53 33.69 31.22 198.99 FVAX 1,054 0.34 0.12 0.07 0.69
18 The average manufacturing employment share (manemp) is 16 per cent. Examples of countries with low shares are Cambodia, Hong Kong and Saudi Arabia. The low manufacturing shares of the first country are attributable to the level of development of the country in 1995. The shift to activities with higher value added explains the low shares of Hong Kong. These activities require less labour, on average. The low shares of Saudi Arabia are due to the resources richness of the country, which is why there has been no need for the country to develop a manufacturing base. Logically, the high shares of manufacturing employment are attributable to countries that have or have had strong manufacturing sectors. Examples of these countries are the Czech Republic, Slovenia and Taiwan.
The average country in the sample has 34 per cent foreign value added in its exports (FVAX). Considering that values range from 7.9 per cent to 68.9 percent, large differences exist between countries. However, due to the likelihood that every country got integrated more in global value chains between 1995 and 2011, the differences are also expected due to small shares in the beginning of the period and larger shares at the end of the period. Indeed, examples of countries with small shares are attributable to Argentina, Italy and Romania in the beginning of the period. However, the largest shares of the sample are attributable to Latvia, Lithuania and Canada. These countries have had large shares over the whole period under consideration. This indicates that these countries source many inputs from abroad to produce their intermediate and final goods for exports.
The log of GDP per capita is quite dispersed, explained partly by the different development stages of the countries in the sample and partly by the possession of natural resources. Cambodia, Indonesia and Vietnam have the lowest levels of GDP per capita. Norway, Saudi Arabia and Luxembourg have the highest level of GDP per capita.
The log of population is even more dispersed, explained by the large heterogeneity in the sample. The sample includes both islands and large populous countries. Examples of small countries are Iceland, Malta and Luxembourg. Examples of populous countries are China, the United States and Indonesia.
19 the dependent variable, the first period (1996-1999) has 53 observations, the second period (2000-2003) has 58 observations, the third (2004-2007) has 59 observations and the fourth period (2008-2011) has 60 observations.
Table 2: Descriptive statistics of the second model
N Mean Std. Dev. Min. Max.
Δ Manempshare 230 -0.009 0.012 -0.045 0.046
Initempsharet*-1 229 0.17 0.05 0.02 0.31
Log GDP per capitat*-1 244 9.68 0.78 7.36 11.06
Log GDP per capitat*-1 – squared
244 94.32 14.72 54.11 122.33
FVAXt-1 244 0.33 0.12 0.08 0.65
Notes: this table shows the number of observations, the mean, the standard deviation, the minimum and the maximum values for the variables included in model 2 (N=61, T=5): the change in the manufacturing employment share between period t and period t - 1, manufacturing employment share in the first year of period t* – 1, GDP per capita (log form) and GDP per capita squared (log form) in the first year of period t* – 1, and foreign value added in exports of period t-1. Since the absolute changes in the share of manufacturing are very small, their values are provided with three decimals instead of two.
The dependent variable of the second model is the absolute change in manufacturing employment between period t and t-1. Due to missing values for certain years, the absolute changes of the manufacturing share between periods are biased upwards. To avoid the attribution of large changes to certain countries, these observations are removed. For example, if a country has values for 1996 and 1997 and for 2002 and 2003, the difference between these two periods is likely to be larger than when all years are present. This is the case for different periods for the following countries: Australia, Cambodia, Cyprus, Croatia, Lithuania, Latvia, Malta, Saudi Arabia, Turkey and Vietnam. Due to missing values, Tunisia has to be dropped in its entirety as no single difference in employment shares between periods could be rightfully calculated.
20 between the periods 1996-1999 and 2000-2003. It increased with 0.046 percentage points. China also saw large increases in its employment, it changed with 0.028 percentage points between the periods 2004-2007 and 2008-2011.
To detect outliers, several procedures are used. A residual versus leverage plot is analysed to visually determine whether outliers are present. Figure 1 shows this plot. Two observations of Taiwan seem to be a bit different from the other observations, but their leverage is small. To further study these two observations, a Cook’s distance test is performed, which identifies several outliers based on their leverage and residual values (Cook, 1977). The Cook’s distance test also detects possible outliers for Taiwan for the years 2009 and 2010. The Cook’s distance for these two observations diverges from the other values. They are relatively larger, but their absolute values do not seem concerning.
When assessing the observations of this country, no irregularities can be observed. Taiwan shows one of the highest shares of manufacturing employment, which for this country is not
21 unexpected. Moreover, the leverage of these points is very small and it is therefore unlikely to influence the estimates. However, to be sure they do not affects the results, the observations are dropped in the estimation to assess the difference.
The final sample of the first model includes 62 countries and 950 observations over the period 1995-2011. In the second model, no outliers are detected. Therefore, the final sample of the second model includes 60 countries and 229 observations over the period 1995-2011.
3.3 Descriptive evidence
As has become evident from the existing literature, Asia and Latin America differ in their industrialization patterns. This section provides descriptive evidence for these differences, assessing the changes in the manufacturing shares for employment. Moreover, it assesses the trends of backward linkages for both regions over the period 1995-2011.
Figure 2 presents the change in the share of manufacturing employment for countries in Asia and Latin America between 1995 and 2011. The twelve Asian countries included are Cambodia, China, Hong Kong, India, Indonesia, Malaysia, the Philippines, Singapore, South Korea, Thailand, Taiwan and Vietnam. Given that Japan is one of the early industrializers, it is excluded from this group. The seven Latin American countries included are Argentina, Brazil, Chile, Colombia, Costa Rica, Mexico and Peru.
From Figure 2 it is most striking that no single country in Latin America has increased its manufacturing share between 1995 and 2011. In addition, it is clear that heterogeneity among Asian countries is greater than among the Latin American countries. First, the share of manufacturing employment in 1995 is quite different for the Asian countries. Second, changes in the share of manufacturing employment also differ more between Asian countries.
22 American countries, it is interesting that Chile and Costa Rica, both among the top three of manufacturing countries in the region in 1995, experienced the largest declines between 1995 and 2011. The countries even had larger shares than successful manufacturing countries like China and Thailand. Mexico is the best performer of the Latin American countries. Starting out with a share larger than 15 per cent in 1995 and only a small decline over the period, it became the country with the largest manufacturing sector of the economy in the region in 2011.
This paper now turns to the observations regarding backward linkages. Figure 3 shows the average trends of both regions. Indeed, the average of backward shares of both regions increased over the period. However, the average of Asia is larger in 1995 and this difference continues to hold between 1995 and 2011. The drop after 2008 signifies the financial crisis.
Figure 2: the change in the share of manufacturing employment between 1995 and 2011
23
Figure 3: the region-average of backward linkages
It is clear that differences exist between Asia and Latin America, it is yet to be determined whether these differences also make sense statistically. The next chapter presents the econometric models.
4. Methodology
Two different econometric models assess the relation between vertical specialization and industrialization patterns within countries. First, this paper wants to explore the relation between vertical specialization and the level of industrialization. Second, it is interested in whether vertical specialization induces industrialization. The two models use a different dependent variable. In the first model, the measure of deindustrialization is the share of manufacturing employment in total employment. Therefore, it can be considered as the level of industrialization. In the second model, following Stöllinger (2016), the dependent variable is the absolute change in manufacturing employment. Since this is a change, it can be considered as a measure of the change in the level of industrialization.
0 5 10 15 20 25 30 35 40 45 50 V ertic al spe cializ ati on ra te (% ) Backward linkages in 1995-2011
Asia: backward Latin America: backward
24
4.1 The share of manufacturing employment in total employment
The following equation functions as the baseline regression and is estimated to assess the relation between backward linkages and the share of the manufacturing sector:
𝑚𝑎𝑛𝑒𝑚𝑝𝑖𝑡 = 𝛼 + 𝛽1𝑙𝑔𝑑𝑝𝑖𝑡+ 𝛽2𝑙𝑔𝑑𝑝𝑠𝑞𝑢𝑎𝑟𝑒𝑑𝑖𝑡+ 𝛽3𝑙𝑝𝑜𝑝𝑖𝑡+ 𝛽4𝑙𝑝𝑜𝑝𝑠𝑞𝑢𝑎𝑟𝑒𝑑𝑖𝑡
+ 𝛽5𝐹𝑉𝐴𝑋𝑖𝑡+ 𝛾𝑖+ 𝛿𝑡 + 𝜀𝑖𝑡 (2)
In equation (2), subscript i denotes country and subscript t denotes time where t is time in years with t = 1,2, …, T = 17. 𝛼 is the average individual effect. 𝛽𝑗 is the regression coefficient with
j = 1, 2, …, 5. 𝛾𝑖 is the deviation from the average individual specific effect for country i, 𝛿𝑡 is the time specific effect at year t and 𝜀𝑖𝑡 is the error term.
The dependent variable manemp indicates the share of manufacturing employment. The variables lgdp and lgdpsquared indicate GDP per capita in log form and the squared term of this log form. The squared term of GDP per capita allows for a curvilinear relationship between GDP per capita and the share of manufacturing employment. It is expected that the sign of the coefficient of this variable is negative, meaning that manufacturing as a share of employment increases first and then decreases again. This would give rise to the famous inverted u-shape. The variable lpop indicates the population of a country and enters the regression in log form and as a squared term as well, following the literature (Rodrik; 2016, Haraguchi; 2017). The variable FVAX denotes the vertical specialization measure. It is expected that the relation between the amount of backward linkages (FVAX) and the relative size of the manufacturing sector is positive.
However, the main interest of this analysis is whether global value chain participation can explain the difference in deindustrialization patterns between Asia and Latin America. Therefore, the estimation of equation (3) is important:
𝑚𝑎𝑛𝑒𝑚𝑝𝑖𝑡 = 𝛼 + 𝛽1𝑙𝑔𝑑𝑝𝑖𝑡+ 𝛽2𝑙𝑔𝑑𝑝𝑠𝑞𝑢𝑎𝑟𝑒𝑑𝑖𝑡 + 𝛽3𝑙𝑝𝑜𝑝𝑖𝑡
+ 𝛽4 𝑙𝑝𝑜𝑝𝑠𝑞𝑢𝑎𝑟𝑒𝑑𝑖𝑡 + 𝛽5𝐹𝑉𝐴𝑋𝑖𝑡+ 𝛽6𝐹𝑉𝐴𝑋𝑖𝑡∗ 𝐿𝑎𝑡𝑖𝑛 𝐴𝑚𝑒𝑟𝑖𝑐𝑎
+ 𝛽7𝐹𝑉𝐴𝑋𝑖𝑡∗ 𝑅𝑜𝑊 + 𝛾𝑖+ 𝛿𝑡+ 𝜀𝑖𝑡
25 In equation (3), subscript i denotes country and subscript t denotes time where t is time in years with t = 1,2, …, T = 17. 𝛼 is the average individual effect. 𝛽𝑗 is the regression coefficient with j = 1, 2, …, 7. 𝛾𝑖 is the deviation from the average individual specific effect for country i, 𝛿𝑡 is the time specific effect at year t and 𝜀𝑖𝑡 is the error term.
Including region dummies allows the relation between vertical specialization and the level of industrialization to be different for the groups in the analysis. The reference group in this estimation is Asia2. Including Asia as the reference group allows for a direct comparison between this region and Latin America. The other countries in the sample are grouped under
RoW, which indicates the “rest of the world”. It is expected that the coefficient of the reference
group Asia is larger than for the other two groups included.
The problem with the estimation of this model is that it is subject to endogeneity issues. In this case, two variables have potential reverse causality problems. Therefore, the estimates of the coefficients are biased and inconsistent. Moreover, the true causal link cannot be established. First, GDP per capita and hence the level of development is related to the share of manufacturing in an economy. At lower levels of income, the manufacturing sector increases. At higher levels of income, the manufacturing sector declines. However, an increase in the manufacturing sector also influences GDP per capita. As people move to the more productive sector of the economy, their income increases. Second, this analysis studies the relation between backward linkages in manufacturing global value chains and the share of manufacturing employment. Given the tradable character of the sector, it could be that countries with an increase in their manufacturing sector need increasingly more inputs to generate output. If they source these from abroad, their backward linkages increases.
To address these issues, one can rely on the use of instruments. An instrument has to be exogenous, theoretical relevant and it should not influence the dependent variable directly. It is very difficult to find a good instrument for the level of development. Kummritz (2016) implements an instrument to address the endogeneity problem of the main explanatory variable vertical specialization. He combines bilateral trade flows of third countries and the distance
2 The countries included in the reference group Asia: Cambodia, China, Hong Kong, India, Indonesia, Malaysia,
26 between industries to compose an instrument. This is considered to be beyond the scope of this paper. The next model addresses the problem of endogeneity by using lagged variables.
4.2 The absolute change in the share of manufacturing employment
The second model of this analysis takes a rough structural change approach, following Stöllinger (2016). Therefore, it assesses whether vertical specialization induces industrialization. It is different from the first model in three ways. First, the dependent variable is the absolute change in the share of manufacturing employment. Second, the timeframe is divided into periods instead of years. Third, it uses lagged variables. The drawback of this model is that taking periods instead of years severely reduces the amount of observations. The model is presented by equation (4):
∆𝑚𝑎𝑛𝑒𝑚𝑝𝑖𝑡 = 𝛼 + 𝛽1 𝑖𝑛𝑖𝑡𝑠ℎ𝑎𝑟𝑒𝑖𝑡∗−1+ 𝛽2𝑙𝑔𝑑𝑝𝑖𝑡∗−1+ 𝛽3𝑙𝑔𝑑𝑝2𝑖𝑡∗−1
+ 𝛽4𝐹𝑉𝐴𝑋𝑖𝑡−1+ 𝛽5𝐹𝑉𝐴𝑋𝑖𝑡−1∗ 𝐿𝑎𝑡𝑖𝑛𝐴𝑚𝑒𝑟𝑖𝑐𝑎 + 𝛽6𝐹𝑉𝐴𝑋𝑖𝑡−1 ∗ 𝑅𝑜𝑊 + 𝛿𝑡+ 𝛾𝑖+ 𝜀𝑖𝑡
(4)
In equation (4), the subscript i denotes country. The subscript t is period 1,2,3,4,5 where period 1 is 1995, period 2 is 1996-1999, period 3 is 2000-2003, period 4 is 2004-2007 and period 5 is 2008-2011. The subscript t* = 1,2,3,4 where t*=1 represents the year 1995, t*=2 represents the year 1996, t*=3 represents the year 2000 and t*=4 represents the year 2004. 𝛼 is the average individual effect. 𝛽𝑗 is the regression coefficient with j = 1, 2, …, 6. 𝛾𝑖 is the deviation from the individual specific effect for country i, 𝛿𝑡 is the time specific effect at period t and 𝜀𝑖𝑡 is the error term.
The variables with subscript t are the averages over that period. These variables are the dependent variable manemp and the independent variable of interest FVAX. The dependent variable now constitutes the absolute change in the share of manufacturing employment between the different periods: ∆𝑚𝑎𝑛𝑒𝑚𝑝𝑖𝑡 = 𝑚𝑎𝑛𝑒𝑚𝑝𝑖𝑡 - 𝑚𝑎𝑛𝑒𝑚𝑝𝑖𝑡. The independent
27 average of four years. However, this analysis gives preference to the measurement of a structural effect.
Lagged variables are used due to the interest in the causal effects of the variables. This model is subject to the same problems of endogeneity as the first model. Including lagged variables rules out the direct effect between vertical specialization and the change in manufacturing employment. To be more precise, the change in manufacturing employment between period t and t-1 cannot directly influence vertical specialization rate in period t.
The variables with subscript t* are one year observations and capture the initial conditions. The values of the share of manufacturing employment and GDP per capita do not change considerably between different years. Therefore, to capture the real initial conditions, the first year of the preceding period is chosen. Initial conditions included in the model are the initial share of manufacturing employment initshare and the initial level of development in log form,
lgdp. Again, the log of GDP per capita is included in its squared term: lgdp2.
The initial share of manufacturing employment (initshare) is expected to have a negative coefficient. This captures the findings that countries are deindustrializing, on average. The squared term of GDP per capita is expected to have a negative coefficient, controlling for the level of development. It is expected that the vertical specialization (FVAX) induces industrialization and that this effect is different for Asia and Latin America.
The models in this analysis are estimated using fixed effects, which is determined by the Hausman test for panel data. The Hausman tests indicates whether fixed effects or random effects should be incorporated (Hausman, 1978). Furthermore, heteroscedasticity is detected in both models by plotting residuals and running the Breusch-Pagan test (Breusch & Pagan, 1979). Therefore, robust standard errors are included, which controls for this. The Wooldridge test identifies autocorrelation in both models. The latter problem demands clustered robust standard errors.
5. Results
5.1 Estimation and results of the first model
28 variables have the expected signs. Indeed, the relation between GDP per capita and the share of manufacturing employment is curvilinear.
Table 3: the share of manufacturing employment as dependent variable
VARIABLES Pooled OLS Pooled OLS
Interaction Fixed Effects Fixed Effects Interaction
Log of GDP per capita 0.488*** 0.586*** 0.325*** 0.322***
(0.0260) (0.0253) (0.0750) (0.0703)
Log GDP per capita – squared -0.0255*** -0.0308*** -0.0158*** -0.0157*** (0.00140) (0.00135) (0.00394) (0.00367)
Log population -0.00707 -0.00153 -0.215* -0.207*
(0.00578) (0.00557) (0.109) (0.108)
Log population - squared 0.000205 -8.49e-05 0.0119** 0.0113**
(0.000291) (0.000275) (0.00555) (0.00564)
FVAX -0.0679*** -0.0379** 0.0447* 0.106
(0.0104) (0.0192) (0.0230) (0.0674)
FVAX * Latin America -0.164*** -0.101
(0.0162) (0.0895) FVAX * RoW -0.0211 -0.0759 (0.0169) (0.0668) Observations 950 950 950 950 R-squared 0.262 0.350 0.645 0.653 Number of countries 62 62 62 62
Country Fixed Effects No No Yes Yes
Time Fixed Effects No No Yes Yes
29 Considering the pooled OLS estimation, the results identify a negative relation between vertical specialization (FVAX) and the share of manufacturing employment. This implies that countries with more backward linkages, e.g. more foreign intermediate goods in a country’s exports, have a relative smaller manufacturing sector. After the inclusion of region dummies, the results indicate a negative relation between vertical specialization and the relative size of the manufacturing sector in Asia. Moreover, the negative relation between vertical specialization and the relative size of the manufacturing is larger for Latin American countries. However, pooled OLS does not consider the time and group structure of the sample. Therefore, unobserved individual characteristics end up in the error term. This leads to an omitted variable bias. Including fixed effects accounts for these differences between individuals that do not change over time. Individual intercepts capture these differences now. Examples of unobserved individual heterogeneity regarding countries are history, geography and national policies.
The inclusion of fixed effects is evident in the difference of the R-squared between the pooled OLS model and the fixed effects model. It increases considerably. However, the results after the inclusion of fixed effects do not support the hypothesis. First, backward linkages in manufacturing global value chains are not related to the share of manufacturing employment in a country. Second, the relation is not different for Asia and Latin America. To be sure that the outliers do not affect the results, the two data points of Taiwan are excluded. As expected, the outliers do not change the results, due to their low leverage.
The descriptive evidence identifies a serious drop in backward linkages after 2008, which indicates that the financial crisis had a large impact on trade. As this is likely to influence the results of the estimations, the first model is estimated for the period 1995-2008. Therefore, it excludes the post-crisis years of 2009, 2010 and 2011. Table 4 presents the results.
30 To assess the differences between Asia and Latin America, region dummies are included in the estimation. The relation between vertical specialization and the relative size of the manufacturing sector is positive for Asian countries. In Asia, a ten-percentage point higher value of vertical specialization (FVAX) is associated with a share of manufacturing that is, 1.6 percentage point larger, on average. Therefore, a country with more backward linkages has a significant higher share of the manufacturing sector in the economy in this region. However, the results do not support the hypothesis that the relation between intermediate goods trade and
Table 4: the share of manufacturing employment as dependent variable (post crisis period excluded)
VARIABLES Fixed Effects Fixed Effects
Interaction
Log GDP per capita 0.308*** 0.299***
(0.0790) (0.0684)
LOG GDP per capita – squared -0.0149*** -0.0145***
(0.00417) (0.00360)
Log population -0.249** -0.231**
(0.122) (0.112)
Log population squared 0.0135** 0.0120**
(0.00591) (0.00562)
FVAX 0.0557** 0.162***
(0.0234) (0.0557)
FVAX * Latin America -0.154*
(0.0781) FVAX * RoW -0.129** (0.0535) Observations 823 823 R-squared 0.592 0.615 Number of countries 62 62
Country Fixed Effects Yes Yes
Time Fixed Effects Yes Yes
31 the share of the manufacturing sector is different for Latin America. Although it is indicative of a considerable smaller effect for Latin America, the results are only marginal significant. Therefore, the right interpretation of the results is that a larger involvement in backward linkages has no relation with the level of industrialization in Latin America.
In addition, an increase in backward linkages in the “rest of the world” is associated with a larger share of the manufacturing sector in a country’s economy. However, this effect is considerably smaller than for the Asian countries. In the “rest of the world”, a ten-percentage point higher value of vertical specialization means that the share of manufacturing is 0.33 percentage points larger.
The interpretation of these results require caution. One cannot infer that an increase in vertical specialization increases the manufacturing sector in a country. The results only indicate that countries in Asia with more backward linkages have significantly higher shares of manufacturing employment. It is still possible that countries with larger manufacturing sectors source more inputs from abroad.
5.2 Estimation and results of the second model
Table 5 presents the results of the estimation of the second model. As explained, the dependent variable is the absolute change in the share of manufacturing employment, it considers four periods instead of years and it uses lagged independent variables.
The results of the second model indicate that vertical specialization (FVAX) has no effect on industrialization. To be more precise, more backward linkages in global value chains, e.g. more foreign intermediate inputs in the manufacturing exports of a country, does not induce industrialization. Furthermore, the difference between Asia and Latin America is not statistically significant.
32 period, the estimation of this model can be improved. It could incorporate both the structural effect of four years and more observations per country. Second, the estimation of the
this model lacks important control variables. This is due the unavailability of data for certain developing countries in the sample. Since the difference in the costs of production mainly drives the relocation of production activities, the addition of control variables that measure this, is important. The ILO has recently started to collect data on the hourly labour costs in the manufacturing sector for a large amount and diverse group of countries. These kinds of initiatives can improve the estimations in the future.
Table 5: the absolute change in the share of manufacturing as a dependent variable
VARIABLES Pooled OLS Pooled OLS
Interaction Fixed Effects Fixed Effects Interaction Initial empsharet*-1 -0.00939 -0.0126 -0.482*** -0.481*** (0.0163) (0.0175) (0.0576) (0.0661)
Log of GDP per capitat*-1 -0.0396** -0.0342 0.160** 0.162**
(0.0199) (0.0224) (0.0737) (0.0750)
Log of GDP per capita – 0.00182* 0.00154 -0.00867** -0.00883**
squaredt*-1 (0.00105) (0.00118) (0.00373) (0.00381)
FVAXt-1 -0.00251 1.16e-05 0.0321 0.0287
(0.00571) (0.0112) (0.0230) (0.0363)
FVAXt-1 * Latin America -0.00559 -0.0266
(0.0100) (0.0490) FVAXt-1 * RoW -0.00260 0.00960 (0.00930) (0.0414) Observations 226 226 226 226 R-squared 0.222 0.224 0.319 0.321 Number of countries 61 61 61 61
Country Fixed Effects No No Yes Yes
Time Fixed Effects No No Yes Yes
33
5.3 Robustness check
The estimation of the first model with the post-crisis period excluded delivers significant results. Countries in Asia with larger backward linkages have larger relative manufacturing sectors. To see whether these results hold under different specifications, this section performs different robustness checks. The first check replaces manufacturing employment by manufacturing value added. The second check excludes the Asian countries one by one from the estimation to see whether individual countries drive the results.
Data for manufacturing value added is from the Industrial Statistics Database (UNIDO, 2017). The replacement of manufacturing employment by manufacturing value added does not change the results, as indicated by column (1) of Table 6. The effect of intermediate goods trade on the share of manufacturing value added is positive and significant for Asia. Again, the relation between backward linkages and the share of manufacturing value added in Latin America is not statistically different from Asia. Moreover, the effect on the share of manufacturing in the “rest of the world” is still significantly different and smaller from the effect of Asia.
Dropping the Asian countries one by one provides interesting insights. First, when all the observations of Taiwan are removed, the significance of the effect for Asia turns from a one per cent significance level to a ten per cent significance level. Column (2) of Table 6 indicates this. The effect for both other regions turn entirely insignificant. Therefore, an individual country drives the significance of the results. Second, excluding either China, India or Indonesia from the observations, the effect of intermediate goods trade on the share of manufacturing in Latin America becomes significant at the five per cent level. This indicates that the results are very sensitive to the countries included in the reference group. Third, the exclusion of all the remaining Asian countries does not change the results.
34
Table 6 Robustness check
(1) (2)
VARIABLES Fixed Effects Fixed effects
Log GDP per capita 0.176** 0.303***
(0.0814) (0.0692)
Log GDP per capita squared -0.00868** -0.0148***
(0.00427) (0.00365)
Log population 0.109 0.231**
(0.161) (0.114)
Log population squared -0.00520 0.0121**
(0.00821) (0.00575)
FVAX 0.184*** 0.148*
(0.0567) (0.0783)
FVAX * Latin America -0.0511 -0.141
(0.0803) (0.0921) FVAX * RoW -0.173*** -0.114 (0.0592) (0.0766) Observations 868 809 R-squared 0.310 0.619 Number of countries 62 61
Country Fixed Effects Yes Yes
Time Fixed Effects Yes Yes
35
6. Concluding remarks
This paper is the first to explore the relation between integration into global value chains and industrialization to explain the divergent industrialization paths of Asia and Latin America. Existing studies are suggestive of the role of production fragmentation in explaining the differences between industrialization patterns, but this had not been researched yet.
A panel analysis of 62 countries between 1995 and 2011 has provided interesting insights. First, this paper finds that more backward linkages in manufacturing global value chains are associated with larger manufacturing shares in countries. This relation is even larger for Asian countries. However, no evidence is found that the relation between backward linkages and the level of industrialization is different for Latin America. Therefore, the results are suggestive of a disproportionate gain in manufacturing shares in Asia due to the region’s integration into global value chains, but they cannot explain the differences between the two regions. However, the results of the effect for Asia are not robust. Excluding Taiwan turns the results insignificant.
Second, this paper finds that integration into global value chains by backward linkages does not induce industrialization. However, due to the short timeframe under consideration to assess structural changes and the lack of important control variables, it is expected that future research can be meaningful in this regard. The control variables have to capture the cost differences between countries in producing manufacturing goods that drive the relocation of activities. Furthermore, the problem of endogeneity has to be better addressed to identify a causal link between vertical specialization and industrialization. To continue, the level of aggregation in this analysis can be considered too high. The preferred level of analysis would be the sector level. This could provide interesting insights about which sectors provide the most opportunities to countries. Moreover, although it has discussed the importance of analysing production fragmentation and the corresponding specialization in activities instead of complete production process, this analysis focuses on a sector in general. It would also be interesting to study which type of occupations and skill groups within this sector benefit most from global value chain integration.
36 knowledge spillovers. However, the positive effects on domestic outcomes are likely to depend on domestic conditions. The effects may depend on the ability of the labour force to adapt quickly to different technologies and to work with the transferred knowledge from the lead firm. Therefore, a sufficient amount of human capital could increase the chances of integration into global value chains. Moreover, a previous study has shown that the flexibility of the labour market plays an important role in preventing the displacement of workers (McMillan et al., 2014). This is relevant, because increased global competition also leads to the exit of firms (Iacovone et al., 2013).
37
7. Bibliography
ADB (2018). Employed, Manufacturing. Asian Development Bank: Statistical Database
System (database), (accessed on 24 April 2018).
Autor, D., Dorn, D., & Hanson, G. (2013). The China Syndrome: Local Labor Market Effects of Import Competition in the United States. American Economic Review, 103(6), 2121-2168.
Balassa, B. (1978). Export incentives and export performance in developing countries: A comparative analysis. Review of World Economics, 114(1), 24-61
Baldwin, R. (2011). Trade and Industrialisation after Globalization’s 2nd Unbundling: how Building and Joining a Supply Chain Are Different and Why It Matters. NBER
Working Paper Series, (17716).
Blinder, A., & Krueger, A. (2013). Alternative Measures of Offshorability: A Survey Approach. Journal of Labor Economics, 31(S1), 97-128.
Breusch, T., & Pagan, A. (1979). A Simple Test for Heteroscedasticity and Random Coefficient Variation. Econometrica, 47(5), 1287-1294.
Clark, C. (1957). The Conditions of Economic Progress. London: Macmillan.
Cook, R. (1977). Detection of Influential Observation in Linear Regression. Technometrics,
19(1), 15-18.
Dasgupta, S., & Singh, A. (2006). Manufacturing, services and premature deindustrialization in developing countries: A Kaldorian Analysis. UNU-WIDER United Nations
Research Paper, (2006/49).
Dedrick, J., Kraemer, K., & Linden, G. (2010). Who Profits From Innovation in Global Value Chains? A study of the iPod and Notebook PCs. Industrial and Corporate
Change, 19(1), 81-116.
Felipe, J., Mehta, A., & Rhee, C. (2014). Manufacturing Matters...but it's the jobs that count. ADB Economic Working Paper Series, (420).
Felipe, J., Mehta, A., & Rhee, C. (2018). Manufacturing matters…but it’s the jobs that count. Cambridge Journal of Economics.
38 Foellimi, R., & Zweimüller, J. (2011). Exclusive goods and formal-sector employment.
American Economic Journal: Macroeconomics, 3(1), 242-272.
Foster-McGregor, N., Stehrer, R., & de Vries, G. (2013). Offshoring and the skill structure of labour demand. Review of World Economics, 149(4), 631-662.
Gereffi, G. (1999). International trade and industrial upgrading in the apparel commodity chain. Journal of International Economics, 48(1), 37-70.
Gereffi, G., & Fernandez-Stark, K. (2016). Global Value Chain Analysis: A Primer. Duke University, Global Value Chains (Center).
Gereffi, G., Humphrey, J., & Sturgeon, T. (2005). The governance of global value chains. Review of International Political Economy, 12(1), 78-104.
Ghani, E., & O'Connell, S. (2014). Can Service Be a Growth Escalator in Low Income Countries? Policy Research Working Paper, (6971).
Grabowski, R. (2017). Premature deindustrialization and inequality. International Journal of
Social Economics, 44(2), 154-168.
Greenstein, J., & Anderson, B. (2017). Premature Deindustrialization and the Defeminization of Labor. Journal of Economic Issues, 51(2), 446-457.
Iacovone, L., Rauch, F., & Winters, L. (2013). Trade as an engine of creative destruction: Mexican experience with Chinese competition. Journal of International Economics, 89(2), 379-392.
Hanson, G., Mataloni, R., & Slaughter, M. (2005). Vertical Production Networks in Multinational Firms. The Review of Economics and Statistics, 87(4), 664-678.
Haraguchi, N., Cheng, C., & Smeets, E. (2017). The Importance of Manufacturing in Economic Development: Has This Changed?. World Development, 93, 293-315. Hausman, J. (1978). Specification Tests in Econometrics. Econometrica, 46(6), 1251-1271. ILO (2018). Employment by sex and economic activity (thousands). ILO database of
labour statistics (database), (accessed on 20 March 2018).
39 Humphrey, J., & Schmitz, H. (2002). How Does Insertion in Global Value Chains Affect
Upgrading in Industrial Clusters?. Regional Studies, 36(9), 1017-1022.
Johnson, R., & Noguera, G. (2012). Proximity and Production Fragmentation. American
Economic Review, 102(3), 407-411.
Kaldor, N. (1967). Strategic factors in economic development. Ithaca, NY: Cornell University.
Kang, S., & Lee, H. (2011). Foreign Direct Investment and de-industrialisation. The World
Economy, 34(2), 313-329.
Koopman, R., Wang, Z., & Wei, S. (2014). Tracing Value-Added and Double Counting in Gross Exports. American Economic Review, 104(2), 459-494.
Kummritz, V. (2016). Do Global Value Chains Cause Industrial Development?. CTEI
Working Paper, (2016-01).
Leontief, W. (1953). Domestic Production and Foreign Trade; the American Capital Position Re-Examined. Proceedings of The American Philosophical Society, 97(4), 332-349. Lewis, W. (1954). Economic Development with Unlimited Supplies of Labour. The
Manchester School, 22(2), 139-191.
Maddison Project Database, version 2018. Bolt, Jutta, Robert Inklaar, Herman de Jong and Jan Luiten van Zanden (2018), “Rebasing ‘Maddison’: new income comparisons and the shape of long-run economic development”, Maddison Project Working paper 10. McMillan, M., Rodrik, D., & Verduzco-Gallo, Í. (2014). Globalization, Structural Change,
and Productivity Growth, with an Update on Africa. World Development, 63, 11-32. Marconi, N., Reis, C., & Araújo, E. (2016). Manufacturing and economic development: The
actuality of Kaldor’s first and second law. Structural Change and Economic
Dynamics, 37, 75-89.
Michaels, G., Natraj, A., & Van Reenen, J. (2014). Has ICT Polarized Skill Demand? Evidence from Eleven Countries over Twenty-Five Years. Review of Economics and
Statistics, 96(1), 60-77.
OECD (2018). Manufacturing share of employment 1970-2009. OECD: STAN Structural
