World Development Vol. 37, No. 1, pp. 127145, 2009 2008 Elsevier Ltd. All rights reserved.

. 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev

doi:10.1016/j.worlddev.2008.03.009

Measuring Globalization of International Trade: Theory and Evidence

IVAN ARRIBAS ` ` Universitat de Valencia and Ivie, Valencia, Spain FRANCISCO PEREZ ` ncia and Ivie, Valencia, Spain ` Universitat de Vale

and
EMILI TORTOSA-AUSINA * Universitat Jaume I and Ivie, Castello de la Plana, Spain
Summary. Measuring globalization requires a Standard of Perfect International Integration as a benchmark that a single world space would reach under conditions of geographic neutrality in international trade. We dene this standard and present indicators for openness, connectedness and integration, for each specic economy, and for the world economy. We apply our indicators to data on trade ows for 59 countries for the 19672004 period. Results show that trade integration is higher than what traditional openness indicators suggest. Several economies nd high levels of integration, but the low degree of openness in some large economies jeopardizes the progress of globalization. 2008 Elsevier Ltd. All rights reserved. Key words geographic neutrality, globalization, international trade, network analysis

1. INTRODUCTION The advance of globalization has received a great deal of attention for some years now. However, to date, no consensus has been reached on the level of globalization attained and how far we are from completing it. A recent state of the art revision of the economic globalization indicators undertaken by the OECD, 2005 clearly shows that the current indicators do not reect the new reality. Its conclusion is that there is no connection between the importance that globalization has acquired and the improvements in the available indicators, which are inordinately based on the old concept of market openness. Measuring globalization is problematic because it is complex in two ways. First, it presents many relevant facets because of the increase in cross-national cultural, political, and social interactions, as well as economic interactions (commercial, nancial trade and investment ows, migration, and the performance of multinational companies). Second, it is not only a question of increasing the openness of countries but also of developing a network of direct and indirect relations between individuals at a distance from each other, which represents a progressive abolition of every possibility of remoteness. McLuhan and Fiore (1968) proposed an early image of this process in the sixties, referring to the birth of a global village, in which national spaces are abolished, and individuals must learn to live in close relation to formerly distant agents. To judge whether we are approaching this scenario, and at what speed, we need standards to characterize a world where frontiers and distance do not matter, and geographical dimensions are irrelevant. With this objective, and following a suggestion made by Frankel (2000), this study proposes a Standard of
127

Perfect International Integration (SPII), allowing us to describe the conditions under which the world economy would operate as a global village. This standard represents a benchmark that not only requires countries to be more open, but also requires them to attain a full and geographically unbiased development of the network of connections linking economies. The maximum value of globalization in this situation must rst be characterized so we can assess the distance that separates the current level of international economic integration from the scenario of complete globalization. With these objectives, and, following a review of the literature related to our aims (Section 2), the core of the article is structured into two sections, one theoretical and the other empirical. Section 3 presents the methodological contents of our approach to measure international economic integration. Section 4 presents the empirical application by considering data on exports of goods for a wide set of countries that account for virtually all world output, and for a relatively long sample period (19672004). After we have analyzed the most important features of the globalization process from the results obtained, and compared these results with other indexes

* This paper is a result of the FBBVA-Ivie Research Program. All the three authors acknowledge the nancial support of the Ministerio de E ducacion y Ciencia (SEJ2007-66581, jointly nanced with Feder funds, SEJ2005-02776 and SEJ2005-01163, respectively). Emili Tortosa-Ausina also acknowledges the nancial support of the Generalitat Valenciana (GV/2007/110) and Fundacio Caixa Castello-Bancaixa (P1.1 B2005-03). All the three authors are grateful for the comments by Rafael Llorca Vivero and Salvador Gil-Pareja, and to Rodrigo Aragon and Pilar Choren for excellent research assistance. Final revision accepted: March 19, 2008.

128

WORLD DEVELOPMENT

of international integration and globalization, Section 5 concludes. 2. THE EXISTING LITERATURE From the wide range of literature on international integration indicators, four issues emerge as most relevant to our objective. First, the focus of the indicators. Second, the analysis of the network connections. Third, the treatment of the geographical biases of the ows. And fourth, the dimension of the globalization considered. This section will focus on these points, providing a synthesized review of the existing literature. In general, international economic integration indicators can be classied into two broad categories: price-based and quantity-based indicators. 1 Many scholars prefer measures of integration based on prices to consider an axiomatic criterion, that is, the compliance with the law of one price (LOOP), in dierent geographical markets. However, empirically the LOOP does not seem to oer a very promising solution, since a unique price would exist only for homogeneous goods, and not for goods which could be dierentiated. 2 The most commonly used integration measure based on quantities is the degree of openness dened as exports plus imports divided by GDP ((X + M)/GDP). Although this measure provides a straightforward approach, from the globalization perspective, its principal disadvantage is that it completely disregards the architecture of trade connections that each country has with the rest of the world. 3 The OECD (2005) emphasizes the importance of this matter remarked above, since the process of integration also advances because of the eect of indirect networks (reected in the increase in the share of exports imported). Other factors are the increasing trade in intermediate goods and the degree of geographic diversication of the trade in the compiling country (OECD, 2005). Indeed, the architecture of the network of world trade ows turns out to be very relevant when assessing integration from a globalization perspective. This is because some of its more relevant features are the multiplicity of ows in many directions, the adherence to the process of all the countries, and the establishment of many other connection paths both direct and indirect, and physical or virtualbetween agents and economies. This increased number of connections may be ecient because of the development of information technology and the dramatic fall in transaction costs. 4 Our measurement is based on quantities (trade volumes) and not on prices. However, we attempt to overcome the shortage of the degree of openness by analyzing the nature of the network relations between economies. This is because any measure of international integration in the age of globalization must take into account the complexity of connections. To uncover the structure of the trade network that economies forge, we can contemplate the relations or ows between them as the vectors of a graph in which the nodes represent the countries. We then analyze the degree of connectedness in the network. Although these techniques are somewhat underused by economists compared with other social sciences, 5 this approach is not new in international economics, and has attracted recent interest. 6 Other works suggest applying complex network analyses concepts, topological properties, and instruments from dierent sciences (from physics to sociology) developed to study the structure and dynamics of international trade, using instruments such as centrality, network density, clustering, assortative mixing, or maximum ow. When applying these

techniques to the study of the World Trade Web (WTW) the focus is on the evolution of the degree of connectedness among countries, the inuence of the level of development on the position (central and peripheral) of the countries in the WTW, and the role of regional connectedness in the globalization context (Kali & Reyes, 2007), and economic growth (Fagiolo, Reyes, & Schiavo, 2007; Kali, Mendez, & Reyes, 2007). 7 The aforementioned papers study the statistical properties of international trade by applying network analysis tools. However, network analysis describes the economy using descriptive statistics that, in principle, do not have any economic interpretationalthough recent papers show that they allow us to explain the growth in the economy (Fagiolo et al., 2007; Kali et al., 2007; Kali & Reyes, 2007). Our analysis of the integration shares two characteristics with the network analysis approach. First, while the number of connections an economy has is important, the way they are distributed is also highly relevant. Second, ows among economies reect only the rst-order relationships, when higher-orders are also relevant since many of those ows move through dierent economies before reaching their nal destination. However, while these analyses consider a set of useful statistics to uncover the topology of the WTW, our methodology diers in that our integration index is specically dened to measure how far/close the world economy is to a Standard of Perfect International Integration. Therefore, our description of the economy as a network allows us to study which properties clearly related to the concept of integration are veried and to analyze the evolution of some economic features. In order to characterize the stage of globalization in which barriers and distances are irrelevant, the third important issue to address is to measure the geographical biases in both directions: through the home economy or by prioritizing some connections over others. The literature analyzing regionalism (and its eects on the intensity of intra-regional and extra-regional trade) suggests approaching this problem via the concept of geographic neutrality. 8 This concept can be dened as the absence of preferential directions in trade ows. Thus, the geographic distribution of a countrys trade is said to be neutral if the weight of every partner in the countrys trade is equal to its weight in the world trade. This concept can be generalized for our objectives, in order to make the supraterritoriality 9 idea eective. According to Scholte (2002), globalization is not a particularly intense form of internationalization, since it refers to a shift in the nature of social space involving a removal of barriers to transworld contacts. It also promotes transplanetary connectivity and supraterritorial links. In fourth place, some research studies rightly emphasize the importance of capturing the multiple facets of globalization, in order to control for the fact that integration is advancing in many directions. And also to understand the interrelations between the components of this process. From this point of view, indexes that contemplate dierent dimensions have been built and subsequently added to frequently used synthetic indicators (see Andersen & Herbertsson, 2003; Dreher, 2006; Heshmati, 2006; Kearney, 2003; Lockwood & Redoano, 2005; Martens & Zywietz, 2006). However, statistics criteria are used to add the partial indicators (principal components or factor analysis), and their meaning is not always clear. We have chosen to accept these limitations (Lockwood & Redoano, 2005) because of the signicant problems caused by the synthetic indicators. Thus, we have focused on one dimension of globalization, both in the construction of the SPII and in the empirical application. The aspect chosen is international trade and, more specically, exports. 10

MEASURING GLOBALIZATION OF INTERNATIONAL TRADE: THEORY AND EVIDENCE

129

Some authors have recently pointed out how relevant it is not only to focus on trade volumes and trade openness, but also on trade policies (Rodrguez & Rodrik, 2001). Our study does not tackle these issues directly. However, we provide some measures which could be added to the literature on outward orientation and growth, since the unconvincing and fragile results (Edwards, 1993) could be partly related to the openness and/or integration measures used. Taking into account these four issues, we will develop a proposal for integration and globalization indicators in Section 3, which we believe will solve the problem of certain gaps in the literature. The core of our contribution is formed by the SPII concept and a treatment of its components (degree of openness and degree of connectedness). This allows us to interpret the economic meaning of both, as well as their importance in the advance of globalization. As can be seen when analyzing the results, they dier from those existing in the literature (in both traditional degrees of openness and those generated by network analysis), precisely because two distinct and relevant dimensions of globalization are combined in one analysis. 3. INTEGRATION INDICATORS: DEFINITIONS AND PROPERTIES The international integration process starts with the openness of economies, but its eects and scope also depend on the structure of current relations between these economies. Relevant aspects of this structure include the number of economies each one is in contact with, whether the relations are direct or indirect, the number of ows between them, and the proportionality of these ows to the size of the economies. Considering globalization and integration as synonymous, we dene the SPII as an extension of Iapadres (2006) concept of geographic neutrality. This concept covers the notion of single space world where the ow from one region to another depends only on their relative sizes. Contrary to some common international economic integration indexes, openness does not necessarily imply globalization according to the geographic neutrality criterion (Iapadre, 2006). An economy whose trading ows with the rest of the world are below those corresponding to its size is as far from being integrated as another economy whose trading ows are above that proportion. Both economies show an imbalance between the domestic ows (i.e., the share of their GDP consumed internally) and the international ows (i.e., the share of their GDP they export), which is not in accordance with the geographic neutrality criterion. Under the neutrality assumption, there exists an optimal value for the international ows, and the following property must be veried: Domestic neutrality (P1): An economy whose domestic demand is proportional to its share of the world economy will have a higher level of integration. Both the global ow of an economy and its distribution are important. In a global village, when there are no transaction costs or regional preferences, the distribution of the ow of an economy between the destination economies should be characterized by their relative sizes. Under geographic neutrality, an economy has no preferences of any kind (social, political, geographical, etc.) for the direction of its ows. They are only determined by the size of the recipient economies, as stated by the following property (P2): Direct international neutrality (P2): An economy that balances its direct relations with other individual economies, in proportion to their sizes, will have a higher level of integration.

Although ows between economies reect only rst-order relations, higher-orders are also relevant. The set of relations established between economies operates like roads between cities. First, they allow economies to be connected even when there is no direct relation between them. Second, ows can reach their nal destination in dierent ways, depending on the intermediating economies they cross. Goods, services, and capital may move from one economy to another several times before arriving at their nal destination. This possibility enables the interconnectivity of the world to increase, and therefore its integration. Indirect international neutrality (P3): An economy that reinforces its relations with other economies through indirect relationships across intermediating economies will have a higher level of integration. An economy can deviate from perfect integration as a result of some of the factors mentioned above. The impact of this deviation on the integration of the world will depend on the size of each economy. When a large economy departs from perfect integration, it reduces global integration to a greater extent than a small economy does. For example, the inuence of the USA on global integration is necessarily higher than, for instance, the inuence of Albania. Thus, the integration index should also be weighted by the size of the economies. Size (P4): The larger the economy, the more relevant its integration will be to the globalization of the world economy (global level of integration). We say that the world meets the SPII if properties P1 to P4 are veried, constituting an extension of Iapadres (2006) concept of neutrality. Given its wider coverage, we call this geographic superneutrality. In order to discover the extent to which economies meet the four properties mentioned above, we must dene an integration index and assess the distance that separates the current level of integration from the SPII. We will proceed in four stages, each one dening dierent indicators, which correspond to the following four subsections. Moreover, the analysis of the four indicators is conducted on two levels. First, the individual level, which focuses on each economy, and the global level, which corresponds to the analysis of all economies. On the second level, the weight of each economy enters the aggregation analysis and allows us to dene our Integration Index. Let us start with some denitions. Let N = {1, . . . ,g} be the set of nodes or economies, and let i and j be typical members of this set. Let g be the number of elements in N, that is, the number of economies in the analysis. Let Yi be the size of economy i 2 N, for example, its GDP. We dene ai as economy is relative weight with respect to the world economy, that P is, ai Y i = j2N Y j . Given a measurable relation between economies, we dene the ow Xij as the intensity of this relationship from economy i to economy j. The ow between economies can be evaluated through either the imports or the exports of goods, capital, or any other ow measured in the same units as Yi. Moreover, in general the ow will be asymmetric, so that Xij will not necessarily be equal to Xji, for all i,j 2 N. We also assume that Xii = 0 for all economy i 2 N. All the denitions in this paper hinge on the ow considered to measure international economic integration. (a) Degree of openness In the rst stage, we characterize the degree of openness. We start with the usual denition found in the literature, but cor-

130

WORLD DEVELOPMENT

rected for domestic bias to take into account the diering sizes of the economies being compared. If the orientation of production toward domestic demand is not biased, then its volume will not be the same for each economy since it depends on its size. In order to remove the domesb tic bias, we dene Y i as the ow from economy i to the world, taking into account the weight in the world economy of the b economy under analysis, namely, Y i Y i ai Y i . Then, we dene the relative ow or degree of openness between economies i b and j as DOij X ij = Y i . Given that Xii = 0, it follows that DOii = 0 for all i 2 N. Denition 1. Given an economy i 2 N, we dene its degree of openness, DOi, as P X j2N X ij DOi DOij : 1 b Yi j2N By denition the degree of openness takes the value of 1 if and only if domestic neutrality (P1) is veried. The degree of openness yields results (in general) ranging in the (0, 1) interval, where a value of 0 indicates that the economy is closed (compared to the measure of ow chosen) and a value of 1 indicates a lack of domestic bias in the economy (total openness). Although the degree of openness in an economy is, in general, lower than 1, some particular economies might surpass this value showing an extremely open character. 11 Dierences in DO among economies can be attributed to dierent obstacles to integration (transport costs, trade barriers, political factors, etc.), among which we also nd size. However, dierences cannot be due to the bias in the measure of openness, since we have corrected for domestic bias. 12 (b) Degree of direct connection In this second stage, we analyze whether the connection of one economy with others is proportional to their sizes in terms of GDP, 13 or whether this connection does not show geographical superneutrality, drawing us away from the scenario of a perfectly integrated world according to the direct international neutrality property (P2). Thus, we dene the degree of direct connection to measure the discrepancy between the trade volumes in the real world and those corresponding to the SPII. In the economic network, the relative ow from economy i to economy j in terms of the total ow of economy i, aij, is given by aij P X ij j2N X ij 2

is the relative weight of economy j in a world where economy i is not considered. P Note that j2N i bij 1 and that bij is the degree of openness between economies i and j in the perfectly connected world, with bii = 0. Let B = (bij) be the square matrix of degrees of openness in the perfectly connected world. Starting from the previously dened matrices, we can dene an indicator that measures the distance between the real distribution of ows and that corresponding to a perfectly connected world. We consider the cosine of the angle of the vector of relative ows with the vector of the ows in a perfectly connected world. Denition 2. Given an economy i 2 N, we dene the degree of direct connection of i, DDCi, as P j2N aij bij 4 DDC i qq P P 2 2 j2N aij j2N bij Although the cosine of two vectors ranges between 1 and 1, the degree of direct connections always takes nonnegative values since both vectors have only nonnegative components. DDC measures whether economies meet P2, providing a single value that equals 1 if and only if an economy meets the property of direct international neutrality. It approaches zero for an economy whose ows are directed toward the smallest world economies. (c) Degree of total connection Indirect relations between economies and their importance are considered in the third stage. To extend the analysis of economic integration in this direction, we dene the degree of total connection. By doing this we evaluate the importance of all direct and indirect relations that economies establish with each other. Both the real world matrix A and the perfectly connected world matrix B consider direct relative ows between economies. However, part of the ow from economy i to economy j might pass through other economies and those indirect ows also contribute to integration. Let An A A . n . A be the n-times product matrix of ma. trix A and let an be the element ij of An. It is not dicult to ij show that an is the relative ow that goes from i to j passing ij through n 1 intermediate economies. Moreover, we can verify that 0 6 an 6 1 for all n P 1. In the same way we dene ij Bn, the elements of which evaluate the ow passing through all economies in a perfectly connected world. Let ci 2 (0, 1) be the proportion of ow that economy i receives from another economy and remains for internal consumption, while 1 ci is the proportion of received ow that an economy re-exports, possibly after some transformation. Alternatively, we can interpret the inverse of ci as the number of times a good is traded (on average) from its production in economy i until it reaches its nal destination. Thus, c = 0.5 is consistent with the assumption that goods are traded twice (either with or without added manufacturing operations). For example, between economies i and j there is only an intermediate economy and two transformations, or trades, are made. Let C be the square diagonal matrix of direct ow proportions, so that the element ii of C is ci and the element ij, for i j, is zero. The matrix of total ows from one economy to another is the sum of the direct and indirect ows. It can be estimated as follows:

(recall that we are assuming Xii = 0). Let A = (aij) be the square matrix of relative ows: the component ij of matrix A is aij. We consider that the world economy is perfectly connected if the ow between two economies is proportional to their relative sizes. An economy that is part of a perfectly connected world will trade with other economies in proportion to the size of the recipient economies. On the other hand, if the world economy is perfectly connected, then the ow from economy i to economy j should b be equal to bij Y i , where bij P Yj
k2Nni Y k

MEASURING GLOBALIZATION OF INTERNATIONAL TRADE: THEORY AND EVIDENCE

131

AC BC

1 X n1 1 X n1

CI Cn1 An ; CI Cn1 Bn

5 6

If economy i veries Assumption 4, then 1 ci = 1 di and hence ci = di. It therefore follows that X i 1 ci M i 1 di Y i 1 ci M i 1 ci Y i 1 ci M i Y i which implies that 1 ci = Xi/(Yi + Mi) or equivalently ci Di =Y i M i : 8 Therefore, under Assumption 4, ci is the proportion of available goods (production plus imports) that are internally consumed by economy i. Obviously, the procedure to estimate ci will hinge on the ow considered. (d) Degree of integration From the concepts mentioned above, we dene the degree of integration. This combines degrees of openness and total connection, provided that both set limits to the integration level achieved. Denition 5. Given an economy i 2 N we dene its degree of integration, DI C , as i q 9 DI C DOi DTC C i i The degree of integration of an economy is the geometric average of its degrees of openness and total connection. Thus, DI depends on both the openness of the economy and the balance in its direct and indirect ows. Moreover, if the economy veries properties P1 to P3, then DI will be equal to 1. 15 q If DI C DOi DTC C , then i i ss C DOi DTC i 1 10 DI C DI C i i and we can interpret each of these two factors as the weight that the degrees of openness and total connection have over the degree of integration. In a given economy, this can be useful to analyze changes over time in the weight of the factors. To characterize the integration of the world, we should consider the share of each economy in the world (property 4) to dene P the global indicator as follows (recall that ai Y i = j2N Y j . Denition 6. We dene the degree of integration (globalization) of the world as X DGI C ai DI C : 11 i
i2N

where I is the identity matrix of order g. Both expressions depend on matrix C. Let aC be the element ij of the matrix AC, and bC be the eleij ij ment ij of the matrix BC. Each element of these matrices is the weighted sum of the direct and indirect ows through any possible number of intermediate economies. We can verify that the above two series are convergent (see Appendix A). Denition 3. Given an economy i 2 N, we dene the degree of total connection of i, DTC C , as i P C C j2N aij bij 7 DTC C qq : i P P aC 2 bC 2 ij ij j2N j2N The degree of total connection ranges in the (0, 1) interval. This measures the distance of the direct and indirect ows of an economy from what its ows would be in a perfectly connected world. Similarly to the degree of direct connection, the total connection should be 1 when the ows of an economy are proportional to the size of the recipient economies (indirect international neutrality). It should be close to zero if the largest economies do not receive any commodities and the smallest receive all the goods. We should bear in mind that if there are no indirect ows, ci = 1 for all economies, then expressions (5) and (6) yield AC = A and BC = B. Therefore, the degrees of total connection and direct connection coincide. The limit case ci = 0 (goods receive an innite number of transformations, or are traded innitely, before reaching their nal destinations) cannot be derived directly from the above expressions. The basic limit theorem of Markov chains 14 is needed to show that when c = 0 the proportion of ow an economy j receives from an economy i is independent of i, that is, all economies export the same proportion of ow to economy j. However, DTCC depends on ci which measure the incidence of indirect ows in the connections between economies; these should be calculated before the degree of total connection. Let us assume that the considered ow among economies under analysis, Xij, are exports. Then recall we dene ci as the proportion of the economy is imports remaining for internal consumption. Likewise, we dene di as the proportion of the economy is production that is consumed by the own economy itself, whereas 1 di is the production that economy i exports. Notice that we are assuming that both ci and di are country dependent. Given an economy i 2 N, let Mi be its imports. Then we can verify that Yi = Di + Xi Mi and Xi = (1 ci)Mi + (1 di)Yi, where the second equality tells us that the economy is exports, Xi, can be divided into two components according to its source. Component one consists of the exports of goods that have been re-elaborated from imported goods, (1 ci)Mi . Component two is the exports of new goods, produced by the country, (1 di)Yi. An additional assumption is needed to estimate ci: Assumption 4. The share of re-exported imports of a given economy equals its share of exported production.

The DGI indicator is the most general quantitative approximation to the international integration of economies. This is because it considers not only the degree of openness, but also the distribution and size of the direct and indirect ows between economies. In light of the dierent concepts included in this denition, the indicator will be considered as a Globalization Index for the world economy, according to properties P1 to P4 (the rst three properties because DGI is an increasing function of DI for any economy; property P4 is veried because DGI is a weighted average of the economies degree of integration, where the weight of each economy depends directly on its size). The degree of integration measures how close the world is to the SPII. It should be equal to 1 when

132

WORLD DEVELOPMENT

all economies are perfectly integrated and achieve their theoretical potential of integration, that is, they have neutral degrees of openness (taking into account domestic demand) and the ows between economies are proportional to the share of each economy in the world. (e) Other global indicators In the previous subsections, we have dened several indicators that characterize the integration of each individual economy and that, as the degree of integration, can also be summarized for the whole economy: Degree of global openness: X DGO ai DOi : 12
i2N

Degree of global direct connection: X ai DDC i : DGDC

i2N

13

Degree of global total connection: X ai DTC C : DGTC C i

i2N

14

4. ON THE EVOLUTION OF INTERNATIONAL ECONOMIC INTEGRATION: EMPIRICAL EVIDENCE The international economic integration indicators dened above may well be used to study the evolution of international trade. In this section, we apply our indicators to trade ows, which requires information on the volume of activity for each country, together with their ow exchanges with the rest of the world. The rst subsection details the problems related to information sources and the decisions taken to overcome them. The remaining subsections present results on degrees of openness, connection, and integration. (a) Statistical sources and selected variables The data were taken from the CHELEM database 16 and correspond to 59 countries that together account for 96.7% of world output and 86.5% of international trade. The variable selected to measure ows between countries is the volume of exports. 17 The available information covers a relatively long period of time, from 1967 to 2004, entirely covering what some authors have termed the second wave of globalization (ORourke & Williamson, 1999, 2002; Maddison, 2001). Although the database also contained information for other countries, it was not available for all our sample years, and we therefore disregarded it. The rst three columns in Table 1 report data on GDP shares for each country in our sample. For the sake of simplicity, and also for reasons of space, tables containing individual information for each country in our sample constrain the reported information to three years. That is to say, the initial year (1967), the nal year (2004), and an intermediate year (1985). 18 In both the tables with aggregated data and in the gures (referring to the world economy as a whole, and to each of the largest economies), the annual evolution is reported. All indicators are reported as percentages. (b) Degree of openness The degree of openness dened by considering both exports and GDP in Eqn. (1) is presented in Table 1 and Figure

1. In addition to each countrys share of world output, Table 1 also reports each countrys degree of openness for the selected years, considering information on exports of goods for years 1967, 1985, and 2004 (columns 4, 5, and 6). 19 Figure 1 shows the evolution of both the indicators for all the countries in our sample, reporting information on weighted mean, unweighted mean, and the median. The lower panels in the gure represent the entire distribution using box plots and violin plots 20 corresponding to the three selected years. This enables us to detect the features of the distributions more thoroughly. For the entire world economy, considering the degree of global openness as dened in Eqn. (12), and the corresponding results in Figure 1, the case of exported goods increased from 8% in 1967 to 20.9% in 2004that is, the indicator multiplied by 2.6. Over time, the increase in the degree of openness is not smooth. We can observe stagnant periods (from 1985 to 1995), and even brief periods of reversal.. The unevenness is accentuated at country level. Although positive annual growth rates dominate, some exceptions also exist, especially in the second part of the period (19862004). The unweighted mean is always higher than the weighted mean (see Figure 1) because the degree of openness for the largest economies is lower, even after including the bias correction as suggested in Eqn. (1). The gap between the mean and the median suggests there are countries with quite an extreme degree of openness, especially in the upper tail. The violin plots reinforce this nding, which is stressed over time, showing that some countries have expanded their openness much more than average. Thus, dispersion in the openness indicators for all countries increases. However, due to the increase in the average, the variation coecient declines. Table 1 shows that there are considerable discrepancies among the largest world economies in terms of degree of openness. Since the denition of the indicator controls for domestic bias, the dierences in openness are not directly attributable to this variable. However, this does not necessarily imply that the size eect is negligible. By the end of the period, the high levels achieved by Canada, China, Germany, the former USSR, the Netherlands, and South Korea should be noted, together with the low levels shown by India, Australia, the USA, and Japan. High degrees of openness in the fastest growing countries are noted for China, Canada, the former USSR, Germany, and Mexico. (c) Degree of connection The value for the degree of direct connection indicator (DDC, Eqn. (4)) matches the degree of total connection indicator (DTC, Eqn. (7)) under the hypothesis of c = 1. We also perform computations following an additional hypothesis. In addition, we assume that each country re-exports (see Feenstra & Hanson, 2004) a share of its exports equal to the share of exports in its total output. Thus, we obtain a specic c parameter for each country, as dened in Eqn. (8), which seems more reasonable than a single c value for all the countries. Because of the lack of information on the actual number of transactions, this hypotheses will help us study the importance of indirect connections for the degree of connection and, in a subsequent stage, for the degree of integration. The evolution of the country-specic cs is reported in Figure 2, which shows that c has decreased as a consequence of the growing indirect transactions between countries. Simultaneously, as shown by the violin plots contained in Figure 2, a comparison of 1967, 1985, and 2004 reveals that discrepancies among countries have widened considerably.

MEASURING GLOBALIZATION OF INTERNATIONAL TRADE: THEORY AND EVIDENCE Table 1. Shares of world GDP (a) and degree of openness (DO) (%) ai 1967 Albania Algeria Argentina Australia Austria BLEU Brazil Brunei Daressalam Bulgaria Canada Chile China, Peoples Rep. Colombia Czechoslovakia (former) Denmark Ecuador Egypt Finland France Gabon Germany Greece Hong Kong Hungary Iceland India Indonesia Ireland Israel Italy Japan Malaysia Mexico Morocco Netherlands New Zealand Nigeria Norway Pakistan Peru Philippines Poland Portugal Romania Singapore South Korea Southafrican Union Spain Sweden Switzerland Taiwan Thailand Tunisia Turkey United Kingdom United States USSR (former) Venezuela Yugoslavia (former) Mean Weighted mean Standard deviation Coecient of variation Median NA 0.15 1.07 1.35 0.52 0.90 1.36 0.01 0.31 2.89 0.31 3.20 0.26 0.63 0.56 0.07 0.26 0.41 5.33 0.01 5.61 0.36 0.12 0.20 0.03 2.17 0.27 0.15 0.17 3.56 5.56 0.16 1.21 0.14 1.11 0.27 0.23 0.42 0.33 0.27 0.33 0.80 0.24 0.48 0.05 0.22 0.61 1.39 1.19 0.77 0.16 0.25 0.05 0.71 4.94 37.11 7.78 0.48 0.51 1.72 5.01 2.91 0.42 1985 0.02 0.49 0.75 1.41 0.58 0.73 1.88 0.03 0.27 3.01 0.14 2.58 0.30 0.40 0.51 0.14 0.40 0.46 4.50 0.03 5.37 0.35 0.30 0.18 0.02 1.90 0.74 0.17 0.20 3.61 11.48 0.27 1.64 0.11 1.12 0.19 0.24 0.54 0.25 0.14 0.26 0.64 0.21 0.40 0.15 0.84 0.51 1.45 0.89 0.83 0.52 0.33 0.07 0.57 3.86 35.72 4.43 0.52 0.34 1.69 4.87 2.87 0.49 2004 0.02 0.20 0.38 1.55 0.73 0.96 1.43 0.01 0.06 2.50 0.23 4.13 0.24 0.37 0.61 0.07 0.17 0.46 5.11 0.02 6.87 0.51 0.43 0.25 0.03 1.72 0.57 0.45 0.29 4.24 11.90 0.29 1.71 0.13 1.46 0.24 0.16 0.63 0.20 0.17 0.22 0.59 0.42 0.18 0.27 1.72 0.49 2.48 0.87 0.91 0.79 0.43 0.07 0.80 5.47 30.26 2.00 0.27 0.26 1.69 4.29 2.53 0.45 1967 NA 19.46 5.44 10.00 14.29 33.07 5.89 58.50 2.92 16.36 11.10 1.54 7.80 4.63 18.70 13.68 9.52 15.42 8.68 30.90 17.30 5.81 39.50 8.92 17.86 2.62 14.65 21.84 12.86 10.18 7.36 32.63 4.24 12.24 24.95 16.64 13.89 15.92 5.59 10.87 10.06 4.61 9.29 4.01 37.91 6.15 9.02 4.11 15.82 19.45 16.77 8.83 9.80 3.14 11.24 5.24 1.48 13.43 10.33 13.30 7.94 10.66 0.80 10.60 DOi 1985 7.26 21.77 8.38 10.97 23.61 58.61 10.37 71.40 2.95 24.22 21.16 6.64 8.96 6.92 25.25 18.42 9.76 24.10 17.14 52.00 29.61 9.50 45.09 14.61 33.73 3.74 21.52 48.53 22.56 16.81 13.00 46.06 14.37 17.44 55.21 20.65 45.41 31.43 6.28 16.33 16.45 5.87 21.16 8.19 76.50 26.05 15.87 12.36 27.92 25.90 47.24 14.93 17.17 7.63 21.83 7.26 5.17 18.59 25.60 22.77 14.21 16.80 0.74 18.42

133

2004 7.36 40.47 20.61 13.22 35.20 82.00 15.53 84.36 38.44 33.12 32.48 36.40 16.48 63.76 28.43 22.94 7.58 31.37 20.58 37.62 34.60 6.85 11.79 54.45 28.16 9.28 28.40 58.46 31.88 20.83 13.12 112.95 28.32 19.30 50.05 18.86 43.41 32.92 12.95 17.83 51.36 31.00 20.77 31.67 92.44 34.87 20.68 18.14 35.50 32.43 53.59 58.48 31.65 16.98 16.41 8.34 31.70 7.40 28.98 32.62 20.85 21.71 0.67 31.00

134

WORLD DEVELOPMENT

40

Degree of openness (DO)

Degree of openness (DO)

30

20

10

20

40

60

80

100

1970

1980
Year

1990

2000

1967

1985
Year

2004

Unweighted mean

Median

Weighted mean

Figure 1. Degree of openness (DO), 19672004. Summary statistics and violin plots. *Violin plots are a mix between box plots and density functions. Box plots show four main features of a variable: center, spread, asymmetry, and outliers. The density trace supplements this information by graphically showing the distributional characteristics of batches of data such as multi-modality.

95

90

Country specific

Countryspecific

85

80

75

70

30

40

50

60

70

80

90

100

1970

1980

1990

2000

1967

1985

2004

Year
Unweighted mean Median Weighted mean

Year

Figure 2. Country-specic cs, 19672004. Summary statistics and violin plots. * Violin plots are a mix between box plots and density functions. Box plots show four main features of a variable: center, spread, asymmetry, and outliers. The density trace supplements this information by graphically showing the distributional characteristics of batches of data such as multi-modality.

Table 2 reports information on the degree of connection for each country (DTC). The same information for all the countries as a whole (DGTC) is reported in Figure 3 (weighted mean). In both the cases, the dierences between the two indicators (DTC and DGTC) are remarkable, even when the number of indirect connections is assumed to be very low. Dierences are more important if the number of indirect connections increases, as shown by the estimation for country-specic c. However, since c is country-specic, departures from the results obtained for c = 1 vary depending on the country analyzed. Some countries DTC increases sharply when accounting for indirect connections (see, for year 2004, Belgium and Luxembourg, Brunei Daressalam, former Czechoslovakia, Hungary, Malaysia, The Netherlands, Norway, Singapore, Taiwan, former USSR, or former Yugoslavia, for instance). On the other hand, there are rather modest dierences for other countries (see, for year 2004, Brazil, Canada, Colombia, India, Israel, Japan, Mexico, Pakistan, or Peru, for instance). This would imply that the full potential for indirect connections is remarkable. A country with a low degree of direct connection with the rest of the world could have a high degree of total connection as a result of the variety of itineraries oered by the world trade network.

The mean values for the degree of direct connection (DDC, or DTC for c = 1) are higher than those for the degree of openness (DO). They are especially high for some countries, many of which exhibit values of over 80%. When we consider the possible existence of indirect connections, the degree of connection increases noticeably. Table 2 reveals this eect for our set of economies. In 2004, the degree of direct connection was 64.6% (weighted mean), whereas the degree of total connection was higher (69.0%) when considering country-specic c. If we consider that c is constant, the time trend for the degree of connection indicators is of moderate growth, that is, countries widen their trade networks with the rest of the world, and attempt to balance them according to the size of their export markets. However, by weighing in not only time but also the number of transactions (c ), we perceive a larger increase, from 59.5% in 1967 to 69.0% by 2004 (for countryspecic c). Although c is dicult to measure and might vary greatly depending on the commodity considered, the evidence suggests that it has decreased substantially over the past 40 years due to the current trends in oshore outsourcing and delocalization (Figure 2). 21

MEASURING GLOBALIZATION OF INTERNATIONAL TRADE: THEORY AND EVIDENCE Table 2. Degree of total connection (DTC) for c = 1 and country-specic c, individual countries, 19672004 (%) DTCi (c = 1) 1967 Albania Algeria Argentina Australia Austria BLEU Brazil Brunei Daressalam Bulgaria Canada Chile China, Peoples Rep. Colombia Czechoslovakia, former Denmark Ecuador Egypt Finland France Gabon Germany Greece Hong Kong Hungary Iceland India Indonesia Ireland Israel Italy Japan Malaysia Mexico Morocco Netherlands New Zealand Nigeria Norway Pakistan Peru Philippines Poland Portugal Romania Singapore South Korea Southafrican Union Spain Sweden Switzerland Taiwan Thailand Tunisia Turkey United Kingdom United States USSR, former Venezuela Yugoslavia, former Mean Weighted mean Standard deviation Coecient of variation Median NA 20.70 54.85 59.69 35.95 41.42 91.61 0.40 25.70 95.67 63.30 15.13 92.72 30.08 41.04 86.75 38.19 44.92 41.61 27.99 54.24 66.22 92.50 24.37 66.22 71.08 44.29 26.86 77.94 58.59 94.72 64.75 95.04 28.64 33.61 44.73 37.97 43.85 70.57 90.33 76.19 52.46 54.00 28.41 22.39 60.36 84.26 74.36 46.99 65.09 95.08 52.24 27.88 80.60 76.27 56.92 34.45 92.41 45.33 55.52 57.86 24.61 44.33 54.12 1985 21.88 66.84 70.92 60.07 32.43 38.94 97.61 29.07 32.63 93.93 85.86 71.46 93.78 24.18 53.16 89.66 28.80 38.01 52.05 70.40 58.16 46.84 93.99 38.69 83.44 86.96 66.36 40.33 94.86 62.92 96.68 62.28 94.38 29.57 33.66 69.97 78.56 35.25 76.82 98.05 95.30 36.70 51.82 73.34 78.57 76.08 96.62 60.25 61.92 57.39 97.23 87.37 23.42 50.33 71.69 63.62 30.07 94.85 30.70 63.33 67.36 24.41 38.53 63.62 2004 13.95 75.53 52.27 62.90 43.58 48.38 88.30 47.83 39.31 88.36 84.55 81.32 87.28 32.60 49.33 88.31 58.88 50.25 54.90 88.89 62.58 48.77 70.51 35.33 52.83 87.87 75.12 78.03 91.54 62.05 82.25 75.74 87.01 32.74 42.19 67.55 89.22 53.02 93.06 92.11 73.03 34.56 42.14 37.31 60.42 76.66 73.42 43.42 68.72 64.16 65.70 86.32 27.65 59.92 77.25 56.97 24.42 82.96 28.51 62.71 64.60 20.84 33.24 62.89 1967 NA 25.66 57.04 63.76 40.48 48.66 91.48 23.16 27.29 95.58 66.48 16.69 92.81 32.36 46.59 87.84 42.31 49.51 44.26 37.19 58.24 66.75 92.59 28.46 68.24 71.90 25.89 32.07 78.76 60.33 94.79 73.58 95.03 32.93 41.36 49.62 43.03 48.22 72.05 90.81 77.55 53.75 56.18 30.17 36.52 84.67 63.07 74.76 51.66 68.21 94.97 55.91 30.79 80.68 77.85 57.71 35.57 92.13 48.31 58.35 59.46 22.74 38.97 56.61 DTCi (country-specic c) 1985 26.63 69.84 74.29 66.99 40.15 53.85 97.51 80.65 35.66 94.16 89.45 74.42 93.88 36.63 59.65 90.76 33.41 47.02 56.45 80.46 64.19 49.39 95.29 50.53 84.71 87.55 74.59 54.93 94.87 66.02 96.89 79.65 94.58 36.98 48.18 77.61 82.21 46.17 79.34 98.31 95.95 41.77 56.22 74.68 89.39 97.00 82.22 63.21 67.20 63.82 97.53 89.28 30.67 52.97 74.18 64.66 34.14 94.99 39.33 69.04 69.91 21.51 31.16 74.18

135

2004 15.42 80.91 63.68 69.18 52.37 66.02 89.68 85.13 49.24 88.85 90.00 87.01 88.25 46.35 59.07 90.38 63.01 61.09 59.77 93.10 69.04 50.61 72.28 47.21 59.23 88.59 82.58 83.25 92.90 66.13 83.37 89.73 87.50 39.10 55.43 73.44 92.87 63.68 93.37 92.90 81.85 41.65 47.27 45.84 79.76 80.54 81.62 49.32 74.66 70.85 77.44 90.12 36.42 63.62 78.73 58.62 61.93 90.18 42.82 70.08 69.01 18.36 26.20 72.28

136

WORLD DEVELOPMENT
Degree of total connection (DTC), country-specific

90

Degree of total connection (DTC), =1

80

70

60

50

1970

1980

1990

2000

50

60

70

80

90

1970

1980

1990

2000

Year
U n w e ig h te d me a n Me d ia n W e ig h ted m ean Unwei ght ed m ean

Year
M edi an Wei ght ed m ean

100

Degree of total connection (DTC), country-specific

1967 1985 2004

Degree of total connection (DTC), =1

80

60

40

20

20

40

60

80

100

1967

1985

2004

Y ear

Ye a r

=1
*

Country-specific

Figure 3. Degree of total connection (DTC), c = 1 and country-specic c, 19672004. Violin plots are a mix between box plots and density functions. Box plots show four main features of a variable: center, spread, asymmetry, and outliers. The density trace supplements this information by graphically showing the distributional characteristics of batches of data such as multi-modality.

As a whole, dispersion in the degree of connection tends to diminish over time, in both absolute and relative terms. It is important to realize that when indirect connections are taken into account, and these increase in number, economies become much more similar in their degrees of total connection, as suggested by the sharp decline of dispersion indicators (Table 2). Figure 3 shows that the values for the weighted degree of total connection (DGTC) are slightly higher than those corresponding to the unweighted mean. In contrast to what occurs with the degree of openness, large economies tend to connect with the rest in a more balanced way than smaller economies do. A further dierence in the degree of openness is that now both the mean and the median are very close, suggesting that both tails of the distribution are not very important for the degree of connection. However, the violin plots in Figure 3 indicate that the distribution of the degree of connection shows a fairly stable dispersion over time, and it is bimodal. Therefore, there are two groups of economies with dierent degrees of connection. The rst group is concentrated around a high degree of connection values, higher than 80%, which is equivalent to being connected in a balanced way with all other countries. In contrast, the mode of the second group is located around lower values, close to 40%. For countries in the second group, what occurs to the indirect connections will be more relevant. It is also interesting to note that the second group has ostensibly been losing weight over time.

The degree of connection (DTC) also varies greatly among the largest economies (see Table 2). Some examples of countries with high degrees of connection by 2004 are Canada, most Asian economies (China, South Korea, India, Japan), Brazil, and Mexico. Some large economies also have low degrees of connection, among which we nd several European countries such as the former USSR, the Netherlands, or Spain. The general tendency for the degree of direct connection is to increase. However, not all countries follow the same pattern, and for some of them the balance in their external connections is declining because they export only to specic trade partners. This is the case of Canada and, notably, of some European countries (Iceland, Spain, Greece, Portugal, former Yugoslavia, former Czechoslovakia, Poland, Romania), some Latin American countries (Argentina, Colombia), and some Asian countries (Thailand, Hong Kong). The decline in the degree of direct connection indicates that these countries trade more with economies whose weight in the exporter countries exports is larger than that corresponding to the importing countries according to their share of world output. The list of countries showing this behavior enables us to establish a hypothesis, the testing of which would require an additional investigation. That is to say that the current international economic integration processes in dierent parts of the world have an impact on the structure of trade connections. In the European case, the eect seems particularly

MEASURING GLOBALIZATION OF INTERNATIONAL TRADE: THEORY AND EVIDENCE Table 3. Degree of integration (DI) for c = 1 and country-specic c, individual countries, 19672004 (%) DIi (c = 1) 1967 Albania Algeria Argentina Australia Austria BLEU Brazil Brunei Daressalam Bulgaria Canada Chile China, Peoples Rep. Colombia Czechoslovakia, former Denmark Ecuador Egypt Finland France Gabon Germany Greece Hong Kong Hungary Iceland India Indonesia Ireland Israel Italy Japan Malaysia Mexico Morocco Netherlands New Zealand Nigeria Norway Pakistan Peru Philippines Poland Portugal Romania Singapore South Korea Southafrican Union Spain Sweden Switzerland Taiwan Thailand Tunisia Turkey United Kingdom United States USSR, former Venezuela Yugoslavia, former Mean Weighted mean Standard deviation Coecient of variation Median NA 20.07 17.28 24.43 22.66 37.01 23.24 4.83 8.67 39.56 26.51 4.83 26.90 11.81 27.70 34.45 19.07 26.32 19.00 29.41 30.63 19.62 60.45 14.74 34.40 13.64 25.48 24.22 31.66 24.42 26.40 45.97 20.07 18.73 28.96 27.28 22.96 26.42 19.87 31.34 27.69 15.56 22.40 10.67 29.13 22.76 23.34 17.48 27.26 35.58 39.93 21.48 16.53 15.90 29.28 17.28 7.14 35.23 21.64 24.26 20.26 9.88 40.71 23.78 1985 12.61 38.15 24.38 25.67 27.67 47.77 31.82 45.56 9.80 47.70 42.63 21.78 28.99 12.94 36.64 40.64 16.76 30.27 29.87 60.50 41.50 21.10 65.10 23.77 53.05 18.05 37.79 44.24 46.26 32.52 35.45 53.56 36.83 22.71 43.11 38.01 59.72 33.28 21.97 40.01 39.59 14.68 33.11 24.51 77.53 50.17 34.75 27.29 41.58 38.55 67.77 36.12 20.05 19.60 39.56 21.49 12.47 41.99 28.03 35.07 29.07 14.51 41.36 35.45 2004 10.13 55.29 32.82 28.84 39.17 62.99 37.04 63.52 38.87 54.10 52.40 54.41 37.92 45.59 37.45 45.01 21.12 39.70 33.61 57.83 46.53 18.28 28.83 43.86 38.57 28.56 46.19 67.54 54.02 35.96 32.85 92.49 49.64 25.14 45.96 35.69 62.23 41.78 34.71 40.53 61.25 32.73 29.58 34.37 74.73 50.60 39.81 28.06 49.39 45.61 59.34 71.05 29.58 31.89 35.60 21.79 27.82 24.77 28.75 42.27 34.52 15.27 36.13 39.17 1967 NA 22.35 17.62 25.25 24.05 40.12 23.22 36.81 8.93 39.55 27.17 5.08 26.91 12.25 29.52 34.66 20.07 27.63 19.60 33.90 31.74 19.70 60.48 15.93 34.92 13.71 18.09 26.47 31.83 24.78 26.41 49.00 20.07 20.08 32.12 28.73 24.45 27.70 20.07 31.42 27.93 15.75 22.85 11.00 37.21 22.82 23.86 17.53 28.59 36.42 39.91 22.22 17.37 15.91 29.58 17.40 7.25 35.17 22.34 25.58 20.61 10.03 39.20 24.61 DIi (country-specic c) 1985 13.91 38.99 24.95 27.11 30.79 56.18 31.80 75.88 10.25 47.76 43.51 22.22 29.00 15.92 38.81 40.89 18.05 33.66 31.11 64.68 43.60 21.66 65.55 27.17 53.45 18.11 40.06 51.63 46.26 33.31 35.48 60.57 36.87 25.39 51.58 40.03 61.10 38.09 22.32 40.06 39.72 15.66 34.49 24.74 82.69 50.27 36.12 27.95 43.32 40.65 67.88 36.51 22.95 20.11 40.24 21.67 13.29 42.03 31.73 37.29 29.86 15.91 42.68 36.51

137

2004 10.65 57.22 36.22 30.25 42.94 73.58 37.32 84.74 43.51 54.25 54.07 56.28 38.13 54.36 40.98 45.53 21.85 43.77 35.07 59.18 48.88 18.62 29.19 50.70 40.84 28.68 48.43 69.76 54.42 37.12 33.07 100.67 49.78 27.47 52.67 37.22 63.50 45.78 34.77 40.70 64.84 35.93 31.33 38.10 85.87 52.99 41.08 29.91 51.48 47.93 64.42 72.59 33.95 32.86 35.94 22.11 44.31 25.83 35.23 45.40 36.10 17.01 37.47 42.94

138

WORLD DEVELOPMENT

strong, especially for most of the countries that joined the European Union in its various enlargements. In most of these cases, the value of the degree of direct connection not only declines, but is also low (below 0.5 in 2004), whereas the world average is higher and has also increased. These ideas might partly explain why Spain or the Netherlands have low degrees of connection. Appropriate answers could therefore relate to the existence of regional agreements with pernicious collateral eects, given that although trade intensity with other agreement members increases, it might decline with respect to nonmembers. Comparing summary statistics reported in the last ve rows of Table 2 shows the relevance of indirect connections in increasing and homogenizing the degree of total connections between economies. For economies with low degrees of direct connection, indirect connections are more relevant, since they can considerably improve the degree of total connection. In addition, when we also consider the indirect itineraries, some economies which showed a tendency toward disconnection now show a more stable evolution due to their strong links to economies that are much better connected to the rest. On the other hand, economies such as former USSR show a reverse pattern when comparing results for c = 1 and for country-specic c.

(d) Degree of integration Integration indicators uncover the combined eect of openness and balance of connection, and are presented in Table 3 and Figure 4. In general, this eect has increased for all economies, with few exceptions. When considering only direct connections, the average increased from 20.3% in 1967 to 34.5% by 2004 (see Table 3). If we take into account indirect connections (country-specic c), the degree of integration rises even more (from 20.6% in 1967 to 36.1% in 2004). However, the degree of global integration is still far from its full theoretical potential, characterized by the SPII. Figure 4 indicates that integration for large economies is lower, as shown by the lower values for the weighted as compared to the unweighted mean. In addition, the progress of integration is slightly less intense among large economies, since the rate of growth of the weighted indicator is slightly lower. The dispersion shown by the degrees of integration is remarkable, although it tends to diminish when the coecient of variation is considered, which controls for the growing average eect. Integration for some countries is quite high, as revealed by the violin plots, which show that the most advanced countries have values over 60%. On the other hand, for the most backward economies it hardly reaches 40%.

60

Degree of integration (DI), country-specific

50

Degree of integration (DI), =1

40

30

20

10

10

20

30

40

50

60

1970

1980
Year

1990
Median

2000
Weighted mean

1970

1980
Year

1990
Median

2000
Weighted mean

Unweighted mean

Unweighted mean

100

Degree of integration (DI), country-specific

80

Degree of integration (DI), =1

60

40

20

1967

1985
Year

2004

20

40

60

80

100

1967

1985

2004

Year

=1
*

Country-specific

Figure 4. Degree of integration (DI), c = 1 and country-specic c, 19672004. Violin plots are a mix between box plots and density functions. Box plots show four main features of a variable: center, spread, asymmetry, and outliers. The density trace supplements this information by graphically showing the distributional characteristics of batches of data such as multi-modality.

MEASURING GLOBALIZATION OF INTERNATIONAL TRADE: THEORY AND EVIDENCE

160

139

1970

1980

1990

2000

Year

DOi / DI i . (=1) DOi / DI i . (country-specific ) DTCi / DI i . (=1) DTCi / DI i . (country-specific )

------------

Figure 5. Evolution of the contribution  the degree of openness of p p DTC =DI i to the  DOi =DI i and the degree of connection p ip degree of integration (means), 19672004 DOi =DI i DTC i =DI i 1 .

The degree of integration has also grown in most cases because of its driving factors: the degree of openness and the balance in the connection. The importance of each factor can be seen from Eqn. (10) and is shown in Figure 5. In general, the contribution of the degree of connection is larger, although its weight decreases over time. On the contrary, the opposite holds for the degree of openness. (e) How do the dierent indicators relate? Table 4 presents Spearman correlation matrices between different indicators of international economic integration, competitiveness, and globalization for years 1967, 1985, and 2004. In the case of the KOF index, computations were not performed for 1967 due to the lack of information (instead, they were performed for 1970). Notable among these results is the low, and negative, correlation between DO and DTC, regardless of its type (c = 1 or country-specic c), which shows their independence from each other. This result is not surprising, since both the indexes convey dierent sorts of information. In contrast, as shown by Table 4, correlations between DO and DI are high, indicating that the degree of openness is quite relevant in explaining the degree of integration distribution. Therefore, the picture we obtain is not entirely coincidental with that obtained when only analyzing the degree of openness. For some countries, the degree of total connection endorses its economic integration with the rest of the world, whereas for others it has the opposite eectthat is, there are countries which are very open but they do not contribute to the general increase of the degree of global integration. The nal picture emerges when controlling for indirect connec-

tions, that is, when considering not only results for c = 1, but also for country-specic c, for which correlations between DO and DTC are higher. However, this result was easy to forecast due to our assumption about the country-specic c parameter, according to which its value hinges on the share of its exports on the sum of imports plus total output. Over time, the relation between these indicators has uctuated, and no pattern seems to prevail. We might partly expect the low correlation between DO and DTC, which reects that both the indices are measuring dierent dimensions of the globalization process, and that economies with a high (low) level of openness tend to export more to small (large) economies. All the three indicators (DO, DTC, and DI) can be compared with each other and also with other indicators of economic internationalization to analyze their similarities and the ability of our indicators to contribute new yardsticks to interpret the evolution of integration. We have considered three sets of those indicators. The rst set includes standard indicators such as the traditionally dened degree of openness (i.e., (X + M)/ GDP), trade balance ((X M)/GDP), and the comparative advantage index ((X M)/(X + M)). The second set considers some weighted statistics of network analysis fagiolo.reyes.schiavo.wp07 such as node strength NS (the exports over GDP scaled so that the maximum over countries will be one), average of nearest-neighbor strengths ANNS (the average of NS of all partners of a given country), weighted average of nearestneighbor node degrees WANND (the weighted average of number of partners of all partners of a given country, where the weight is dened by NS), and the weighted clustering coefcient CC (the number of triangles around a country weighted using weights associated to its three sides). The last set contains the KOF index, an index of globalization covering three main dimensions: economic integration, social integration, and political integration (see Dreher, 2006). As we might expect, the correlation between DO and (X + M)/GDP is high, and quite close to unity. Nonetheless, the pattern is decreasing, suggesting both imports and exports could be evolving dierently. Correlation between DO and (X M)/GDP, or (X M)/(X + M) is far less important than in the case mentioned above. In addition, the pattern here is of increasing correlation (in contrast to what occurred when comparing DO with (X + M)/GDP). As outlined above, explanations may be related to the fact that the X M dierence has not remained constant over the last 40 years. The degree of connection (DTC), similarly to what occurred when it was compared with the DO, presents a slightly negative correlation with (X + M)/GDP, for all the three years. In turn, its correlations both with the trade balance and the comparative advantage indexes are positive, higher and increasing over time, stressing the role of imports. Correlations are also higher for country-specic c s. The degree of integration shows similar relatively high correlations with (X + M)/GDP, the trade balance ((X M)/GDP), and the comparative advantage ((X M)/(X + M)). Since DI contains information on both DO and DTC, it is more strongly correlated with the indicators of trade balance and comparative advantage, that is, the two indicators containing the X M subtraction. Therefore, we end up with an indicator of international economic integration that is positively and increasingly correlated with traditional indicators of economic internationalization. Specically, by 2004, and considering country-specic c, correlations are always around 0.70 for either (X + M)/GDP, (X M)/GDP, or (X M)/(X + M). The NS index shows a close to 1 correlation with the DO index given that, except for the domestic bias correction, both are equally dened. Thus, the analysis of the correlation be-

%
60 80 100

120

140

140

WORLD DEVELOPMENT Table 4. Spearman correlation matrices among the dierent indicators, 1967, 1985, and 2004 (1) (2) .1414 1.000 (3) .0768 .9720 1.000 (4) .7148 .4660 .5162 1.000 (5) .8532 .2950 .3772 .8986 1.000 (6) .9552 .1761 .1136 .6450 .7886 1.000 (7) .0932 .0472 .0358 .1218 .1579 .1324 1.000 (8) .2947 .0230 .0225 .2587 .3107 .0581 .9465 1.000 (9) .9947 .1396 .0772 .7106 .8494 .9582 .0772 .2761 1.000 (10) .3506 .0565 .0600 .3239 .2475 .3568 .1771 .0922 .2917 1.000 (11) .2632 .2725 .2732 .4454 .3331 .2533 .0798 .1268 .2801 .0835 1.000 (12) .5373 .1199 .0431 .3972 .4936 .4429 .2175 .3826 .5264 .0313 .0367 1.000 (13)a,b .4218 .0771 .0367 .3064 .3285 .4099 .0687 .0225 .4066 .2661 .0161 .7222 1.000 (13)b .4897 .3314 .2999 .2646 .3074 .5120 .1353 .0892 .4650 .2833 .0654 .6967 1.000 (13)b .2541 .3264 .3305 .0959 .1307 .2886 .0553 .1158 .2431 .2253 .2873 .6553 1.000

Year 1967 (1) 1.000 (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) Year 1985 (1) (1) 1.000 (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) Year 2004 (1) (1) 1.000 (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)

(2) .0528 1.000

(3) .1712 .9399 1.000

(4) .8602 .5062 .5995 1.000

(5) .9015 .4186 .5379 .9910 1.000

(6) .9595 .0603 .0483 .7680 .8124 1.000

(7) .3394 .3027 .4123 .4723 .4770 .1213 1.000

(8) .4123 .3327 .4393 .5344 .5370 .1980 .9523 1.000

(9) .9973 .0508 .1710 .8574 .8979 .9601 .3397 .4150 1.000

(10) .5355 .0745 .0464 .4301 .4271 .5567 .0094 .0578 .5321 1.000

(11) .3111 .0642 .1265 .3751 .3754 .2510 .3430 .3642 .3222 .0389 1.000

(12) .5231 .0065 .0219 .4113 .4227 .5255 .0034 .0074 .4930 .3087 .0137 1.000

(2) .0834 1.000

(3) .0853 .9429 1.000

(4) .8393 .4056 .5254 1.000

(5) .9129 .2596 .4228 .9666 1.000

(6) .8482 .3023 .1658 .5969 .6809 1.000

(7) .5598 .4391 .5667 .7260 .7279 .1748 1.000

(8) .5476 .4334 .5645 .7087 .7205 .1863 .9577 1.000

(9) .9977 .0847 .0842 .8388 .9088 .8558 .5492 .5336 1.000

(10) .7519 .3009 .1595 .5741 .6469 .7252 .2814 .2980 .7583 1.000

(11) .2286 .3206 .2973 .0015 .0964 .2519 .0021 .0236 .2220 .2339 1.000

(12) .3437 .1434 .1534 .2347 .2745 .3015 .1646 .1759 .3064 .2409 .0395 1.000

(1) DO; (2) DTC (c = 1); (3) DTC (country-specic c); (4) DI (c = 1); (5) DI (country-specic c); (6) (X + M)/GDP; (7) (X M)/GDP; (8) (X M)/ (X + M); (9) NS; (10) ANNS; (11) WANNS; (12) CC; (13) KOF index. a The correlations in this column correspond to the year 1970, due to KOF index availability for the years before. b Missing countries, due to unavailability of KOF index: Yugoslavia (former), Hong Kong, Brunei Daressalam, USSR (former), and Czechoslovakia.

tween DO and the network analysis indexes are similar to that in Fagiolo et al. (2007). Briey, the high and negative correlation between DO and ANNS, and the high, yet less important, positive correlation with WANND indicate that the more open countries export to closed countries, but ones that are in contact with many other countries. The DTC indexes (both, c = 1 or country-specic c) present a slight correlation with the four network analyses indexes, where the sign, positive or negative, depends on the year analyzed. The low correlation of both second-order trade indexes ANNS and WANND with the higher-order index DTC indicates that second-order does not suce to uncover the whole trade interrelations structure among countries. Therefore, higher-order indexes are needed. Moreover, the decreasing

correlation pattern between DTC and WANND over time suggests that the relevance of higher-order links has increased more than that of second-order links. The correlation of DI with NS, ANNS, WANND, and CC falls between that corresponding to DO and DTC, as the definition of DI indicates. Finally, we have also compared our indicators with the KOF index of globalization (Dreher, 2006). This index is closer to the concept of globalization found in the media than to the concept of international economic integration which is, as suggested by Rodrik (2000), self-evident to economists. Specically, whereas our indexes only take into account the trade-related aspects of globalization, the KOF index is far more comprehensive, as it is made up of economic

MEASURING GLOBALIZATION OF INTERNATIONAL TRADE: THEORY AND EVIDENCE Table 5. Relative positions between degree of openness (DO) and degree of total connection (DTC), 19672004 Upper limit DO 0.403 (a) c = 1 0.089 0.151 0.210 0.311 1.129 0.23 0.16 0.16 0.22 0.23 0.474 (b) Country-specic c 0.089 0.151 0.210 0.311 1.129 0.29 0.19 0.19 0.20 0.13 0.542 0.20 0.17 0.23 0.21 0.19 0.598 0.20 0.19 0.23 0.20 0.17 Upper limit DTC 0.692 0.24 0.22 0.17 0.23 0.13 0.753 0.24 0.16 0.18 0.25 0.16 0.878 0.16 0.19 0.22 0.17 0.27 0.897 0.10 0.18 0.19 0.21 0.31 0.983 0.16 0.27 0.21 0.17 0.18 0.985 0.16 0.29 0.21 0.13 0.23 448 448 447 448 451 449 448 448 448 449

141

Number

globalization (34%, comprising both data on actual ows and restrictions), social globalization (37%, comprising data on personal contact, information ows, and cultural proximity), as well as political globalization (28%). Results show that the KOF index is positively related to DO, although its magnitude diminishes over time. On the other hand, it is negatively related to the DTC. In this case, the relation is rather unstable over time. These dierences were easy to predict, as both indexes overlap only to a certain degree. In addition, results were never going to coincide, as there were some countries for which the KOF index was unavailable. Although correlation coecients convey relevant information, more details are available in Table 5, which provides the distribution of DTC conditional on the distribution of DO. Specically, we estimate conditional probability matrices for each value of c. This type of matrix involves dividing the space of indicators into dierent classes. The matrices track changes in the distribution of one indicator (say, DTC) as the other (say, DO) moves from one class to other. The class limits, or grids, are chosen in accordance with a certain criterion. We consider ve classes each encompassing 20% of the values of an indicator, arranged in increasing order, that is, class 1 covers the lowest, and class 5 covers the highest openness. The conditioned probability then uses an unweighted average of observed frequencies to estimate

the probability that a country in one class according to DO will be in another class according to DTC. 22 Thus, we have evidence on the dierent paths followed by dierent countries to achieve a certain degree of integration (DI), that is, it might be due to either higher openness, or to a higher degree of connection, or to a combination of the two in similar proportions, etc. The possible combinations are multiple. Results are shown in Table 5a and b. Given c, if there is no dierence in the values of any two rows, we can conclude that the indicators DO and DTC are independent, that is, a countrys degree of openness does not give us any knowledge about its degree of connection and vice versa, as can be seen in reality. Moreover, each value in a xed row is close to 20%. A given class of DO is equally likely to belong to any class of DTC. This is because of the independence between these indicators, as well as between classes. For instance, the upper-left cell in Table 5a indicates that the 20% least open countries, with DOi < 0.089, have a 0.23 probability of having a low degree of connection, DTCi < 0.403. But in fact, the probability of their having a medium or a high degree of connection is 0.24 and 0.16, respectively, all probabilities lying close to 0.20. These tendencies are similar for country-specic c, that is, the general tendency is for any class of DO to have an equal probability of belonging to each class of DTC. Finally, the last column in Table 5a and b shows the number of countryyear

Table 6. Degree of openness, degree of total connection, and degree of integration, weighted means, country groups according to geography, 2004 Region Western Europea Eastern Europeb North Americac South Americad Africae Asia (South and Southeastern)f Asia (Eastern)g Australasiah DO c=1 27.80 32.64 11.12 17.26 25.76 34.94 21.82 13.98 59.85 37.68 60.73 82.73 68.46 82.05 80.29 63.52 DTC Country-specic c 65.64 57.66 62.24 86.32 73.14 87.13 83.51 69.75 c=1 39.41 33.24 25.51 37.06 40.80 46.89 40.16 29.76 DI Country-specic c 41.59 42.21 25.81 37.95 42.39 49.20 41.15 31.18

a France, BLEU, Germany, Italy, Netherlands, United Kingdom, Ireland, Denmark, Finland, Norway, Sweden, Iceland, Austria, Switzerland, Spain, Greece, Portugal. b Former Yugoslavia, Former USSR, Bulgaria, Former Czechoslovakia, Hungary, Poland, Romania, Albania, Turkey, Israel. c United States, Canada, Mexico. d Venezuela, Ecuador, Brazil, Argentina, Chile, Colombia, Peru. e Southafrican Union, Algeria, Morocco, Tunisia, Egypt, Nigeria, Gabon. f Indonesia, Singapore, Malaysia, Philippines, Thailand, Brunei Daressalam, India, Pakistan. g Japan, South Korea, Hong Kong, Taiwan, China, Peoples Rep. h Australia, New Zealand.

142

WORLD DEVELOPMENT

pairs in each DO class. For instance, the rst row indicates that there were 449 countryyear pairs with DOi < 0.089. Finally, we can also analyze how the indicators relate to different country groupings. We nd that it is dicult to identify any sort of clear association between the average values of the two dimensions of integration according to geographical areas (Table 6), according to economic size (GDP) (Table 7), or according to per capita GDP (Table 8). However, this classication reveals distinctive features for some of the indicators. In particular, we note how the Eastern European economies stand out for their high levels of openness and low levels of connectedness. In contrast, those in South and Southeast Asia excel in both variables. We can also point to low openness in North America, Australasia, and South America, although for this case DTC is the highest indicator. The highest DI levels are found in Asia (South and Southeast), whereas North America ranks lowest. The impact of both economic size and per capita GDP on degree of openness presents an inverted-U shape, with the highest levels corresponding to intermediate stages. From this perspective, we do not note any particular pattern for either DTC or DI. 5. CONCLUSIONS: IS GLOBALIZATION ADVANCING? The aim of this study was to present international economic integration and globalization indicators that take into account both the growing degree of openness in economies, and the development of direct and indirect network connections. To do this, we approached the characterization of the indicators and their properties from a network analysis perspective. We considered several conditions of geographic neutrality, dening the distance separating each countrys economy and the

world economy from a Standard of Perfect International Integration. When we applied the indicator of integration to our set of countries, we obtained a measure of the level of globalization achieved. If integration reached the level of the SPII, the relations between economic agents in dierent countries would not be biased or inuenced by location, and we would have arrived at the stage known as the global village. To illustrate the potential of the SPII, we applied the proposed methodology to export ows, which provided us with some responses to the question of the distance that separates us from the situation of complete trade globalization. The distance to the theoretical potential of trade integration is still considerable, since we have not reached the halfway point. However, the ground covered over the last forty years is quite remarkable, as it represents advances in international economic integration of more than 75%. Results also indicate that dierences between countries in this vein are notable. Leading positions can be observed for some, especially for some small European Union economies (Belgium and Luxembourg, Ireland) or Southeast Asia (Malaysia, Singapore, Thailand, or Brunei Daressalam), in which the total integration indicators are quite high, twice the average. The methodology proposed therefore oers a starting point to assess the importance of the advance of globalization, and also the contribution made to it by the two components that either jointly constitute the integration process or limit its scope: the degree of openness (DO) and each economys balance in the connections network with the other economies (DTC). The results point rstly to the fact that domestic bias aecting trade (which limits the degree of openness) represents the highest limit to integration. Although its importance is declining, this hindrance is more important for large economies,

Table 7. Degree of openness, degree of total connection, and degree of integration, weighted means, country groups according to GDP, 2004 Group
a

DO c=1

DTC Country-specic c 68.58 71.10 66.88 75.55 69.36 c=1 32.02 43.75 41.46 48.95 37.87

DI Country-specic c 33.20 46.88 43.87 53.06 39.70

Group Group Group Group Group

a b

1 2b 3c 4d 5e

17.80 33.97 31.22 43.66 24.91

64.92 63.77 60.77 66.64 64.53

United States, Japan, Germany, United Kingdom, France, Italy, China (Peoples Republic), Canada, Spain, Former USSR, South Korea, India. Mexico, Australia, The Netherlands, Brazil, BLEU, Switzerland, Sweden, Turkey, Taiwan, Austria, Norway, Denmark. c Poland, Indonesia, Greece, Southafrican Union, Finland, Ireland, Thailand, Hong Kong, Portugal, Argentina, Former Czechoslovakia, Israel. d Malaysia, Venezuela, Singapore, Former Yugoslavia, Hungary, New Zealand, Colombia, Chile, Philippines, Algeria, Pakistan, Romania. e Peru, Egypt, Nigeria, Morocco, Ecuador, Tunisia, Bulgaria, Iceland, Albania, Gabon, Brunei Daressalam.

Table 8. Degree of openness, degree of total connection, and degree of integration, weighted means, country groups according to per capita GDP, 2004 Group
a

DO c=1

DTC Country-specic c 66.76 66.64 72.88 73.35 85.72 c=1 28.86 39.19 47.83 36.00 45.87

DI Country-specic c 29.71 40.84 50.58 42.66 47.28

Group Group Group Group Group

a b

1 2b 3c 4d 5e

14.76 25.69 37.54 26.86 27.91

64.17 61.60 66.29 57.48 81.33

Norway, Switzerland, Ireland, Denmark, Iceland, United States, Sweden, Japan, United Kingdom, Austria, The Netherlands, BLEU. Finland, France, Germany, Australia, Canada, Italy, Hong Kong, Spain, Singapore, New Zealand, Greece Israel. c Portugal, Brunei Daressalam, South Korea, Taiwan, Hungary, Former Czechoslovakia, Mexico, Poland, Chile, Malaysia, Gabon, Turkey. d Former Yugoslavia, Venezuela, Argentina, Southafrican Union, Romania, Bulgaria, Brazil, Tunisia, Former USSR, Thailand, Algeria, Peru. e Albania, Ecuador, Colombia, Morocco, China (Peoples Rep.), Philippines, Indonesia, Egypt, India, Nigeria, Pakistan.

MEASURING GLOBALIZATION OF INTERNATIONAL TRADE: THEORY AND EVIDENCE

143

which are proportionally much more closed than what might be justied by the size of their domestic markets. Due to the higher weight of large economies in the aggregate, the most relevant eect is that of the limits to openness on the globalization indicator. Second, the eect of bias on trade among economies toward certain areas (which limits the direct connection balance) is, in general, more limited than the eect of the degree of openness. However, we have detected that some regional integration processesespecially in Europeemphasize the orientation of many of its recent members exports toward the internal market, to the detriment of developing more balanced trade networks with the rest of the regional world markets. Other forces or barriers are operating in this way to restrict the advance of globalization. Third, the system of indicators suggested shows that the expansion of indirect tradevigorously boosted by the reduction in transport costs and ICT developmentmay well represent a relevant factor in increasing the degree of total connection for many economies and, as a result, their degree of integration. This factor is more relevant for economies that are less directly related to all the others, since they can be integrated in the world trade network through indirect connections. In the case of Europe, some southern economies may be reinforcing their integration through intense commercial relations with other European Union partners that have higher levels of total connection. Finally, the patent heterogeneity of the degrees of openness and connection for dierent sized economies causes the globalization indicator to be aected by the lower degree of integration of some of the largest economies. These results lead us to pose other interesting questions. First, we may inquire into the likely causes for the

dierences between countries in terms of their degrees of integration, openness, and connection. The literature on international economic integration has explored many factors, but always under the assumption that openness and integration are one and the same thing. Once the role of the degree of connection has been introduced, we can reconsider the relevance of economic size, language, colonial or political relations, currency, trade or tari agreements, etc. We can also examine other factors such as economic and technological development, specialization, or human capital endowments, and their importance in relation to the degree of openness or connection, and their eventual impact on integration. In addition, the analysis performed in this study suggests that the international economic integration indicators presented can complement other traditional indicators in the study of the international competitiveness of economies, and the relation between integration and growth. Our study contemplates only some of the potentially relevant features of the trade network for integration, although there could be more. For instance, it might be of interest to analyze in greater depth whether integration and its eects are inuenced by the central or peripheral position of countries with respect to all ows. It might also be important to analyze the existence of regional trade networks within the global network, with much more intense interior relations, and their contribution to globalization. Finally, it must be stressed that the methodology developed in this study can be applied to other international networks of connections such as nancial ows (foreign direct investments, capital ows, or banking investments). If data are available, other dimensions of international economic integration could be analyzed using a similar approach.

NOTES
1. Other indirect approaches take into account the barriers to integration. However, these are not true indicators of international integration but explanatory variables for their limits, that is, for home-country bias, or for other biases such as geographical and ow-orientation biases (Brahmbhatt, 1998; Frankel, 2000; Knetter & Slaughter, 1999). 2. Since imperfect competition is at the core of the new theories on international trade (Krugman & Obstfeld, 2002), to date, international economic integration measures under conditions of imperfect competition are unavailable. Although econometric estimations on the ability to explain deviations from the LOOP are manifold (Knetter & Slaughter, 1999), they do not solve the key problem: the lack of a benchmark to measure integration that does not depend on perfect competition. 3. Justifying the intensity of trade between countries is at the origin of the alternative theories on international trade, which emphasize the importance of the dierent factors of endowment and the dierences in productivity. A panorama in Brakman, Garretsen, Van Marrewijk, and Van Witteloostuijn (2006, chaps. 3 and 4). 4. Many scholars share the opinion that the main drivers of international economic integration in the private sphere are technological change, and the decline in transportation and communication costs. Nonetheless, in the public sphere the main drivers are associated with the gradual removal of political barriers to trade and investment, and capital and human ows (Frankel & Rose, 2000). Certain trends, such as the development of ecommerce networks, or the increasing policy of outsourcing stages in the production process, represent a breakdown in the vertically integrated mode of production (Feenstra, 1998; Feenstra & Hanson, 1996, 1997, 1999). These trends suggest that indirect connections are important and can contribute to the acceleration of the globalization process, thanks to the reduction in transport and transaction costs, and greater reliance on international markets as a mechanism of resource allocation (Coase, 1937; Grossman & Helpman, 2002, 2005; Williamson, 1975). However, the debate continues on the consequences of globalization (Bhagwati, 2004a, 2004b; Rodrik, 1998a, 1998b; Salvatore, 2004; Stiglitz, 2002). 5. See, for instance, the literature on social networks, examples of which include Annen (2003), Hanneman and Riddle (2005), Karlin and Taylor (1975), Wasserman and Faust (1992), Wellman and Berkovitz (1988), or Rauch (2001). 6. Several studies highlight the importance of information owing through cultural, political, or economic ties (Combes, Lafourcade, & Mayer, 2005; Greaney, 2003; Pandey & Whalley, 2004; Rauch, 1999, 2001; Rauch & Casella, 2003; Rauch & Trindade, 2002). 7. Most of this literature values the importance of trade ows establishing a threshold (binary links) (Garlaschelli & Loredo, 2005; Kastelle, Steen, & Liesch, 2006; Kim & Shin, 2002; Serrano, Boguna, & Vespignani, 2006). However, some recent studies apply concepts of weighted network analysis (Barrat, Barthelemy, Pastor-Satorras, & Vespignani, 2004a, 20 04b; Barthelemy, Barrat, Pastor-Satorras, & Vespignani, 2005) and specically consider the intensity of the ows. Directed and indirected analysis is also distinguished, taking into account whether the symmetry of the relationship between nodes in both directions (i.e., exports and imports) justies that the importance of ow does not depend on the direction of ow itself.

144

WORLD DEVELOPMENT 15. From a theoretical point of view the opposite does not hold. It is possible to have an economy, say i, which is over-open (DOi > 1) so that even if its degree of total connection is lower than 1 (DTCC < 1) its degree of integration will be equal to 1. However, data show that this case is very exceptional since only Malaysia has a degree of openness higher than 1. 16. Information on CHELEM (Comptes Harmonises sur les Echanges et lEconomie Mondiale, or Harmonised Accounts on Trade and The World Economy) database is available at URL http://www.cepii.fr/anglaisgraph/ bdd/chelem.htm. 17. The computations for indicators based on imports do not alter the general results, although they may dier for some specic countries. These results are not reported due to space limitations, but are available from the authors upon request. 18. Results on all indicators for the remaining sample years are available from the authors upon request. 19. Our results have been obtained by analyzing ows of goods only, not goods and services, since information on the destination of exports is unavailable in the case of services. In addition, the literature deals with the trade of goods and services dierently. See, for instance, Mirza and Nicoletti (2004). 20. Violin plots are a mix between box plots and density functions estimated nonparametrically via kernel smoothing. See Hintze and Nelson (1998). 21. See, for instance, Feenstra (1998), Feenstra and Hanson (1996, 1999), or Grossman and Helpman (2005). 22. In other words, entry l in each row k, pkl, represents the probability that a country in class k according to DO will be in state l according to DTCi. They are computed as pkl = Nkl/Nk, where Nkl is the number of countries in class k and l for DO and DTCi, respectively, and Nk is the total number of countries in class k.

8. See Gaulier, Jean, and Unal Kesenci (2004), Iapadre (2006), and the references there for a discussion on the measures of regional trade intensity and their limitations. 9. In Section 3, we dene a similar criterion to neutrality. However, we take into account that the weights to be considered are not those for each country in terms of trade, but those in terms of its share of world demand and/or production. This criterion of geographic neutrality is more demanding, since it does not consider the degree of openness of each trading country as data entering the analysis. We name it geographic superneutrality. 10. This option can be justied technically as opposed to the alternative of working with imports, or the total amount of both because import networks can be considered symmetric to those of exports. Therefore, the direction of the ows can be ignored in general. However, in specic cases this does not work. For a more in-depth analysis we should not forget Rodriks cautionary (1999) that the best reection of countries barriers to integration is imports. 11. The current statistical measure of DO does not guarantee its value will lay in the (0, 1) range (see Table 1, which shows that Malaysias total exports are larger than its GDP after correcting the domestic bias). A possible solution is to re-scale in order to exclude values beyond unity. However, given that only one country shows that behavior, we have maintained the natural denition of the degree of openness. 12. We use DO instead of DOi when general statements on the degree of openness are being made, or references to the variable itself, which do not hang on any specic country. The same rule will be applied to the other indicators. 13. The dependence of both the number and magnitude of exchanges on economic size is the focus of international trade analyses based on gravity models and widely used in the literature (Feenstra, Markusen, & Rose, 1998, 2001; Hummels & Levinsohn, 1995; Rauch, 1999). P P 14. By denition it is veried that j2N aij j2N bij 1, thus both matrices A and B dene Markov chains and it can be proven that they are recurrent irreducible aperiodic Markov chains.

REFERENCES
Andersen, T. M., & Herbertsson, T. T. (2003). Measuring globalization. IZA Discussion Paper 817, Institute for the Study of Labor (IZA), Bonn, Germany. Annen, K. (2003). Social capital, inclusive networks, and economic performance. Journal of Economic Behavior & Organization, 50, 449463. Barrat, A., Barthelemy, M., Pastor-Satorras, R., & Vespignani, A. (2004a). The architecture of complex weighted networks. PNAS, 101(11), 37473752. Barrat, A., Barthelemy, M., & Vespignani, A. (2004b). Weighted evolving networks: Coupling topology and weight dynamics. Physical Review Letters, 92(22). Barthelemy, M., Barrat, A., Pastor-Satorras, R., & Vespignani, A. (2005). Characterization and modeling of complex weighted networks. Physica A, 346, 3443. Bhagwati, J. (2004a). Anti-globalization: Why?. Journal of Policy Modeling, 26, 439463. Bhagwati, J. (2004b). In defense of globalization. Oxford: Oxford University Press. Brahmbhatt, M. (1998). Measuring global economic integration: A review of the literature and recent evidence. Working paper, The World Bank, Washington. Brakman, S., Garretsen, H., Van Marrewijk, C., & Van Witteloostuijn, A. (2006). Nations and rms in the global economy: An introduction to international economics and business. Cambridge: Cambridge University Press. Coase, R. (1937). The nature of the rm. Economica New Series, 4, 386405. Combes, P.-P., Lafourcade, M., & Mayer, T. (2005). The trade-creating eects of business and social networks: Evidence from France. Journal of International Economics, 66, 129. Dreher, A. (2006). Does globalization aect growth? Evidence from a new index of globalization. Applied Economics, 38(10), 10911110. Edwards, S. (1993). Openness, trade liberalization, and growth in developing countries. Journal of Economic Literature, 31(3), 13581393. Fagiolo, G., Reyes, J., & Schiavo, S. (2007). The evolution of the world trade web. Working Paper Series 17, Laboratory of Economics and Management (LEM), SantAnna School of Advanced Studies, Pisa. Feenstra, R. C. (1998). Integration of trade and disintegration of production in the global economy. Journal of Economic Perspectives, 12(4), 3150. Feenstra, R. C., & Hanson, G. H. (1996). Globalization, outsourcing, and wage inequality. American Economic Review, 86(2), 240245. Feenstra, R. C., & Hanson, G. H. (1997). Productivity measurement and the impact of trade and technology on wages: Estimates for the U.S., 19721990. Working Paper 6052, NBER, Cambridge, Massachusetts. Feenstra, R. C., & Hanson, G. H. (1999). The impact of outsourcing and high-technology capital on wages: Estimates for the United States, 19791990. Quarterly Journal of Economics, 114(3), 907940. Feenstra, R. C., & Hanson, G. H. (2004). Intermediaries in entrepot trade: Hong Kong re-exports of Chinese goods. Journal of Economics & Management Strategy, 13(1), 335.

MEASURING GLOBALIZATION OF INTERNATIONAL TRADE: THEORY AND EVIDENCE Feenstra, R. C., Markusen, J. R., & Rose, A. K. (1998). Understanding the home market eect and the gravity equation: The role of dierentiating goods. Working Paper 6804, NBER, Cambridge, Massachusetts. Feenstra, R. C., Markusen, J. R., & Rose, A. K. (2001). Using the gravity equation to dierentiate among alternative theories of trade. Canadian Journal of Economics, 34(2), 430447. Frankel, J. A. (2000). Globalization of the economy. Working Paper 7858, NBER, Cambridge, Massachusetts. Frankel, J. A., & Rose, A. K. (2000). Estimating the eects of currency unions on trade and output. Working Paper 7857, NBER, Cambridge, Massachusetts. Garlaschelli, D., & Loredo, M. (2005). Structure and evolution of the world trade network. Physica A, 355, 138144. Gaulier, G., Jean, S., & Unal Kesenci, D. (2004). Regionalism and the regionalisation of international trade. Working Paper 16, Centre dEtudes Prospectives et dInformations Internationales (CEPII), Paris. Greaney, T. M. (2003). Reverse importing and asymmetric trade and FDI: A networks explanation. Journal of International Economics, 61, 453465. Grossman, G. M., & Helpman, E. (2002). Integration versus outsourcing in industry equilibrium. Quarterly Journal of Economics, 117(1), 85120. Grossman, G. M., & Helpman, E. (2005). Outsourcing in a global economy. Review of Economic Studies, 72(1), 135159. Hanneman, R., & Riddle, M. (2005). Introduction to social network methods. Technical report, University of California at Riverside, Department of Sociology. Heshmati, A. (2006). Measurement of a multidimensional index of globalization. Global Economy Journal, 6(2), 128. Hintze, J. L., & Nelson, R. D. (1998). Violin plots: A box plot-density trace synergism. The American Statistician, 52(2), 181184. Hummels, D., & Levinsohn, J. (1995). Monopolistic competition and international trade: Reconsidering the evidence. Quarterly Journal of Economics, 110(3), 799836. Iapadre, L. (2006). Regional integration agreements and the geography of world trade: Measurement problems and empirical evidence. In P. De Lombaerde (Ed.), Assessment and measurement of regional integration (pp. 6585). London: Routledge. Kali, R., Mendez, F., & Reyes, J. (2007). Trade structure and economic growth. Journal of International Trade & Economic Development, 16(2), 245269. Kali, R., & Reyes, J. (2007). The architecture of globalization: A network approach to international economic integration. Journal of International Business Studies, 38, 595620. Karlin, S., & Taylor, H. M. (1975). A rst course in stochastic processes. New York: Academic Press. Kastelle, T., Steen, J. T., & Liesch, P. W. (2006). Measuring globalisation: An evolutionary economic approach to tracking the evolution of international trade. In DRUID summer conference. Kearney, A. T. (2003). Whos up, whos down?. Foreign Policy, 134, 6073. Kim, S., & Shin, E.-H. (2002). A longitudinal analysis of globalization and regionalization in international trade: A social network approach. Social Forces, 81, 445471. Knetter, M. M., & Slaughter, M. J. (1999). Measuring marketproduct integration. Working Paper 6969, NBER, Cambridge, Massachusetts. Krugman, P. R., & Obstfeld, M. (2002). International economics: Theory and policy. Addison Wesley. Lockwood, B., & Redoano, M. (2005). The CSGR globalisation index: An introductory guide. Working Paper 155/05, Centre for the Study of Globalisation and Regionalisation (CSGR), University of Warwick, Coventry. Maddison, A. (2001). The world economy: A millenial perspective. Development centre studies. Pars: OECD. Martens, P., & Zywietz, D. (2006). Rethinking globalisation: A modied globalisation index. Journal of International Development, 18, 331350. McLuhan, M., & Fiore, Q. (1968). War and peace in the global village: An inventory of the current spastic situations that could be eliminated by more feedforward. New York: Bantam Books/Random House. Mirza, D., & Nicoletti, G. (2004). What is so special about trade in services? Research Paper Series 2004/02, Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham, Nottingham.

145

OECD. (2005). Measuring globalisation. OECD handbook on economic globalisation indicators. Statistics. Paris: OECD Publishing. ORourke, K. H., & Williamson, J. G. (1999). Globalization and history: The evolution of a 19th century atlantic economy. Cambridge, Massachusetts: MIT Press. ORourke, K. H., & Williamson, J. G. (2002). When did globalization begin?. European Review of Economic History, 6, 2350. Pandey, M., & Whalley, J. (2004). Social networks and trade liberalization. Working Paper 10769, National Bureau of Economic Research, Cambridge, MA. Rauch, J. E. (1999). Networks versus markets in international trade. Journal of International Economics, 48, 735. Rauch, J. E. (2001). Business and social networks in international trade. Journal of Economic Literature, 39, 11771203. Rauch, J. E., & Casella, A. (2003). Overcoming informational barriers to international resource allocation: Prices and ties. The Economic Journal, 113, 2142. Rauch, J. E., & Trindade, V. (2002). Ethnic Chinese networks in international trade. Review of Economics and Statistics, 84(1), 116130. Rodrguez, F., & Rodrik, D. (2001). Trade policy and economic growth: A skeptics guide to the cross-national evidence. In B. Bernanke, & K. S. Rogo (Eds.), Macroeconomics annual 2000. Cambridge, MA: MIT Press for NBER. Rodrik, D. (1998a). Globalisation, social conict and economic growth. World Economy, 21(2), 143158. Rodrik, D. (1998b). Symposium on globalization in perspective: An introduction. Journal of Economic Perspectives, 12(4), 38. Rodrik, D. (1999). The new global economy and developing countries: Making openness work. Number 24 in Policy Essays. Overseas Development Council, Washington, DC. Rodrik, D. (2000). How far will international economic integration go?. Journal of Economic Perspectives, 14(1), 177186. Salvatore, D. (2004). Globalization, growth and poverty: Editors introduction. Journal of Policy Modeling, 26, 421424. Scholte, J. A. (2002). What is globalization? The denitional issueAgain. Working Paper 109, Centre for the Study of Globalisation and Regionalisation (CSGR), University of Warwick, Coventry. Serrano, A., Boguna, M., & Vespignani, A. (2006). Commercial pathways and the global organization of the international trade system. Stiglitz, J. E. (2002). Globalization and its discontents. New York & London: Norton. Wasserman, S., & Faust, K. (1992). Social network analysis: Methods and applications. Number 8 in Structural analysis in the social sciences. Cambridge: Cambridge University Press. Wellman, B., & Berkovitz, S. (1988). Social structure. Cambridge: Cambridge University Press. Williamson, O. E. (1975). Markets and hierarchies: Analysis and antitrust implications. New York: Free Press.

APPENDIX Proposition and proof Here, we demonstrate that the series given in (5) and (6) are convergent. P Proposition 7. Given a matrix C = (cij) such that j2N 6 1 for P1 n1 n all i 2 N it is veried that the series C is n1 C1 C convergent, where C is a diagonal matrix with elements in the diagonal in the interval (0, 1).

Proof. Let k k1 be the matrix norm dened as kCk1 = max{jcij j:i,j 2 N}. Clearly kCk1 6 1, which implies that P1 k(1 C)Ck1 6 (1 mini2N{ci}) < 1 and the series n1 n C is convergent. h n1 C1 C

Available online at www.sciencedirect.com