Uncovering the drivers of world-system mobility
A network analysis of commodity trade 1980-2016
Martine Postma, s2543214 Mutua Fidesstraat 17
9714 CB Groningen
m.g.postma@student.rug.nl
Supervised by: dr. D. Akkermans Co-assessor: M. Papakonstantinou
1
ABSTRACT
This thesis conducts a network analysis of commodity trade of 79 countries over the period 1980-2016. A measure of world-system mobility is constructed, and consequently regressed on the underlying composition of a country’s total exports that is based on a classification of commodities by technological intensity. Countries that successfully made movements upward in the identified network of commodity trade that makes up the world-system are the countries that successfully altered the composition of their total exports in favor of commodities produced at higher levels of technological intensity.
2
TABLE OF CONTENTS
ABSTRACT ... 1
1. INTRODUCTION ... 3
2. LITERATURE REVIEW ... 5
2.1 Capability building and economic development ... 5
2.2 The technological structure of a country’s commodity exports and economic development ... 6
2.3 The world-system, roles, positions and economic development ... 7
2.4 Hypotheses ... 8
2.4.1 Control variables ... 9
3. DATA ... 10
3.1 Network analysis and international trade ... 10
3.2 Empirical model and variables of interest ... 12
3.3 Data sources, data collection and final sample ... 13
3.4 Construction of variables ... 14
4. METHODS ... 16
4.1 Social network analysis – identification of roles and positions ... 16
4.1.1 Calculation of world-system mobility ... 18
4.2 Panel data analysis ... 19
4.2.1 Descriptive statistics ... 19
4.2.2 Regression assumptions... 20
5. RESULTS AND DISCUSSION ... 21
5.1 Network roles and positions: the structure of the international division of labor ... 21
5.2 Panel data analysis ... 26
3
1. INTRODUCTION
The global economic map is ever-changing. As Peter Dicken put it in his ‘Global Shift’: “old geographies of production, distribution and consumption are continuously being disrupted and new geographies are continuously being created. The new does not totally obliterate the old. On the contrary, there are complex processes of path dependency at work; what already exists constitutes the preconditions on which the new develops” (Dicken, 2011, pp.14). Dicken’s work provides a comprehensive analysis of the effects of globalization on people and places, thereby highlighting the interactions among players in the global economy, and their roles in shaping unequal development. Central to his text is the observation that over the course of history, a global division of labor evolved that has far reaching consequences for the wealth and poverty of nations. That is, increasing international trade and fragmentation of global value chains created significant opportunities for some countries to enter paths towards economic development and catching up with historically rich countries (i.e. the ‘manufacturing miracles’ in East Asia, and the ‘BRICS’), while there are also countries that remain heavily impoverished, and have seen none of such chances.
A stream of research that also focuses on the dynamics that come with globalization, further expanding trade ties and the consequences thereof on the wealth and poverty of nations is that of the world-system. The world-systems perspective has since its emergence based on the works of Immanuel Wallerstein dominated the sociological study of economic development. At the conceptual level, the world-systems view builds on the notion that the world economy follows an international division of labor. This international division of labor can be represented as a hierarchical system of states that occupy different positions and fulfill different roles within it. The intuition is the following: different types of countries engage in different patterns of commodity trade that determine their relative positions, and different roles in turn follow from the type of commodity that dominates their exports (Mahutga, 2006). Main themes that have been touched upon within the world-systems literature are the extent to which roles and positions in the international division of labor matter for development, and whether under the consequences of globalization, countries have been able to improve their positions, a process that is conceptualized as world-systems mobility. World-systems mobility is argued to follow from opening up to trade (Kim and Shin, 2001), and ascribed to efforts of developing countries to engage in “industrial upgrading” (Mahutga, 2006; Mahutga & Smith, 2011). An empirical analysis of uncovering the economic mechanisms that drive mobility, however, has always been left open as a topic for future research. This thesis aims to fill up this gap.
Building on the intuition that roles played in the international division of labor determine a country’s position therein, and that this has consequences for economic development, changing this role should offer an explanation for patterns of mobility. Therefore, the research question that will be answered is the following: to what extent does changing a country’s role played in
the international division of labor promote world-system mobility?
4 research question, and proposes that this is a process that is path-dependent, difficult to change, but crucial for economic development. In brief, upward mobility in the world-system is proposed to follow from going through processes of technology learning and capability building, and a consequent shift of the composition of total exports towards more high- and medium technology commodities.
Empirically, this paper builds on previous literature on identifying the international division of labor that makes up the world-system (Smith & White, 1992; Mahutga, 2006; Lloyd, Mahutga & de Leeuw, 2009; Mahutga & Smith, 2011). Taking these studies as a foundation, a network analysis of total commodity trade over the period 1980-2016 is conducted. This gives a continuous representation of the structure of the international division of labor, to accordingly construct a measure of world-system mobility that captures movement within this structure. This measure of world-system mobility is then regressed on the underlying composition of total commodity trade by technological intensities as identified by Lall (2000b).
Findings indicate that enlarging the share of commodities that are produced under more sophisticated technologies in total exports fosters upward mobility in the world-system. No significant support is found for the opposite effect of producing and exporting commodities on the lower-end spectrum of technological intensity. This paper concludes that those countries that successfully enter trajectories of technology learning and capability building, and so are able to change their role in the international division of labor by producing more technology-intensive commodities enjoy upward mobility in the world-system.
5
2. LITERATURE REVIEW
What constitutes economic development? Neoclassical models of international trade predict that under two factors of production (capital and labor), countries should produce and export according to their comparative advantage. Consequently, and if all countries do so, trade will be beneficial for all. Similarly, these models assume that every country has access to the same pool of technology. That is, international technology markets are by assumption efficient, meaning each firm in any country can access technology without additional costs or effort. In this context, economic development follows from open, unrestricted trade to be able to access mature, foreign technologies supplemented by qualitative, supportive institutions that protect intellectual property rights and provide education for example. This view is increasingly seen as oversimplified. In shining another light on what constitutes development, this section summarizes and combines three alternative views.
2.1 Capability building and economic development
The first relates to the review of Evenson & Westphal (1995), who argue that there are two assumptions made in neoclassical treatments on technological change in the process of economic development that are implausible. The first is that a technology consists of a set of techniques, or ‘blueprints’ that can be easily transferred. The second is that all technology comes from developed countries and flows to developing countries. This first assumption would mean that obtaining technologies requires no further costs and efforts, the second that developing countries lack the capabilities to adapt and modify technologies.
This view on technological change and economic development is further grounded the capability approach, that shows that there is a difference between capacity (e.g. physical technology, like machinery, blueprints) and capability (the ability to use physical technology efficiently). That is, the capability approach builds on the assumption that firms in developing countries have imperfect information on technological alternatives, which makes finding new technologies for production a time-consuming, insecure and costly process. Thereby, once technology is found and imported, developing countries often lack the necessary skills and knowledge to capture its tacit elements, where “tacit” technology is technology that cannot be easily codified and implemented. Comparative advantage and economic development as such depends to a greater extent on the ability of developing countries to foster technological learning rather than simply on relative factor endowments (Lall, 2000a).
6 leads to positive externalities for other entrepreneurs in the economy, who can then imitate them. The role of a country in this framework is to enable entrepreneurs to learn what to produce by stimulating the process of cost discovery. This way, countries are able to alter their patterns of specialization. Countries that fail to do so will likely experience delayed economic development.
In short, following the above perspectives, development follows from building an environment that fosters learning what can be produced, by enabling firms to capture the tacit elements of technologies. The ‘tacitness’ of technologies, however, differs greatly and by product.
2.2 The technological structure of a country’s commodity exports and economic development
The composition of a country’s export patterns has strong implications for economic development. Lall (2000b) observes that export performance by developing countries has been highly diverse over time. Some countries have successfully made the transition from low levels of development to becoming a rich country by aggressively increasing quantities of exports, as well as upgrading quality by shifting towards products with higher levels of technological intensity. Other countries have been stagnating for years, while there are also countries that rest in the middle with increasing quantities of exports, but no further diversification in terms of quality. Her reasoning departs from the notion that different export structures have different implications for economic growth and development. More technology-intensive commodities, for example, are highly income inelastic, create new demand and substitute faster than older products. Thereby, and related to the literature summarized above, they have greater potential for learning, and larger spillover effects that aid in building skills and generating knowledge that can then be applied in other activities. Simpler, more labor-intensive or ‘low technology’ commodities, on the other hand, offer far less learning- and spillover potential. They can, however, enjoy rapid quantity growth: technologies can be easily diffused, facilitating the shift from high to low wage areas. However, this is not a sustainable platform for growth and is more of a ‘once-for-all’ boost to exports that needs the be followed up by expanding into more technology-intensive activities. Building on the capabilities view on economic development, Lall (2000b) further argues that shifting export structures towards more technology-intensive commodities is the result of long trajectories of learning, agglomeration and institution-building, which makes moving from a low- to high-technology structure difficult, but crucial to economic development.
7 enjoy faster economic growth than those countries that do not. Following their models, growth is the outcome of switching from low productivity activities to high productivity activities. Examples of products (HS 6-digit) that require smallest productivity are vegetable products, or vanilla beans. Products that require highest productivity are ‘flat rolled iron or non-alloy steel, coated with aluminium’ and ‘tyre cord fabric of viscose rayon’. The conclusion of their analysis is that gains from globalization and international trade greatly depend on the ability of countries to position themselves at higher points along the spectrum that captures the quality of their export baskets, thus to shift towards producing and exporting products with higher technological intensities.
2.3 The world-system, roles, positions and economic development
The world-systems view to economic development was already shortly highlighted in the introduction. It departs from the notion that different countries have different perspectives for economic development based on their positions in the world-system, that follows from their patterns of commodity trade. Broadly speaking, these positions can be grouped into a core, periphery and semi-periphery. The world-system thus consists of a hierarchical organization of states where the core is most advanced, the periphery the least and the semi-periphery rests in the middle. The central idea is that each of these three zones takes on a different ‘role’ in the international division of labor, that reflects their level of- and barriers to development (Wallerstein, 1974, in: Lloyd, Mahutga & de Leeuw, 2009). Within the world-system, core nations that typically include the major economic powers of Western Europe, the United States and Japan dominate global economic networks. This global dominance is reflected in national economic diversity, wealth, and being at the forefront of technological advancements. Peripheral nations are argued to be dependent on- and disadvantaged relative to the core and to the semiperiphery. They suffer from domestic economic weaknesses, which manifest themselves in poverty, technological and educational shortfalls and commodity exports that consist greatly of primary products. The semi-periphery shares characteristics of both the core and the periphery, to the extent that it is argued to dominate the periphery, and to be dependent on the core. Countries in the semi-periphery often find themselves in processes of industrialization, paired with improvements in national institutions and human capital outcomes (Chase-Dunn, 1989, in: Kick & Davis, 2001).
8 Given that positions in the world-system are determinant to levels of- and barriers to development, positions are likely to change over time. Part of the world-systems view theory is the concept of mobility, where upward mobility results in higher positions in the calculated network of commodity trade.
Smith & White (1992) were the first to empirically assess world-system mobility. They conduct a network analysis of commodity trade flows, and find that over the period 1965-1980 there was considerable mobility upward mainly in what they identified as the semi-periphery, which they suggest to be attributed to the rise of a ‘new international division of labor’ that followed from the relocation of manufacturing activities out of the core. That is, mobility seems to be the result of starting patterns of deindustrialization in the core, and rapid rise of export manufacturing in some other countries, mainly the semi-periphery. Their results lend first support for the fact that indeed movements up to the higher spheres in the system are by prototypical industrializing countries like Brazil, Singapore and South Korea that increasingly concentrate on heavy industry, and that within the lower spheres, countries hardly experience movements upward and mainly focus on low technology, low wage manufacturing.
Mahutga & Smith (2011) aim to reveal how the international division of labor determines gains from participation therein, and find significant support for the notion that a country’s prospects for long-term development are a function of their positions in the international division of labor, as well as their patterns of mobility. They further show that there is convergence between the semi-periphery and the core in terms of their export patterns. That is, due to globalization, many countries in the semi-periphery moved from specialization in primary goods, to light manufacturing towards high technology manufacturing, which is consequently proposed to be a key mechanism explaining their rapid economic growth. At the same time, they find evidence for stagnation in the periphery whose exports remained to be dominated by primary or low-wage light manufactured goods They conclude that mobility in the world-system is a real path towards economic growth and development, and that its determinants should become subject for future research.
2.4 Hypotheses
The precedent overview of the literature presented three main views on barriers to economic development. The first two relate to the idea that countries that lack the skills and capabilities to master the ‘tacitness’ of foreign technologies, and consequently shift towards producing more high technology goods will experience delayed development. The third, and related, is that of the world-systems view that states that different positions follow from different roles played in the international division of labor (i.e. the type of commodities that a country exports) that are determinant to their levels of- and barriers to development, and that over time, countries can improve their positions and enjoy mobility.
9 entails shifting towards producing and exporting more technologically intensive commodities. The following hypotheses can be formulated:
hypothesis 1: shifting towards more high- and medium technology exported commodities
positively impacts world-systems mobility.
hypothesis 2: shifting towards more low technology and resource based exported
commodities negatively impacts world-systems mobility.
2.4.1 Control variables
10
3. DATA
This chapter will be structured as follows. First, a more detailed foundation of network analysis of commodity trade is warranted, as well as an introduction of the classification of commodities by technological intensity to identify the international division of labor that is often referred to (section 3.1). Then, based on the formulated hypotheses, the empirical model that will be tested as well as the included variables will be introduced (section 3.2), followed by data collection, data sources and specification of the final sample (section 3.3). After, construction of all variables will be explained (section 3.4).
3.1 Network analysis and international trade
11 For this study, positions within the world-system thus need to be derived based on total commodity trade data, that need to be subdivided into five different groups of commodities that reflect different levels of technological intensity to be able to determine whether countries in the higher spheres of the world-system indeed produce and export more high technology commodities, and to determine whether mobility will follow from changing the underlying composition of total exports in different commodities by technological intensity. In section 4.2, the step by step procedure of identifying world-system positions and calculating mobility will be provided.
Classification Examples
Primary products
Fresh fruit, meat, rice, cocoa, tea, coffee, timber, coal, crude petroleum, gas, ore concentrates and scrap
Manufactured products
Resource-based manufactures
Prepared meats/fruits, beverages, wood products, vegetable oils, base metals (except steal), cement
Low technology manufactures
Textile fabrics, clothing, footwear, leather manufactures, simple metal structures, furniture, toys
Medium technology manufactures
Passenger vehicles and parts, commercial vehicles, motorcycles and parts, synthetic fibres, chemicals
High technology manufactures
Data processing and telecommunications equipment, television sets, transistors, turbines
Table 1: Classification of commodities by technological intensity
Source: Lall (2000b)
What follows is a short explanation of each of the manufactured product groups that together represent the international division of labor (from: Lall, 2000b):
labor-12 intensive goods. The majority of the reported resource based manufactures arises from each individual country’s availability of natural resources, and usually do not require capital, scale, and skill-intensive technologies. Primary goods, finally do not require much further explanation. These as well stem in the vast majority from each country’s availability of natural resources and do not require any processing.
A final note is that the theory builds on the notion that different countries are capable of producing different commodities, which drives their outlooks for upward mobility. Construction of the main concepts, however, will be based on commodity exports, thereby implicitly assuming that exports are the same as production. Which is not the case. Within high technology commodities for example, electronics are heavily susceptible to having many stages of production being performed in lower wage regions due to their high value-to-weight ratios. The most classical example here is that of the iPod, where mainly design and R&D take place in the USA, and the rest of all production stages is dispersed globally. Result is that these do show up in the high technology exports of countries that add relatively few value (only perform the assembly stage). The two main limitations from relying on exports are the following (Lall, 2000b): (1) commodity exports cannot distinguish between differences in quality (e.g. handmade toys versus mass production toys); (2) commodity exports do not reveal underlying processes (e.g. high-tech, sophisticated processes versus simple assembly). These two limitations could be overcome using data on trade in value added that actually do reflect production. Despite of various initiatives to construct databases that contain data on value added, for now these data are not available for large samples of countries comprising a fair balance between developed and developing countries, covering a larger timespan. Thus, for practical reasons, and assuming that commodity exports still reasonably capture what a country is capable of producing (Hausmann et al., 2006), commodity trade data will be used.
3.2 Empirical model and variables of interest
Based on the precedent overview of the literature and hypotheses, as well as the above outlined research design, the model that will be estimated is the following:
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖 = 𝛽𝛽0+ 𝛽𝛽1∆ℎ𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖+ 𝛽𝛽2∆𝑚𝑚𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖−𝛽𝛽3∆𝑚𝑚𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖−𝛽𝛽4∆𝑟𝑟𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖−𝛽𝛽5∆𝑡𝑡𝑝𝑝𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖+
𝛽𝛽6∆ℎ𝑚𝑚𝑚𝑚𝑚𝑚𝑡𝑡𝑖𝑖𝑖𝑖+ 𝛽𝛽7∆𝑓𝑓𝑓𝑓𝑚𝑚𝑚𝑚𝑓𝑓𝑖𝑖𝑖𝑖+ 𝜀𝜀𝑖𝑖𝑖𝑖 (1)
where
i = country indicator k = time indicator1
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 = the change in distance to the USA in the world-system in t and t+1 divided by t ∆ℎ𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡 = difference in the share of high technology exports in total exports in time t+1 and t ∆𝑚𝑚𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡 = difference in the share of medium technology exports in total exports in t+1 and t
13 ∆𝑚𝑚𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡 = difference in the share of low technology exports in total exports in t+1 and t
∆𝑟𝑟𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡 = difference in the share of resource based manufactured exports in total exports in t+1 and t
∆𝑡𝑡𝑝𝑝𝑡𝑡𝑡𝑡𝑡𝑡 = difference in the share of primary goods exports in total exports in t+1 and t ∆ℎ𝑚𝑚𝑚𝑚𝑚𝑚𝑡𝑡 = difference in the share of high technology imports in total imports in t+1 and t ∆𝑓𝑓𝑓𝑓𝑚𝑚𝑚𝑚𝑓𝑓 = difference in FDI inflows expressed as a share of a country’s GDP in t+1 and t Shifting towards exporting more high- and medium technology commodities is expected to have a positive impact on mobility. Shifting towards exporting more low technology and resource based manufactured commodities is expected to have a negative effect on mobility. High technology imports and FDI inflows are expected to act as channels through which learning and technology upgrading takes place to be able to export more high- and medium technology commodities, thus to positively impact mobility. Measurement and calculation of all variables will be discussed in section 3.4.
3.3 Data sources, data collection and final sample
Data are collected over the period 1980 - 2016, at five different points (1980, 1990, 2000, 2010 and 2016). The chosen timeframe captures a period of increased relocation of great parts of developed countries’ manufacturing sector to developing countries, and hence very likely shifts in the composition of a country’s total exports.
Data on the composition of a country’s total exports are supplied by the United Nations Commodity Trade Statistics Database (Comtrade), classified conforming to the Standard International Trade Classification system, Revision 2 (SITC Rev.2). SITC groups commodities at different levels, ranging from 1-digit (aggregated, very general) to 5-digit (very specific). The different groupings of aggregation reflect the materials used for production, the processing stage, how the products are used, the importance of each of the commodities in terms of world trade, and technological change. Because the classification of Lall (2000b) is used, commodities belonging to each of the five categories at the 3-digit level are collected. Specifically, import data are used, because these are typically better measured than export data. Durand (1953) reports that the reason for this higher accuracy of import data is that countries are more tended to record their imports better for the purpose of tariffs. Important to bear in mind is thus that always reference is made to exports by a given country i to j, measured as imports by j from i. Using Revision 2, rather than the most recent and more detailed Revision 4 is simply to allow for comparison within the timespan under consideration.
14 industrializing countries like Brazil, China, Taiwan, Mexico, South Africa and Russia were not included yet as reporters. All other samples are thus restricted to that of reporting countries in 1980, but Brazil, China, Taiwan and Mexico are manually added to the matrices for 1980, relying on export data from the rest of the world to this group of countries. Not one country reported to have exported goods to South Africa in 1990, and to Russia in both 1980 and 1990, so these two were totally excluded. The sample obtained after manually adding these four countries is the one that will be applied for all further time slices. Because then still, there are countries that appeared as a reporter in 1980, but missed in some of the other time slices, these countries were added by taking the export values of other countries to these particular countries. This approach has as main disadvantage that the countries reported exports that are actually above or below the value that would have been reported by the country itself as imports. Also, due China’s ‘One Nation’-policy, data for Taiwan are not reported directly. The Comtrade database does include aggregated ‘Other Asia, n.e.s.’, and Lloyd et al. (2009) report that this group reasonably matches data annually reported by Taiwan, be it imperfect because it may include trade data from countries other than Taiwan. Taking this as sufficient support, Other Asia n.e.s. was coded to TWN2.
Taking into account the above described considerations, the final sample consists of 79 countries that is consistent over time. In total, the included countries are representative of all world regions, and contain a fair composition of developed and developing countries.
Data on FDI inflows as a share of GDP are retrieved from the World Bank. Data were not available for all 79 countries in the sample and for all points in time. There were a lot of missing data mainly in the poor countries included, but also on for example Taiwan and China, which restricts the amount of observations.
3.4 Construction of variables
Using the commodity trade data collected, six squire, symmetric, directed input output matrices are constructed that represent pairwise trade flows between reporters (columns) and partners (rows)3 for each year. Five of these six matrices represent the different categories of
technological intensity, and the remaining matrix consists of the totals of the five. The total trade matrix for each year will serve as input for the procedure to calculate mobility. The five matrices that that consist of the trade flows in the different categories will be used as input for calculating the independent variables.
Construction of Mobility follows from the continuous representation of the calculated network of total commodity trade, whose procedure is discussed step-by-step in section 4.1.1. At the conceptual level, mobility is the result of change in distance between an individual county and
2 See also: UN Trade Statistics (2010). Taiwan, Province of China Trade data (China, Data availability, Taiwan). Retrieved April 11, 2018, from https://unstats.un.org/unsd/tradekb/Knowledgebase/Taiwan-Province-of-China-Trade-data
15 the highest level of the network measured (in this study, the USA) at time t and time t+1, divided by the distance in time t (Mahutga & Smith, 2011).
To construct the composition of a country’s total exports, total exports are first calculated by taking the row sums of the input output matrix that represents the totals of the five commodity groups for each country in each individual year:
𝑚𝑚𝑚𝑚𝑚𝑚𝑡𝑡𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖 = ℎ𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖+ 𝑚𝑚𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖+ 𝑚𝑚𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖+ 𝑟𝑟𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖+ 𝑡𝑡𝑝𝑝𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖 (2)
Measurement of the share that each of the commodity groups makes up in total exports follows the same procedure for all five different commodity groups, as well as for high technology imports. The following equation shows how this is done for high technology exports:
∆ℎ𝑚𝑚𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖 = ∑ 𝑖𝑖𝑡𝑡𝑖𝑖𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡∑ ℎ𝑖𝑖𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖+1𝑖𝑖𝑖𝑖+1−∑ 𝑖𝑖𝑡𝑡𝑖𝑖𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡∑ ℎ𝑖𝑖𝑡𝑡𝑡𝑡𝑡𝑡𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 (3)
where subscript i denotes an individual country and subscript k denotes the period between t and t+1.
Data on FDI inflows already are supplied as a share of GDP. The change in FDI inflows between two points of time will be calculated as follows:
∆𝑓𝑓𝑓𝑓𝑚𝑚𝑚𝑚𝑓𝑓𝑖𝑖𝑖𝑖 = 𝑓𝑓𝑓𝑓𝑚𝑚𝑚𝑚𝑓𝑓𝑖𝑖𝑖𝑖+1− 𝑓𝑓𝑓𝑓𝑚𝑚𝑚𝑚𝑓𝑓𝑖𝑖𝑖𝑖 (4)
where again subscript i denotes an individual country and subscript k denotes the period between t and t+1.
16
4. METHODS
This section is subdivided into: (4.1) identification of network roles and positions and calculation of world-system mobility and (4.2) panel data analysis.
4.1 Social network analysis – identification of roles and positions
Lloyd et al. (2009) propose a methodology for identifying network roles and positions using social network analysis techniques. This methodology, that consists of three broad steps (see Figure 1, pp. 602), will be followed and executed in software package UCInet4, and visualized in NetDraw5. All steps will be thoroughly discussed in the proceeding pages.
The analysis starts with the 79*79 total commodity trade matrices for 1980, 1990, 2000, 2010 and 2016 as input. These will be first transformed by taking base log10 numbers to reduce skewness in later Correspondence Analysis. The matrix that represents the totals of all five technology categories is used, because this will display the overall network of trade ties among individual countries, and what position each individual country takes on the basis of their total trade relationships with other countries.
The second step is then to choose a measure of equivalence, to assess similarity between countries based on their trade patterns with other countries. Available measures of equivalence are structural equivalence and regular equivalence. Structural equivalence relates to the degree of similarity of two actors and the exact same other actors. Regular equivalence relaxes the definition of structural equivalence in the sense that for actors to be regularly equivalent, they would have ties with similar others, instead of the exact same others. To give an illustrative example: even though the USA and the UK are generally considered similar in terms of their patterns of trade with other (developing) countries, they would never be considered structurally equivalent, because the USA has ties with Latin American countries, whereas the UK has ties to African countries (Lloyd et al., 2009). For this study, the measure used is that of regular equivalence, following the algorithm of White & Reitz (1983). The regular equivalence 𝑀𝑀 between countries i and j at iteration t+1 is given by:
𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖+1 =∑𝑔𝑔𝑘𝑘=1𝑚𝑚𝑡𝑡𝑡𝑡𝑚𝑚=1𝑔𝑔 ∑𝑖𝑖=1𝑅𝑅 𝑀𝑀𝑘𝑘𝑚𝑚𝑖𝑖 (𝑖𝑖𝑖𝑖𝑖𝑖𝑀𝑀𝑘𝑘𝑚𝑚𝑖𝑖𝑖𝑖 +𝑖𝑖𝑖𝑖𝑖𝑖𝑀𝑀𝑘𝑘𝑚𝑚𝑖𝑖𝑖𝑖 )
∑𝑔𝑔𝑘𝑘=1𝑚𝑚𝑡𝑡𝑡𝑡𝑚𝑚∗ ∑𝑅𝑅𝑖𝑖=1(𝑖𝑖𝑖𝑖𝑖𝑖𝑀𝑀𝑡𝑡𝑡𝑡𝑘𝑘𝑚𝑚𝑖𝑖+𝑖𝑖𝑖𝑖𝑖𝑖𝑀𝑀𝑡𝑡𝑡𝑡𝑘𝑘𝑚𝑚𝑖𝑖) (5)
This algorithm finds the best possible matching of relationships between country i and j, and is provided in software package UCInet. UCInet returns an equivalence matrix with a number between 0 and 1 in each cell - 0 represents maximal dissimilarity, 1 represents maximal
4 Borgatti, S.P., Everett, M.G. & Freeman, L.C. (2002). UCInet 6 for Windows: Software for Social Network Analysis. Harvard, MA: Analytic Technologies. Version used: v6.332
17 similarity. This matrix of regular equivalencies is of crucial importance, because it represents the input for all further steps.
The third step is usually to use the matrix of regular equivalencies as input for a hierarchical clustering routine, as to arrive at similar groups of countries that can be classified as core - periphery. Due to the complexity of the network of trade data, it is very well possible that it is not immediately apparent what the different clusters are, i.e. that are no clear partitions, which is a phenomenon conceptualized as “chaining” (Mahutga & Clark, 2018). In such cases, one can analyze the regular equivalencies with reduced ranked decompositions, like a correspondence analysis. Which is how we will proceed.
Correspondence analysis comes from a family of techniques that draws upon singular value decomposition computing routines, and aids in giving simplified, continuous representations of data used as input. UCInet has a built-in correspondence analysis routine, and Lloyd et al. (2009) provide a more technical breakdown of the steps performed in correspondence analysis. Roughly, correspondence analysis uses the matrix of regular equivalencies and presents it in a multi-dimensional Euclidean space, giving each country a set of coordinates that places them close to the ones to which they are similar, and far from the ones to which they are dissimilar. A simplified explanation of the procedure is as follows6: UCInet breaks down input matrices into one matrix that summarizes all information in the rows of the input matrix by taking row averages, another matrix that summarizes all information in the columns by taking column averages. It then computes the expected value of each cell from the original matrix by multiplying the row average of a cell by the column average, and then dividing it by the overall average, and consequently calculates residuals by subtracting the expected value for each cell from its original value. UCInet thus returns the residuals as coordinates, with a coordinate being defined as ‘scores for each point on each dimension adjusted both for point marginals and dimension weights (eigenvalues)’. The coordinates are then visualized into a scatter plot, where actors with similar coordinates are being put close together. The horizontal axis of this scatter plot is the first dimension, that represents the highest amount of variance. A dimension is simply some description of the information provided in the matrices used as input.
The output of correspondence analysis thus shows the dimensions and the amount of variation explained by each dimension, as well as the coordinates of each country that places it in an Euclidean space that can be visualized in a scatter plot. This plot uses the first two dimensions as axes, and positions countries on the basis of their coordinates along the dimensions. In short, interpreting the output from correspondence analyses thus depends on the amount of variation explained by each dimension, and the corresponding Euclidean space that is provided in a scatter plot. Ideally, the first one or two dimensions explain almost all of the variance. Roughly speaking, a country that has a coordinate of zero has no deviations from the average country, and the further away a country moves from the origin (0,0) - the more differentiated it is. Because the CA here is based on the regular equivalencies matrix, the distance between
18 countries in terms of their coordinates can be interpreted as the extent to which they are similar to each other.
4.1.1 Calculation of world-system mobility
World-system mobility will be derived from the continuous representation of the network of total commodity trade that results from correspondence analysis. Here, the measurement of Mahutga & Smith (2011) will be taken and slightly adapted. They take the coordinates that were obtained from correspondence analysis and define world-system mobility as the result of changes in distance to the core group of countries over 10-year intervals. Note that the group that represents the core in their analysis is identified after the hierarchical clustering procedure that groups similar countries into equivalent groups. This step was deliberately skipped in this study, and we will only build on the output from correspondence analysis. They measure the distance between a particular country and the core at time t by subtracting the first dimensional coordinate for country i at time t from the average first dimensional coordinate for all core countries. This measure of distance is then used to assess the mobility of non-core countries over time by first subtracting the distance at time t+1 from the distance at time t, and then dividing it by the distance to the core in t. Here, distance to the USA will be measured instead of distance to the core, because the USA very reasonably represents what is generally perceived as the core for decades already. Distance can be calculated as follows:
𝑓𝑓𝑖𝑖𝑖𝑖 = 𝑡𝑡𝑈𝑈𝑈𝑈𝑈𝑈𝑖𝑖− 𝑡𝑡𝑖𝑖𝑖𝑖 (6)
where i denotes an individual country, t denotes the year, 𝑡𝑡𝑈𝑈𝑈𝑈𝑈𝑈𝑖𝑖 is the first dimensional coordinate of the USA at time t, and 𝑡𝑡𝑖𝑖𝑖𝑖 is the first dimensional coordinate of country i at time
t. Using this measure of distance, mobility is then calculated as follows:
𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖 =𝑑𝑑𝑖𝑖𝑖𝑖−𝑑𝑑𝑑𝑑𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖+1 (7)
19
4.2 Panel data analysis
4.2.1 Descriptive statistics
Obs. Mean Std. dev Min Max
Mobility 316 -0.039 1.154 -20.320 1.009 htexp 316 0.015 0.105 -0.721 0.603 mtexp 316 0.013 0.074 -0.460 0.751 ltexp 316 -0.003 0.085 -0.381 0.382 rbexp 316 0.005 0.114 -0.496 0.637 pgexp 316 -0.030 0.139 -0.713 0.701 htimp 316 0.015 0.084 -0.258 0.314 fdiin 248 0.009 0.078 -0.547 0.502
Table 2: Descriptive statistics of all variables included
Under the specified calculation, the mobility number for China was as high as -20 between 2010 and 2016, which is highly unrealistic and distortionary. Therefore, this value was replaced for the average mobility number of that period. Further reported tests and results will be based on this replaced value. Furthermore, the numbers for the group that constitutes the core countries (the UK, Germany, France, the Netherlands, Japan and Italy) are overstated for every year. That is, their relative positions remain about equal, and their distance to the USA remains small, but the specified calculation of mobility shows that the magnitude of change7 is either extremely positive or extremely negative, meaning that these countries often show up as outliers.
As for the changes in each of the five commodity export groups, the minimum and maximum values represent extremes. These are in the grand majority of cases the result of either countries that generally have very low trade values, but experience one year of ‘luck’, or because in some years missing reporters were included with the help of export data, instead of import data (see section 3.3). To give a few illustrative examples: within high technology exports there were some remarkable positive and negative peaks for Niger, Malawi and Cabo Verde. Niger, for example, exported a large amount of radioactive commodities to Germany in 1980 and to France in 2000, 2010 and 2016. This was not maintained in 1990, resulting in extreme peaks in 1990 and 2000. Malawi is another country that benefited from a big amount of exports of radioactive goods. Cabo Verde was not a reporter in 1990, meaning here the data were filled up with what other countries recorded as exports, which might have yielded diverging results in between 1980 and 1990, and 1990 and 2000.
FDI inflows have the lowest amount of observations. This variable had a considerable amount of missing data, mainly within the relatively poor countries that were included in the sample.
20 4.2.2 Regression assumptions
The empirical model as specified in equation (1) will be tested with panel data analysis. To do so, first, a number of diagnostic checks need to be satisfied. In short, this requires to check for linearity, the normality of the error term of the specified model, multicollinearity, autocorrelation and heteroscedasticity. All tests and supporting graphs are included in Appendix A.4.
21
5. RESULTS AND DISCUSSION
Results are again subdivided into two main parts: the results from social network analysis (section 5.1) and regression results (section 5.2).
5.1 Network roles and positions: the structure of the international division of labor
To recap, the structure of the network of total commodity trade that represents the international division of labor will be analyzed on the basis of the results from the identification of positions in the world-system that followed from correspondence analysis. CA provided a continuous representation of the network by assigning row and column coordinates to each country. A coordinate captures a score for each country on each dimension identified, adjusted for marginal and dimension weights (eigenvalues). Row coordinates follow from the first dimension, column coordinates from the second. First dimensional (row) coordinates for each country are included in Appendix A.3, column 3. Previous studies of network roles and positions in the international division of labor universally found that the networks identified from correspondence analysis conformed to a core - periphery structure in which there is an overall well-connected core group of countries, and an isolated group of peripheral countries on the basis of two indicators (Smith & White, 1992; Mahutga, 2006; Lloyd et al., 2009 Mahutga & Smith, 2011). First, core-periphery structures typically demonstrate large variance explained by the first dimension of correspondence analysis, suggesting that this first dimension supports a core-periphery structure. Second, a hierarchical ordering of countries along this first dimension is suggestive of the fact that the first dimension captures overall ‘core-ness’ in the world-system. What follows is an overview of findings that characterize the calculated network of this study. Table 3 below reports the first five dimensions from correspondence analysis, along with their singular values and variance explained.
1980 1990 2000 2010 2016
Dimension 1 Singular value 0.022 0.022 0.026 0.024 0.024
Var. explained 69.2% 72.6% 73.9% 75.2% 75.3%
Dimension 2 Singular value 0.002 0.002 0.002 0.002 0.002
Var. explained 7.6% 7.1% 5.5% 5.4% 5.1%
Dimension 3 Singular value 0.001 0.001 0.001 0.001 0.001
Var. explained 4.3% 4.0% 4.0% 4.0% 4.3%
Dimension 4 Singular value 0.001 0.001 0.001 0.001 0.001
Var. explained 3.0% 2.1% 1.9% 2.1% 1.8%
Dimension 5 Singular value 0.001 0.000 0.000 0.000 0.000
Var. explained 1.7% 1.3% 1.4% 1.1% 1.3%
22 On average, the first dimension explains 73.24% of the variance of the regular equivalencies matrix. This is lower than was found in previous studies, where the first dimension explains on average around 90% of the variance of the regular equivalencies matrix (Mahutga, 2006; Lloyd et al., 2009; Mahutga & Smith, 2011). This means that further interpretation of row coordinates, as well as later Mobility numbers comes with the caveat that there is variance in the overall ‘core-ness’ of the world-system that is not explained by the first dimension.
The second gauge as to assess whether the analysis conforms to a core-periphery structure is to visualize how countries are positioned in terms of their coordinates along the first and second dimensions. After performing correspondence analysis, UCInet returns 2D scatter plots. These scatter plots use the first dimensional (row) coordinates as the x-axis, and the second dimensional (column) coordinates as the y-axis. All scatter plots are included in Appendix A.5. The scatter plots show an ordering of countries from left (core) to right (periphery). This lends further support for the fact that the first dimension explains overall ‘core-ness’, in this case again with the caveat that not all variance is explained. The scatter plots also show considerable dynamics over time. That is, countries indeed change their positions.
To visualize the overall network of total commodity trade, consider Figures 1 and 2 below. These figures are the result of fitting the first dimensional coordinates over the matrix with total trade numbers for a given year. Only the figures for 1980 and 2016 are included to provide an intuition of the world-system, and an overall picture of changes that occurred.
23
24
25 In 1980, the group of countries on the ultimate left are the United States, Germany, France, the United Kingdom, Japan, Italy and the Netherlands. This group is followed by a cluster of countries that consists of some other European countries (Belgium, Sweden, Spain), as well as rapidly industrializing, rising economies (China, Taiwan, Hong Kong, Singapore and South Korea). From there on, the further you move to the right, the less well connected countries become. Interestingly, Saudi Arabia appeared among the group of core countries in 1980 (Figure 1), whereas in all other years it was consistently at approximately the same, lower position. Reason for Saudi Arabia to be positioned among all core countries in 1980 can be attributed to the ‘1980s oil glut’, that followed the 1973 and 1979 oil crises. Between 1980 and 1986, the OPEC countries cut their oil production nearly in half at various moments in an attempt to keep the oil prices high. The result was a peak in oil prices that resulted in an extraordinarily high total export value for Saudi Arabia in 1980 across all commodity groups that contain oil. While other countries also saw sharp increases in their total export value over the same period, Saudi Arabia showed similar trade ties as core countries do. For example, it had very high export values to other core countries like Germany, France and the Netherlands, while in other years export values to these core countries are far less. This resulted in their ‘unexpected’ position for 1980.
In 2016, some considerable changes took place already. China entered to a core position, on the same level as Germany and the United States, and much of the rapidly industrializing countries also entered the group of core countries. On the right side of the chart, not much has changed.
In short: the spatial ordering of countries shows that countries become more ‘core-like’ when moving from right to left. There are also more dynamics on the overall left-hand side than on the right-hand side of Figures 1 and 2 (see also Appendix A.5). Over the period as a whole, the average mobility number of the top 50% of countries was 0.049, compared to -0.001 of the bottom 50%. Figure 3 illustrates the pattern of mobility for a group of eight countries between 1980 and 2016.
Figure 3: Patterns of mobility of a selected group of countries, 1980-2016.
Note: the years on the horizontal axis represents time periods, thus 1980 = 1980-1990
-0,2 -0,1 0 0,1 0,2 0,3 0,4 0,5 1 9 8 0 1 9 9 0 2 0 0 0 2 0 1 0
26 As Figure 3 illustrates, most countries that are located in upper-middle part of the network experience upward mobility over the entire period. The countries taken as an example are South Korea, India, Malaysia, Thailand and Mexico, which are countries that are known for their paths of rapid industrialization and export-oriented growth. On the lower end of the network, there is stagnation and downward mobility, as is illustrated by Central African Republic and Malawi in the above chart. The next section will present the results on the regression analysis of the drivers of mobility.
5.2 Panel data analysis
5.2.1 Correlations
1. Mobility 2. htexp 3. mtexp 4. ltexp 5. rbexp 6. htimp 7. fdiin 1. Mobility 1.0000 2. htexp 0.1202** 1.0000 3. mtexp 0.0048 -0.0181 1.0000 4. ltexp -0.2081*** -0.1421** 0.0112 1.0000 5. rbexp -0.0578 -0.2421*** -0.2571*** -0.2357*** 1.0000 6. htimp -0.1006* 0.3662*** 0.1182** 0.0000 -0.3000*** 1.0000 7. fdiin 0.0223 0.1537** 0.1194* -0.0366 -0.0591 0.0982 1.0000
Table 4: Pairwise correlations. Number of observations = 248
Note: * significant at 10% level, ** significant at 5% level, *** significant at 1% level
Table 4 shows that overall, the correlations are low. There are thus no more signs of multicollinearity, and there is heterogeneity in the data. The difference in high technology exports is positively correlated with Mobility, as expected. Medium technology exports show almost no relation, which is in line with the pattern we saw in the scatter plots included in Appendix A.3.1. Low technology exports show a negative correlation with Mobility, as expected. Resource based manufactured exports, again, do show the expected sign but a moderate relation. High technology imports show a negative significant correlation with Mobility, against expectations. FDI inflows only shows a very moderate positive relationship. The correlations give a first indication of the relationships, however, conclusions can only be drawn after the regression analysis.
5.2.2 Regression results
27
dependent variable: Mobility
1 2 difference htexp 0.244 (0.090)*** difference mtexp 0.254 (0.110)** difference ltexp -0.044 (0.104) difference rbexp -0.059 (0.057) difference htimp 0.297 0.159 (0.141)** (0.135) difference fdiin 0.070 0.013 (0.111) (0.103) constant -0.013 -0.015 (0.018) (0.020) Observations 248 248 Countries 75 75 R-squared 0.3267 0.3694 within 0.4077 0.4261 between 0.0537 0.1709
year dummies? Yes Yes
Wald statistic 118.92*** 128.67***
Table 5: regression results 1980-2016, random effects
Note: cluster robust standard errors in parentheses, * significant at 10% level, ** significant at 5% level, *** significant at 1% level
The Wald statistic is significant at Prob > chi2 = 0.0000, which means that the model can be interpreted. The R-squared shows that the included variables are responsible for 36.94% of the variation in Mobility. The within R-squared is higher than the between R-squared. This indicates that over time, most of the variation in the model is explained due to differences within countries.
Increasing the share of high technology commodity exports in total exports enters the regression with a positive, significant coefficient. This implies that, ceteris paribus, countries that increase the share of high technology exports in their total exports experience upward mobility in the world-system. Increasing the share of medium technology goods in total exports is also found to significantly impact world-system mobility. Both increasing the share low technology and resource based manufactured commodities in total exports negatively impacts world-system mobility, although the effects are not found to be statistically significant and the coefficients show a less strong effect than that of shifting towards exporting more high- and medium technology commodities does.
28 5.2.3 Robustness check
The sub-sample of rapidly industrializing Asian countries could bias the results upwards, as their patterns of growth and development potentially increases their weight in the sample. Therefore, the analysis is repeated, dropping all Asian countries from the sample. Table 6 below reports the results.
dependent variable: Mobility
1 2 difference htexp 0.137 (0.066)** difference mtexp 0.143 (0.103) difference ltexp 0.056 (0.077) difference rbexp -0.039 (0.044) difference htimp 0.127 0.117 (0.114) (0.115) difference FDIin 0.081 0.046 (0.105) (0.102) constant -0.012 -0.020 (0.016) (0.018) Observations 207 207 Countries 63 63 R-squared 0.4238 0.4427 within 0.4796 0.4926 between 0.0784 0.1369
year dummies? Yes Yes
Wald statistic 125.47*** 142.18***
Table 6, regression results 1980-2016 with Asian countries excluded
Note: cluster robust standard errors in parentheses, * significant at 10% level, ** significant at 5% level, *** significant at 1% level
High technology exports still enter with a positive, significant effect, however, the size of the coefficient is smaller. Medium technology exports also have a smaller coefficient, and lost their significance. Low technology exports now have a positive coefficient, which is at odds with the expectations and with the results reported in Table 5. Resource based manufactured exports, as well as high technology imports and FDI inflows show similar results.
29
5.3 Discussion
In what follows, the results as presented in sections 5.1 and 5.2 will be discussed and placed into theoretical perspective.
First, the network of total commodity trade showed a similar hierarchical ordering of countries as was found in previous studies. It is also found that countries on the left side of the network (i.e. the core and semi-periphery) are better connected, and on average more mobile than countries on the right side of the network (i.e. the periphery) (Smith & White, 1992; Mahutga, 2006; Lloyd et al., 2009; Mahutga & Smith, 2011). As for the mechanisms behind mobility, the results show that shifting towards more high- and medium technology exports significantly impacts upward mobility in the world-system, which lends support for hypothesis 1. Shifting towards low technology commodities and resource based manufactured commodities are not found to significantly impact mobility. Hypothesis 2 is thus not found to be statistically confirmed, however, the coefficients have the expected sign. High technology imports enters with a positive coefficient, FDI inflows with a small positive coefficient. Both are not found to be significant, but their coefficients do lend support for the expectation that technology diffusion and learning could take place through these two channels, and so also lead to upward mobility in the world system.
Why are the effects for high- and medium technology commodities stronger (in terms of both coefficient and level of significance) than those of low technology and resource based manufactured commodities? As was argued in section 2, low technology goods are produced under technologies that are easily diffusible and can be easily codified, which facilitates relocation from high to low wage and allows countries to capture and master the technologies with ease. This can result in rapid increases in export growth (Lall, 2000b). Accordingly, there might be countries that enjoyed such ‘once-for-all’ boosts in total exports, resulting in higher positions in the overall network as well as increases in the share of low technology exports in total exports, leading to a less strong negative effect of increasing low technology exports on mobility than expected. A possible explanation of the result on resource based manufactured goods can be that this group of commodities is very sensitive to fluctuations in prices. Mainly after 2000, the prices of natural resource based commodities like metals increased tremendously (see for example the ‘2000s commodities boom’). Because all data are constructed based on trade values in dollars, changes in price can make especially natural resource exporters end up higher in the overall network, and paired with an increase in the share of natural resource based manufactures in total exports this might result in a less negative effect of resource based manufactured exports on mobility than expected.
30 at higher levels of productivity and offer more potential for technology spillovers and learning that can be applied in future activities. Here, we join Lall (2000b) and Hausmann et al. (2006), and argue that the type of goods a country specializes in thus has important implications for its development. Countries that succeed in changing their roles in the international division of labor can thrive, whereas countries that do not are likely to experience delayed economic development.
31
6. CONCLUSIONS
The aim of this paper was to uncover the drivers of upward mobility in the world-system. Specifically, changing the role played in the international division of labor, as represented by the underlying composition of a country’s commodity exports was expected to offer an explanation. Does shifting towards exporting more high- and medium technology commodities constitute upward mobility, whereas shifting towards exporting more low technology and resource based manufactured commodities leads to downward mobility? The main results can be summarized as follows. The network analysis of commodity trade revealed that first of all, its structure and positions countries take within it have changed considerably. Most countries that changed their positions are on the overall left-hand side of the networks, i.e. rapidly industrializing countries like the East Asian ‘manufacturing miracles’, as well as Brazil and Mexico for example. Regression analysis showed that mobility is partly the result of increasing the share of high- and medium technology exports in total exports. The opposite effect is not found to be significant for low technology- and resource based manufactured exports. This comes with the note that when excluding the sub-sample of Asian countries from the analysis, the results only moderately hold.
Limitations of the study are the following. First, data were collected at five different points in time, which thus reflects a one-moment picture for each of these points. This could be improved upon by taking averages over periods, to balance out irregularities in a given year, like years with peaks in for example natural resource based commodity prices. An example here is that Saudi Arabia showed up between the core countries in 1980, due to a peak in oil prices. Second, because mobility is constructed based on the continuous representation of the network of total trade, the choice of sample may have impacted some of the results. That is, if other countries would have been included in the sample, and some of the countries in the current sample would have had more trade ties with them, they might have ended up higher in the overall network. Third, with increasing fragmentation of production globally, using total export data could bias the results, because these do not always reflect the underlying production processes that take place in a country, see here the discussion in section 3.1. Fourth, the current construction of mobility shows to have room for improvement. Some of the outcomes, especially in the higher spheres of the network where positions remain stable over time do not accurately reflect reality, see section 5.2.1. Finally, the tested empirical model builds on the theory of capability building by countries to be able to master the tacit elements of foreign technology and consequently to shift towards more high-technology exports. Present analysis includes two channels that might enable such technology diffusion: high technology imports and FDI inflows. However, technology diffusion is not an automatic process, but a process that requires investments and intervention. This can be improved upon by including more control variables that would capture the channels that allow for capability building by countries. As already argued, such control variables could be human capital, R&D expenses, or proxies for institutional quality.
32 production in different countries, shifting the focus of world-systems research towards global value chain participation could offer fruitful insights. The second would be to extend the empirical model to include more control variables that facilitate learning and capability building from participation in such global value chains.
All in all, the global economic map as represented by the calculated networks of commodity trade in 1980, 1990, 2000, 2010 and 2016 indeed has changed considerably. Complex processes of path dependency are at work, and future research can try to get a better picture of what this path looks like.
Acknowledgements – I would like to thank Dirk Akkermans for his supervision. Further
33
REFERENCES
Bock, T. (2018). How Correspondence Analysis Works (A Simple Explanation). Retrieved April 16, 2018 from https://www.displayr.com/how-correspondence-analysis-works/
Borgatti, S.P. (2002). Netdraw Network Visualisation. Analytic Technologies: Harvard, MA. Version used: v2.161
Borgatti, S.P., Everett, M.G. & Freeman, L.C. (2002). UCInet 6 for Windows: Software for Social Network Analysis. Harvard, MA: Analytic Technologies. Version used: v6.332
Borgatti, S.P., Everett, M.G. & Johnson, J.C. (2013). Analyzing Social Networks. Sage Publications.
Chase-Dunn, C. (1989). Global formation: Structures of the world-economy. Cambridge, UK: Blackwell.
Dicken, P. (2011). Global shift (6th ed.). New York: The Guilford Press.
Durand, E. D. (1953). Country Classification. International Trade Statistics, 117-129. Roy George Douglas Allen and J.E. Ely, editors. John Wiley.
Evenson, R., & Westphal, L.E. (1995). Technological change and technology strategy. Ch. 37. In: Behrman, J., Srinivasan, T.N. (Eds.), Handbook of Development Economics, vol. 3A. North-Holland, Amsterdam, 2209 – 2229.
Hausmann, R., & Rodrik, D. (2003). Economic Development as Self-Discovery. Journal of
Development Economics, 72, 603-633
Hausmann, R., Hwang, J., & Rodrik, D. (2006). What you export matters. Journal of Economic
Growth, 12, 1-25.
Keller, W. (2004). International Technology Diffusion. Journal of Economic Literature, 42(3), 752-782
Kick, E.L., & Davis, B.L. (2001). World-system Structure and Change. American Behavioral
Scientist, 44(10), 1561-1578.
Kim, S., & Shin, E. (2002). A Longitudinal Analysis of Globalization and Regionalization in International Trade: A Social Network Approach. Social Forces, 81(2), 445-468
Lall, S. (2000a). Technological change and industrialization in the Asian newly industrializing economies: achievements and challenges. In: Kim, L., Nelson, R.R. (Eds.), Technology, Learning and Innovation. Cambridge Univ. Press, New York, pp. 13 – 68
34 Lloyd, P., Mahutga, M. C., & Leeuw, J. D. (2009). Looking Back and Forging Ahead: Thirty Years of Social Network Research on the World-System. Journal of World-Systems Research,
15(1), 48-85.
Mahutga, M. C. (2006). The Persistence of Structural Inequality? A Network Analysis of International Trade, 1965-2000. Social Forces, 84(4), 1863-1889.
Mahutga, M.C., & Clark, R. (2018). Theory and Method in Macro-Comparative Social Network Anlaysis. Forthcoming in Zachary Neal and Celine Rozenblat (Eds), Handbook on Cities and
Networks, Camberley Surrey, UK: Edward Elgar
Mahutga, M. C., & Smith, D. A. (2011). Globalization, the structure of the world economy and economic development. Social Science Research, 40(1), 275-272
Smith, D. A., & White, D. R. (1992). Structure and Dynamics of the Global Economy: Network Analysis of International Trade 1965-1980. Social Forces, 70(4), 857.
United Nations, 2018. COMTRADE. United Nations, New York
UN Trade Statistics (2010). Taiwan, Province of China Trade data (China, Data availability, Taiwan). Retrieved April 11, 2018, from
https://unstats.un.org/unsd/tradekb/Knowledgebase/Taiwan-Province-of-China-Trade-data
Wallerstein, I. (1974). The modern world system: Capitalist agriculture and the origins of the
European world-economy in the sixteenth century. New York: Academic Press.
Wasserman, S., & Faust, K. (1994). Social network analysis: methods and applications. New York, NY: Cambridge University Press.
White, D. R. & Reitz K. P. (1983). Graph and semi-group homomorphisms on networks of relations. Social Networks, 6, 193-235.
World Bank (2018). World Bank Open Data. Retrieved April 12, 2018
35
APPENDIX
Appendix A.1 – Classification of commodity groups by technological intensity on page 37 Appendix A.2 – Variables, measurement and source on page 38
Appendix A.3 – Results from social network analysis on pages 39 – 48 Appendix A.4 – Regression assumptions on pages 49 – 52
36
Appendix A.1 – Classification of commodity groups by technological intensity
This classification was retrieved from Lall (2000b)
CATEGORY EXAMPLES SITC Rev.2 three digit
commodity code
Commodities
Fresh fruit, meat, rice, cocoa, tea, coffee, timber, coal, crude petroleum, gas, ore concentrates and scrap 001, 011, 022, 025, 034, 036, 041, 042, 043, 044, 045, 054, 057, 071, 072, 074, 075, 081, 091, 121, 211, 212, 222, 223, 232, 244, 245, 246, 261, 263, 268, 271, 273, 274, 277, 278, 281, 286, 287, 289, 291, 292, 322, 333, 341 Manufactures Natural resource-based manufactures
Prepared meats/fruits, beverages, wood products, vegetable oils, base metals (except steal),
petroleum products, cement, gems, glass 012, 014, 023, 024, 035, 037, 046, 047, 048, 056, 058, 061, 062, 073, 098, 111, 112, 122, 233, 247, 248, 251, 264, 265, 269, 423, 424, 431, 621, 625, 628, 633, 634, 635, 641, 282, 288, 323, 334, 335, 411, 511, 514, 515, 516, 522, 523, 531, 532, 551, 514, 515, 516, 522, 523, 531, 532, 551, 592, 661, 662, 663, 664, 667, 681, 682, 683, 684, 685, 686, 687, 688, 689
Low technology manufactures Textile fabrics, clothing, footwear, leather manufactures, travel goods pottery, simple metal structures, furniture, jewelry, toys, plastic products 611, 612, 613, 651, 652, 654, 655, 656, 657, 658, 659, 831, 842, 843, 844, 845, 846, 847, 848, 851, 642, 665, 666, 673, 674, 675, 676, 677, 679, 691, 692, 693, 694, 695, 696, 697, 699, 821, 893, 894, 895, 897, 898, 899
Medium technology manufactures Passenger vehicles and parts, commercial vehicles, motorcycles and parts, synthetic fibers, chemicals and paints, fertilizers, plastics, iron and steel, pipes and tubes, engines, motors, industrial machinery, pumps, ships, watches
781, 782, 783, 784, 785, 266, 267, 512, 513, 533, 553, 554, 562, 572, 582, 583, 584, 585, 591, 598, 653, 671, 672, 678, 786, 791, 882, 711, 713, 714, 721, 722, 723, 724, 725, 726, 727, 728, 736, 737, 741, 742, 743, 744, 745, 749, 762, 763, 772, 773, 775, 793, 812, 872, 873, 884, 885, 951
High technology manufactures Data processing and
telecommunications equipment, television sets, transistors, turbines, power generating equipment, pharmaceuticals, aerospace, optical and instruments, cameras
