Do Birds of a Feather Trade Together? Homophily in Preferential Trade Agreements Network

Transcription

1 Do Birds of a Feather Trade Together? Homophily in Preferential Trade Agreements Network Taedong Lee, University of Washington, Seattle Byoung Inn Bai, Seoul National University A paper prepared for presentation at the International Political Economy Society Conference November 13 14, 2009 College Station, Texas 1

2 Introduction The world has witnessed an unprecedented outburst of Preferential Trade Agreements (PTAs) since the mid 1990s. Although economic regionalism has long been an enduring feature of the international political economy, recent development outweighs the previous waves in its pervasiveness (Mansfield and Milner 1999). The World Trade Organization (WTO) reports that almost 300 additional PTAs have been notified to the WTO since its inception in 1995 and that as of December 2008, total number of notification reached 421 of which 230 PTAs were in force. 1 Virtually every member country of the WTO is now a party to at least one. Against this backdrop, scholarly efforts to explain this new development have been flourishing. Equipped with the Vinerian terms of trade effect (Viner 1950), economists have long debated over the welfare implications of PTAs for both members and the world as a whole (Bhagwati 1993; Bhagwati and Panagariya 1996; Krugman 1993; Lawrence 1996; Summers 1991). Largely couched in terms of the magnitude of trade diversion effect, economic explanations attribute the rise of economic regionalism to the expected welfare gains from PTAs and to the concomitant domino effects (Baldwin 1997; Whalley 1998; Winters 1996; Yi 1996). Political scientists, on the other hand, have called upon both international and domestic political factors. Internationally, the changes in global power distribution (Mansfield 1998) and the multilateral trade regime (Mansfield and Reinhardt 2003) have been singled out while the quest for liberal economic and political reforms has been considered as the major driving forces behind the new wave of regionalism (Birch 1996; Haggard 1997; Mansfield and Milner 1999; Remmer 1998; Whalley 1998). Yet, notwithstanding its significance and contribution, extant literature has paid little attention to an important aspect of PTA formation: the choice of partners. Despite the general trend toward economic regionalism, countries are not indifferent to their 1 2

3 partners. That is, the new wave waves unevenly. While extant literature is full of theoretical conjectures about the likely PTA partners, only a handful of empirical studies have been conducted. Economists have addressed this issue through various modifications of the standard gravity model that locate trade flow among PTA members on the left hand side of the equation (Haveman and Hummels 1998; Soloaga and Winters 2001). This strategy, however, is not immune from the endogeneity problem (Magee 2003; Baier and Bergstrand 2007). Political scientists, on the other and, have focused on the specific variable of interest such as regime type (Mansfield, Milner and Rosendorff 2002) instead of taking directly into account all the relevant variables. We attempt to fill this gap by utilizing the concept of homophily and heterophily in social network analysis. Both homophily and heterophily attribute the tendency to form social ties to the similarity or dissimilarity in actors attributes. Extant literature suggests that the similarities or dissimilarities in economic and political attributes are crucial in choosing PTA partners. On the basis of the Vinerian distinction between trade diversion and trade creation effects, economists have long proposed that PTA partnership is contingent upon such attributes as geographic proximity, the size and sectoral composition of national economy, and the level of economic development. Political scientists, on the other hand, have pointed out that the similarities in political institutions matter. We test all of these hypotheses in light of the homophily and heterophily in PTA network and find that regime type matters significantly while economic variables except geographic proximity produce mixed results. 2 This finding confirms the previous work that finds strong association between democracy and PTA partnership. The remainder proceeds as follows. Next section reveals the homophily assumptions in the extant literature on economic regionalism and elaborates hypotheses. Section three provides research design and descriptive analysis of the existing PTA 2 In our empirical analysis, we only consider Free Trade Agreements (FTAs) that account for over 90% of PTAs. 3

4 network. Section four analyzes the determinants of PTA network by fitting the Exponential Family Random Graph (ERG) model and discusses the findings. Section five concludes. Choosing Partners: Homophily and Heterophily in PTA Network PTA has a two sided quality. It represents a move toward freer trade among partners, but, at the same time, is discriminatory against non members as preferential treatments are only extended to partners (de Melo and Panagariya 1993, 4). This distinctive nature of economic regionalism constitutes the Vinerian terms of trade effect. Viner (1950) demonstrates that the creation of new trade between partners unambiguously increases the welfare of each partner as they are able to undermine each other s less efficient industry. Yet, this putatively positive effect of economic regionalism may be offset as members are enabled to displace a more efficient outside supplier by taking advantage of tariff preference they enjoy in a partner country. On the basis of this discussion, extant literature provides a number of predictions about which countries are likely to form PTAs. These predictions, in general, fall into two groups. One line of reasoning predicts that only welfare enhancing arrangements are likely to be signed, while the other emphasizes political factors that buttress those arrangements. The baseline assumption of these predictions is that any PTA requires a simultaneous assent of partners. With this restriction, the former stresses that PTAs are likely to be formed if both countries are able to minimize the negative terms of trade effect of trade diversion. Such attributes as geographic proximity (Summers 1991; Krugman 1991; Frankel 1997), the symmetry of economic size (Michaely 1998), relative factor endowments (Ethier 1998; Krueger 1999) have been singled out as the key factors that determine overall welfare impact of a PTA. The latter line of reasoning explores the conditions under which PTAs are 4

5 politically feasible. Grossman and Helpman (1995) and Levy (1997) argue that PTAs are likely to be formed when they generate substantial welfare gains for the median voters or supportive interests groups. Since PTAs require the assent of all governments, it is vital for the influential social groups to be indifferent or supportive to the proposed arrangement. They argue that these conditions are likely to be met when there is relative balance in the potential trade between the partner countries (Grossman and Helpman 1995) and the countries involved have capital labor ratios that are on the same side of the world median capital labor ratio (Levy 1997). While these discussions deduct the political feasibility of a PTA from the economic attributes of the partner countries, other scholars focus on the impact of political institutions, especially that of political regime. A group of scholars argue that democracies are the likely candidates of PTA partnership because of their sensitivity to gross social welfare (Mitra, Thomakos and Ulubasoglu 2002) or electoral constraints (Mansfield et al. 2002). While there are considerable disagreements, existing literature on economic regionalism shares a common view that the likelihood of PTA partnership depends upon the relative attributes of participating countries. The point of disagreement in the extant literature relates to whether countries are likely to form a PTA on the basis of their similarity or dissimilarity in these attributes. This common view is in essence tantamount to the homophily (heterophily) principle in social network analysis, which assumes that the similarity (dissimilarity) of actors attributes are the determining factor of social ties (McPherson, Smith Lovin and Cook 2001). The attributes that determine the likelihood of PTA partnership fall into three categories: geographic proximity, economic attributes (size, development level and sectoral composition) and political institution. We now elaborate testable hypotheses by discussing each of these attributes in more detail. Geographic Proximity One of the key features alleged to determine the likelihood of PTA formation is 5

6 geographic proximity. In fact, the very notion of economic regionalism hinges on the importance of geographic proximity despite its inherent elusiveness (Katzenstein 1997, 8 11; Viner 1950, 123). Scholarly debate on the role of geographic proximity relates to the putatively negative effect of trade diversion. While Viner (1950) and the following works stipulate that trade diversion produces welfare losses, a group of scholars have expressed deep suspicion on this claim (Baldwin and Venables 1995). Summers (1991), for example, argues that trade diversion is not something to be worried when considering PTAs. According to him, the risk of trade diversion is minimal if the members of a PTA are natural trading partners, that is, they are geographically proximate and major trading partners with each other. Krugman (1991) draws the same conclusion by noting that, despite the potential for trade diversion, a PTA between geographically proximate countries can be welfare enhancing if considering transport costs. Later works by Frankel (1997) and Frankel, Stein and Wei (1995, 1996, 1998) distinguish natural, unnatural and supernatural PTAs and conclude that PTAs are likely to be formed between geographically proximate countries. With high intercontinental transport costs, according to these works, PTAs between geographically close countries are always welfare enhancing because it creates more trade among member countries while trade diversion is minimized due to intercontinental transport costs. Thus, to the extent that governments are concerned about the general welfare of citizens, they are willing to form PTAs with neighboring countries in order to reap the benefits from preferential trade arrangements. This logic is not immune to criticism, however. Nitsch (1996) challenges this view by noting that it holds only if intra continental transport costs are zero. Since it is unlikely, he concludes, the net benefits of a PTA among geographically closer countries depend upon the magnitude of intra continental transport costs relative to inter continental costs. In a more analytic sense, Bhagwati (1995) and Bhagwati and Panagariya (1996) 6

7 point out that the idea of natural trading partners is neither symmetric nor transitive. According to them, both the United States and Mexico, for example, may be considered as natural trading partners in this logic, but in reality the relationship between them is not symmetric. While the United States can be considered as the natural trading partner of Mexico, the reverse is not true. Thus, it is hard to hold that a PTA between geographically proximate countries is ceteris paribus superior to that between distant countries (Bhagwati and Panagariya 1996). The debate between these two logics is in essence an empirical one. If a PTA between geographically closer countries can help countries reap the benefits of trade creation while appeasing the negative impact of trade diversion, we should expect that the existing network of PTAs reflects geographic distribution. If, on the other hand, the critiques are right in expecting zero or minimal effect of transport costs, geographic proximity would not be an important factor in considering PTA formation. H1. Countries in the similar region (geographical proximity) are more likely to form PTA ties. Economic Attributes: Size, Development Level and Sectoral Composition At the heart of theoretical conjectures about the likelihood of PTA partnership lie the relative economic attributes of potential partners. Both explanations that focus either on the welfare implications or on the political sustainability of PTAs assume the fundamental decisiveness economic attributes in determining PTA formation. Among others, the size, development level and sectoral composition of national economy are the mostly mentioned attributes. Yet, scholars are sharply divided over whether the symmetry or asymmetry in these attributes contributes the likelihood of PTA formation. A group of scholars argue that PTAs are likely to be formed among those countries with similar economic size (Furusawa and Konishi 2007; Kennan and Riezman 7

8 1990; Krishna 1998; Michaely 1998). The logic behind this conjecture is quite straightforward. Given that a PTA allows the improved access to another market, participating countries may expect the positive change in terms of trade stemming from the increased export to the partner countries (Bagwell and Staiger 1999). This benefit will be larger as the partner countries have bigger domestic markets. If the market sizes of participating countries are asymmetric, however, forming a PTA would be a difficult job since any improvement in terms of trade for one country would come only at the expense of the other. The larger country may expect negative change in terms of trade in making a PTA with smaller countries. Thus, to the extent that a PTA requires simultaneous assent of both parties, it is highly likely that a PTA will be formed between those countries with similar economic sizes. Others argue that both the welfare benefits and the political feasibility of a PTA are contingent upon the relative symmetry in the development level and sectoral composition of national economies. Venables (2003) points out countries with comparative advantages closer to the world average will do better in a PTA than do countries with more extreme comparative advantages. Since the negative effect of trade diversion will vary along with the members factor endowments relative to the rest of the world, a PTA between the countries with similar capital labor ratios is likely to produce more benefits than losses. Conversely, a PTA between poor countries with comparative advantages below the world average may distort their true comparative advantages so that trade diversion will undermine trade creation. Levy (1997) also demonstrates that the political feasibility of a PTA increases as the members are on the same side of the world median capital labor ratio and the prospect for intra industry trade is considerable. Since intra industry trade allows countries to reap gains from product variety without having to suffer from industry dislocation, a PTA between those countries with similar sectoral composition is politically more sustainable than the one that inter industry trade is prevailing (see also Krugman and Obstfeld 2000). 8

9 Taken together, these logics suggest that PTAs are likely to be formed between homogeneous economies in terms of the size, development level and sectoral composition of national economy. Yet, others raise a serious question to this proposition. First, in light of the standard Ricardian trade model, Ethier (1998) and Krueger (1999) argue that PTAs between heterogeneous economies are likely to be more welfareenhancing than those between homogeneous economies. Since countries with similar economic attributes have less scope for gains from trade based on comparative advantage, they conclude that North South PTAs are more likely to be prevailing than North North or South South PTAs. Secondly, those who emphasize strategic considerations behind PTA formation suggest that PTAs are likely to be formed between heterogeneous economies despite asymmetric welfare gains to the members. In economic models discussed above, PTAs between small developing and large advanced economies produce asymmetric welfare gains to the small developing economies. The expected gains for these economies range from the easier access to larger market, technology and skill transfer, to the increased inward investments (Schiff and Winters 2003). While large advanced economies are not able to expect similar economic gains, they nevertheless may be willing to form such a PTA out of strategic considerations that take advantage of it as a means to enhance their voice or bypass the deadlock in the multilateral negotiations (Mansfield and Reinhardt 2003). Also, sometimes PTAs are formed out of purely foreign policy considerations as in the US Israel and US Jordan PTAs (Rosen 2004). In sum, extant literature exhibits considerable disagreement over whether similarity or dissimilarity in economic attributes determines the choice of PTA partners. As in the case of geographic proximity, this issue is ultimately an empirical one. H2. PTA formation is contingent on the similarity in economic size, development level and sectoral composition. 9

10 Political Institutions In addition to economic attributes, extant literature suggests that countries with similar political institutions are likely to be PTA partners. The intuition behind this reasoning is that all the expected benefits from a PTA rely on the members willingness and ability to liberalize trade simultaneously. The positive effect on terms of trade, for example, may turn into an adverse one if only one party liberalizes. Yet, since trade liberalization and the concomitant policy coordination are costly, countries ability to form a PTA will be constrained by their political institutions that may influence on the relative easiness of striking and maintaining a deal. For this reason, it has been noted that institutional similarity matters in promoting economic regionalism (Nogués and Quintanilla 1993; Katzenstein 1997). Out of various aspects of political institutions, regime type has been singled out as the major determinant. A group of scholars suggest that trade liberalizing PTAs are likely to be formed between democracies due to their trade policy profiles. Wintrobe (1998), for example, claims that democracies are less rent seeking than non democracies. Since protectionism is merely a form of rent seeking, he concludes that democracies are likely to pursue free trade. Thus, to the extent that PTAs involve in mutual trade liberalization, democratic countries are likely to be partners to each other. Mitra et al. (2002), on the other hand, suggest that democratic countries tend to place greater weight on social welfare than do non democracies, which imply that democracies are willing to pursue PTAs to the extent that they are welfare enhancing. Mansfield et al. (2002) provide more systemic analysis as to why democracies are more likely to be PTA partners with each other. Assuming that both democratic and nondemocratic leaders would like to maximize the rents stemming from protectionism, their model suggests that both leaders face with the problem of assuring voters that the government is not being too extractive. This problem is more severe in democracies due to the prospect of losing elections. This electoral constraint provides the incumbent 10

11 executive with strong incentives to sign up PTAs since they allow executives to commit themselves credibly to actions that voters would otherwise find incredible. Thus, democracies are likely to be more prone to PTAs especially with other democracies. While these logics commonly suggest that democracies tend to pursue trade liberalization against each other, Verdier (1998) argues exactly the opposite. According to him, political conflicts engendered by free trade are likely to lead democratic countries to adopt more protectionist policy against each other except when intraindustry trade dominates their trade flow. As in Grossman and Helpman (1995) and Levy (1997), he argues that only those democratic countries with similar factor endowments are likely to manage free trade between them since it will guarantee more political feasibility. With respect to the impact of regime type on choosing PTA partners, this logic suggests that regime types play only a marginal or negligible role. Again, extant literature is divided over whether regime types matter in determining the likelihood of PTA partners. H3. Democracies are likely to be PTA partners with each other. With Whom Do Countries make PTAs? Data and Network Description Extant literature, in sum, hypothesizes that countries are likely to form PTAs on the basis of their relative attributes. We now turn to the empirical examination of these hypotheses by describing the network structure of PTA ties in the world. In our analysis, we only consider the formal bilateral Free Trade Agreements (FTAs) notified to the GATT/WTO for the following two reasons. First, while GATT/WTO reports other types of PTAs such as customs union, more than 90% of the contemporary PTAs are FTAs. Due to its pervasiveness, we only focus on FTAs. Second, we treat existing PTAs such as EU and NAFTA as single entities since a number of PTAs are formed between them and individual countries. Thus, while we do not consider the PTA ties within EU and NAFTA, 11

12 we do include the PTAs between these regional groups and individual countries such as FTAs between the EU (27 members) and Algeria, between EFTA (Iceland, Liechtenstein, Norway, and Switzerland) and Korea and so on. Since this choice rules out some customs unions from the beginning, we exclusively focus on FTAs for the sake of coherence. Our data reveal that the FTA network has 71 vertices and 157 edges, that is, 71 political and economic entities around the world have 157 bilateral FTAs with each other. In the network, we find two distinctive components: one consists of densely connected 68 entities and the other has three African countries (Kenya, Tanzania, and Uganda). That is, three African countries form FTAs among themselves without any FTA ties with outside countries. The density of a network is one of the network properties showing the degree of the pairwise ties between actors. Higher density in a network refers that there are more ties between actors. The density of FTA network is Given the maximum possible number of FTA ties with 71 countries are 2485, the density of the FTA network is calculated as 157/2458= In order to verify the hypotheses from the existing literature, we categorize 71 entities with five different vertex attributes: Region, GDP, OECD, Industrial Sector, and Regime Type. Existing literature contends whether countries with geographic proximity and similar political and economic attributes are the likely candidates for PTA partners. We use each of these vertex attributes as a proxy for contending hypotheses. As discussed in the previous section, we collapse down the economic attributes into three categories: economic size ( GDP ), development level ( OECD ) and sectoral composition ( Industrial Sector ) while Region and Regime Type directly address the hypotheses on geographic proximity and regime type. We now describe existing FTA network on the basis of these vertex attributes. 3 Density=L/g(g-1)/2=2L/g(g-1) (Wasserman and Faust 1995), where L is the number of lines present and g is the number of nodes. Thus FTA network density is 2*(157)/71(71-1)=

13 Homophily in Geography The vertex attribute Region is to test the hypothesis on homophily in geographic proximity. If geographic proximity matters in forming FTAs as many have discussed, we should see the clustering of FTAs among those countries within the same region. According to the generally accepted geographic classification scheme, world consists of eleven regions: North America, Middle America, South America, West Europe, East Europe, East Asia, South Asia, Southeast Asia, Oceania, Africa, and Southwest Asia North Africa (Kim and Shin 2002). Since the sample sizes are small in some regions like North America, and South Asia, we collapsed eleven world regions into six (see Appendix 1). Figure 1 illustrates the strong tendency for FTA ties to be homophilous by regions. The sub network of FTA in African region is isolated with the main FTA network. Only three nodes, Uganda, Tanzania and Kenya, form a small and separated network. While South Africa is an African country, it does link to the EC instead of the African sub network. Asian countries are mainly connected to Asian countries, while there are a few ties between Asian countries and European ones like the EFTA Korea and the EFTA Singapore FTA. Japan, for instance, has three FTAs with other Asian countries (Thailand, Singapore, and Malaysia.) out of four FTAs that Japan currently has. There is no FTA between Asian countries and countries in Africa, South West Asia/North Africa, and Eastern Europe transitional countries. Among other regions, Eastern Europe transitional region demonstrates strong homophily by region. Forty out of sixty FTA of this region are formed within the region. However, figure 1 illustrates that there are two FTA clusters which consist of central Asian countries and Eastern European countries respectively. Central Asian countries such as Kyrgyz Republic, Kazakhstan, and Uzbekistan make FTA cluster one another. This Central Asian cluster is close to Russia, while Eastern European countries such as Albania, 13

14 Croatia, and Moldova have close ties with European Countries. Moldova plays a role in connecting FTA ties between Central Asia and Eastern Europe clusters. Geographic proximity is significant in European region as well. Given that Western Europe is integrated as the EU, FTA ties in European region covers non EU members like Norway, Switzerland, and Iceland. Most of FTA ties of European region are connected to countries in South West Asia/North Africa and Transition regions (twenty nine out of forty seven FTA ties of the Europe), not to Africa, Asia or America (one, two and four FTA ties respectively). The probability of homophily by region is (=77/2485) and the ratio out of total ties is (=77/157). Considering most regions have tendency to form FTA with neighboring regions, economic geography is a noteworthy determinant of FTA formation. Figure 1. FTA Network Homophily and Mixing Matrix by Regions Note: Black: Africa, Red: Asia, Green: America, Dark Blue: Europe, Sky Blue: SW Asia/N Africa, and Violet: Transition 14

15 Africa Asia America Europe SW Asia/ N. Africa Transition Africa Asia America Europe 3 (2%) (12%) 7 (4%) 21 (13%) 1 (1%) 2 (1%) 4 (3%) 11 (7%) SW Asia/ N. Africa Transition (1%) 8 (5%) 17 (11%) 4 (3%) Total (8%) 8 (5%) 40 (26%) Total Note: Percentage of total FTA ties (157) in parentheses. Since the matrix is symmetric, we omit the symmetric part. Homophily in Economic Attributes As discussed before, existing literature on economic regionalism exhibits a considerable disagreement as to the impact of economic attributes on PTA partnership. Some argue that countries with different economic attributes are likely to form PTAs in order to fully exploit the benefits from varying comparative advantages while others insist that the feasibility of PTA heavily relies on the similarity in their economic attributes. Out of many attributes, three stand out: economic size, development level and sectoral composition. We examine each of these elements. In order to test the impact of economic size on FTA formation, we use GDP as a proxy for economic size. 4 While GDP is a continuous variable, we categorize countries with four different groups by using quartile. This operationalization should be considered as relative rather than absolute. As for the development level, we utilize OECD membership with the assumption that OECD members are likely to enjoy similar development level 4 We also tested the amount of trade (export+import) as another measure for economic size. Since the amount of trade variable is highly correlated with the GDP (correlation coefficient.97), we only use the natural log of GDP. 15

16 compared to non members. While this dichotomous variable does not suffice to cover all the aspects of development level, it can be justified on the basis of the common belief that OECD is in general a forum for the developed countries. There are thirty OECD members in total of which eleven members 5 are in the FTA network. We coded EU and EFTA as OECD member since the most of EU members and EFTA members are OECD members. As for the hypotheses on sectoral composition, we categorize each economic entity into three groups: agriculture, manufacture and service sector based countries. We first calculate the relative weight of each industrial sector in GDP and compare it to the world average (3% of agriculture, 28% of manufacturing, and 69% of service sector). Final categorization was conducted on the basis of whether a country exhibits significantly higher portion of a certain sector than world average. In Albania, for instance, agriculture covers 23 % of GDP while manufacturing and service sector mark 22 % and 56%, respectively. Thus, we categorize Albania as agriculture based country. In this categorization scheme, there are thirty five agricultural countries, nineteen industrial countries and seventeen service based countries. Figure 2 illustrates FTA network by GDP. Size of nodes reflects that of GDP. Mixing matrix by GDP illustrates that there are two noteworthy trends in the relationship between economic size and FTA formation. First, countries tend to form FTA ties with those countries that have different economic sizes. There are thirty one homophily FTAs in terms of economic size (ten FTAs between smallest, three between second largest, four between third largest, and fourteen between the largest economic scales). Homophily in economic size accounts for twenty percent of total FTAs. Second, strong gravity appears in the network. 102 out of 157 FTAs are linked to largest economies. Countries in each category have more FTAs with countries in 4 th quartile of GDP than any other categories. 5 The OECD members in the FTA networks are Australia, Canada, EU, EFTA, Japan, Mexico, Norway, South Korea, Switzerland, Turkey and the United States. 16

17 The strong gravity toward big economies casts doubts on the argument that a PTA is likely to be formed between countries with similar economic sizes. While the largest share of FTAs is between countries in the 3 rd and in the 4 th quartile of GDP, FTAs with the biggest economies are prevailing in each quartile. That is, despite the expectation that a PTA between large and small countries is less likely, the existing FTA network reveals that majority of FTAs is formed between countries with asymmetric economic sizes. Figure 2. FTA Network Homophily and Mixing Matrix by GDP SriLanka Laos Indonesia Malaysia Pakistan Buthan India ThailandNewZealand ASEAN Japan China puanewguinea Australia Panama Singapore Korea Uganda Kenya ElSalvador CostaRica Bahrain Honduras Mexic o USA Peru Canada Guatemala Oman Nicaragua Chile Jordan Morocco EFTA Syria Lebanon Switzerland Egypt Norway PalestinianAuthority FaroeIslands Tanzania DominicanRep Israel EC IcelandAndorra Turkey Macedonia Bulgaria Romania Croatia Albania Bosnia Tunisia Montenegro SouthAfrica Liechtenstein Algeria Serbia Kosovo Moldova KyrgyzRepublic Armenia Uzbekistan Turkmenistan KazakhstanGeorgia Azerbaijan Russia Ukraine 1 st quartile 2 nd quartile 3 rd quartile 4 th quartile Smallest (1 st) 2 nd 3 rd (Biggest) 4 th Total (6%) (8%) (9%) (13%) (2%) (8%) (19%) (3%) (24%) (9%) Total

18 Table 1 shows that OECD members are likely to make FTA ties with non OECD members. In particular, the EC makes twenty two FTA ties (out of twenty six FTAs that the EU made) with Non OECD member. The U.S. presents the similar pattern. Australia, Canada, Mexico, and Singapore are OECD members while other twelve partners are non OECD members. Both economic giants, EC and the U.S, present tendency to form FTA ties with Non OECD countries except for FTAs with neighboring countries (Switzerland, Iceland, and Norway for the EU, and Canada and, Mexico for the U.S.). Compared to seventy three FTA ties between Non OECD countries and OECD countries, there are relatively few, sixteen, FTA ties between OECD countries. Probability of FTA ties between OECD countries is (= 16/2485). FTA ties between OECD countries takes approximately ten percent of total FTA ties (16/157=0.102). Interestingly, there are substantial FTA ties between non OECD. The ratio of ties between non OECD is 68/157=0.435 and probability of the ties is 68/2485= The percentage of homophily by OECD/non OECD countries is 54% (=84 (68 between Non OECD + 16 between OECD) /157). Table 1. FTA Network Homophily and Mixing Matrix by OECD Non OECD Total OECD Non OECD (44%) (46%) OECD (10%) Total The mixing matrix in Table 2 presents a weak evidence of homophily by industrial composition. Among others, manufacture based economies are less likely to form FTAs with other manufacture based economies. Only six FTAs are formed between manufacture based countries. The probability of homophily by industrial composition is 18

19 0.025 (=64/2485) and the ratio in total FTAs is (=64/157). Instead, we see more FTAs between countries with different industrial compositions. Agriculture based economies make thirty three and thirty two FTAs with manufacture based economies and service based economies respectively. Service based economies also form more FTA ties with economies with different industrial composition. Table 2 FTA Network Homophily and Mixing Matrix by regime types Agriculture Manufacture Service Total Agriculture 32 (20%) 33 (21%) 32 (20%) 97 Manufacture 6 (4%) 28 (18%) 34 Service 26 (17%) 26 Total Political Institutional Homogeneity In order to test hypothesis on the homophily of regime type, we utilize the Freedom House democracy index as of This index categorizes countries with three types: free, partly free and not free. These three categories indicate the overall status of regime types, combining political rights and civil liberties rating (Freedom house 2008). In this FTA network, there are twenty nine free countries, twenty two partly free countries, and nineteen not free countries. Table 3 illustrates FTA Network Homophily by regime types. FTAs between free countries demonstrate strong homophily. The probability of FTA ties between free countries is (= 54/2485) and the ratio of the total ties is (=54/157). Countries in partly free regime types tend to make FTA ties with free countries instead of partly free countries or not free countries. Countries in not free regime shows similar pattern. Forty out of forty three FTA ties from not free countries link to free countries (twenty one FTAs) and partly free countries (nineteen FTAs). Three FTA between not free 19

20 countries are made between China and Pakistan, Kyrgyz Republic and Kazakhstan, and Kyrgyz Republic and Uzbekistan. The probability of ties between Not Free countries is (=3/2485) and the ratio of ties between Not free in the FTA network is (=3/157). The probability of homophily by regime type is (= ( )/2485) and the ratio of homophily by regime is (=72/157). Homophily by regime types is mainly determined by FTA ties between free countries rather than partly free or not free countries. Both non free countries and partly free countries tend to choose democracies as FTA partners rather than countries with similar regime type to themselves. Table 3. FTA Network Homophily and Mixing Matrix by regime types Free Free 54 (35%) Partly Free Not Free Partly Free 44 (28%) 16 (10%) Not Free 21 (13%) 19 (12%) 3 (2%) Total Total The descriptive analysis presented above reveals that homophily in region is the strongest attribute in FTA network. That is, countries that are geographically close tend to form FTAs with each other. Other attributes, however, exhibit mixed results. With respect to the attribute of economic size, heterophily instead of homophily turns out to be prevailing in the FTA formation. Most FTAs, 102 out of 157, are formed between the largest economies and all others. Homophily in development level and sectoral composition is ambiguous at best. The majority of FTAs is formed between non OECD countries (44%) or between OECD and non OECD countries (46%) while there is no distinctive trend as far as sectoral composition is concerned. Taken together, our descriptive analysis reveals that economic attributes produce 20

21 mixed results at best with respect to the PTA formation. Political regime, on the other hand, turns out to be strong factor in determining FTA partnership. As the literature suggest (Mansfield et al. 2002), democratic countries tend to be active in forming FTA ties with other countries and especially with other democracies. FTA ties that involve in democratic countries account for 75% of total FTAs (119 out of 157) and 35% of them are between democratic countries. In sum, the gravity model of democracy works in the formation of FTA ties. We now turn to more systematic analysis of this result by fitting the ERG models. Fitting ERG models Mixing matrixes in the previous section are able to present a tendency showing the number of FTA ties between different attributes. We now move on to more systemic analysis using the exponential family random graph model. Exponential family random graph model (hereafter ERG model) is to examine the probability of observing a set of network edges. This model aims to predict the joint probability that a set of edges (ties) exists on nodes in a network (Handcock et al 2008; Hunter et al 2008). Contrary to conventional regression models which have focused on attributes (gender, democracy, trade amount, etc) of the nodes (individuals, organizations, states, etc), ERG models in social network analysis address a form of relational data. Relations in the ERG models can be presented in dichotomous variable which indicates propensity of ties or no ties between nodes (Keohely et al 2004). In the FTA context, ERG model is an appropriate approach to examine the questions on how similarities and dissimilarities in nodal attributes as network structures influence the absences and presenses of FTA ties. For instance, ERG models can predict the probability having a tie between the countries sharing similar socio economic attributes such as regime type. 21

22 The ERG model is fitted by the following formula: Logit (Yij=1)=θTδ[g(y, X)]ij (1) Where Yij is an actor pair in Y (the random set of relations), and δ[g(y, X)] is the change in a vector of network statistics, g(y, X), when the value of yij is toggled from 0 to 1. The θ coefficients which determine the impacts of covariate are parameters to be estimated (Goodreau, Kitts, and Morris 2009). We have five models which add one explanatory variable one by one. In this way, we are able to compare the model fits among different models. Model 1 includes only the edge term. This model is for the baseline to compare other models. The coefficient estimate in Model 1 tell us the log odd of tie is.271 corresponding probability of this log odd is (=exp( 2.71)/(1+exp( 2.71)). This is equivalent to the overall network density. In Model 2, we include region term to test hypothesis on geographic homophily. All two terms are statistically significant in 95% level. The probability that correspond to the log odd of a tie that is homogeneous by region is 1.35 (= ); corresponding probability of this log odd is (=exp( 1.35)/(1+exp( 1.35)). When we include region term in the Model 2, we see a dramatic increase in model likelihood relative to model 1. Maximum likelihood estimate of model 1, , decrease to that of in model 2. AIC (Alkaike Information Criteria), an index of model fit, also decrease from 1142 to 997, which means Model 2 has better model fit than Model 1, considering inclusion of additional predictor. 22

23 Table 4. Homophily Models on FTA ties Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Edges 2.71 (.08)** 3.50 (.13)** 2.85 (.16)** Region (.17) ** (.18)** GDP.32 (.22) 3.03 (.17)** 2.36 (.19)**.34 (.22) 3.19 (.18)** 2.40 (.18)**.36 (.22) 3.10 (.18)** 2.37 (.19)**.37 (.22) OECD 1.08 (.19)** Political regime Free countries Partly Free Not Free 1.10 (.19)**.55 (.18)**.86 (.19)**.92 (.21)**.51 (.30) 1.28 (.60)* 1.17 (.19)**.52 (.17)** Sector Composition.34 (.19) Maximum Likelihood AIC Model 3 includes edge, region, GDP and OECD term. The coefficients of this model indicate the log odd of different types of ties. Negative sign in OECD term tell us that non OECD countries are more likely than OECD countries to make FTA ties in this network. The log odd of a tie that is completely heterogeneous (the two members differ from each other in region, GDP, and OECD) is 2.85 which is equivalent to the probability of One that is homogeneous in all three attributes is 1.88 (= (.32)+( 1.08), the probability of (=exp( 1.88)/(1+exp( 1.88))). That is, if two countries are in all the same nodal attributes, for example, both are OECD countries, largest economic scale, and in Asian region, the probability of having a FTA tie is around 13%. Among other variables, regional homophily increase the propensity of making a ties, since the log odds of a tie that is homogeneous by region is 0.48 (= ), which is equal to the probability of (= exp( 0.48)/(1+exp( 0.48)). 23

24 Homophily in political regime in model 5 is positively associated with FTA tie formation. However, this homophily term reflects assortative mixing uniform homophily within the attribute there is the same tendency for within political regime edges (for example), regardless of which political regime is considering about. Thus, Model 5 disintegrates the subcategories of political regime to figure out which categories of political regime homophily promote FTA tie formation. This selective mixing in dyad ties within attributes categories are differentially likely to cross category ties (Goodreau et al 2009) in model 5 shows that only free countries are positive and statistically significant in FTA formation. The inclusion of homophily term of industrial composition in model 5 does not improve the model fit much. In addition, homophily in industrial composition is not statistically significant. While countries of similarities in region, OECD, regime type are likely to form FTA ties, FTA ties is not contingent on similarities in industrial composition. As the theory of comparative advantage predicts, countries are likely to choose FTA partners that have different industrial composition. Conclusion We have analyzed the determinants of FTA formation. While the outburst of FTAs proliferates scholarly attention to it, scant attention has been paid to the question of which pair of countries is likely to form an FTA and for what purpose. We have attempted fill this gap by utilizing the concept of homophily in social network analysis. Since FTAs are in essence a network among a group of countries, it is appropriate to apply tools developed in the field of social network analysis. The concept of homophily allows us to identify the factors that contribute to the FTA formation. Existing literature suggests that countries with shared attributes are likely to form an FTA. As is summarized in section two, those attributes are geographical proximity, similar 24

25 economic size and development level, and institutional homogeneity. We test empirically the relative strength of all these hypotheses. Both our descriptive analysis and the estimates of ERG model confirm that geographical proximity do matter in making FTA ties. As theory suggests, geographical proximity may allow countries to save on transport costs and alleviate negative effect of trade diversion. While our analysis cannot confirm whether transport costs are the major concerns behind the FTA formation, our result gives credit to the idea that geographical proximity matters. Other attributes, however, produce mixed results. Unlike the expectation that countries with similar economic institutions are likely to be strong candidates for an FTA formation, existing FTA network is not on the basis of homogenous economic institutions. Yet, the homogeneity of political regimes turns out to be a factor that influences the likelihood of FTA ties. Given that the majority of FTA ties are between at least partly democratic countries, it seems fair to say that democratic countries do tend to form FTAs. As theory suggests, this may be the case because institutional constraints of democracies make FTAs attractive in resolving commitment problem. The expectation that the similarity in economic size and development level matters turns out to be weak. As our analysis shows, this factor contributes to the FTA formation, but its impact is very weak and mixed. Majority of the existing FTA ties is between developed and developing countries followed by the FTAs between developing countries. Our analysis indicates that developed countries tend not to form an FTA between them while economic theory suggests that FTAs are likely to be between developed countries. This finding may reflect the fact that the motivation behind an FTA goes far beyond economic analysis of trade creation and diversion. While our analysis cannot confirm the major motivation behind an FTA formation, it nevertheless shows that FTA formation is more than economic calculation. We expect further research is necessary in this issue. Future research might also examine the evolution of FTA network. We analyze the static FTA network as of However, the FTA network has been evolved over 25

26 time. For instance, network structures including participating entities, their attributes, and FTA ties have been changed. We believe our inquiry into how similarities and dissimilarities of economic entities influence FTA formation is a productive avenue for future research on FTA network evolution. Our empirical analysis on choosing FTA partners tests competing theories of FTA formation by using social network analysis. Contrast to dyadic analysis, social network analysis estimates probability of presence and absence of FTA ties considering network structures such as homophily. Our analyses suggest that birds of a feather that are located in similar region have a tendency to form FTA ties together. Another evidences of this research show that not the similarities but the higher levels of democracy and larger economic mass are likely to attract more FTA ties. Thus, our research concludes that the gravity model of trade predicting more trade among closely located entities and large mass of economy also explains FTA formation. 26

27 References Bagwell, Kyle and Robert W. Staiger An Economic Theory of GATT. American Economic Review 89(4): Baier, Scott L. and Jeffrey H. Bergstrand Do Free Trade Agreements Actually Increase Members International Trade? Journal of International Economics 71(1): Baldwin, Richard E The Causes of Regionalism. The World Economy 20(7): Baldwin, Richard E. and Anthony J. Venables Regional Economic Integration. In Handbook of International Economics Vol. 3, eds. Gene M. Grossman and Kenneth Rogoff, Amsterdam: Elsevier. Bhagwati, Jagdish Regionalism and Multilateralism: An Overview. In New Dimensions in Regional Integration, eds. Jaime de Melo and Arvind Panagariya, New York: Cambridge University Press. Bhagwati, Jagdish "U.S. Trade Policy: The Infatuation with Free Trade Areas." In The Dangerous Drift to Preferential Trade Agreements, eds. Jagdish Bhagwati and Anne Krueger, Washington, D.C.: American Enterprise Institute for Public Policy Research. Bhagwati, Jagdish and Arvind Panagariya "The Theory of Preferential Trade Agreements: Historical Evolution and Current Trends." American Economic Review 86(2): Birch, Melissa H Economic Policy and the Transition to Democracy in Paraguay. In Economic Policy and the Transition to Democracy: The Latin American Experience, eds. Juan Antonio Morales and Gary McMahon, New York: St. Martin s Press. de Melo, Jamie and Arvind Panagariya Introduction. In New Dimensions in Regional Integration, eds. Jaime de Melo and Arvind Panagariya, New York: Cambridge University Press. Ethier, Wilfred Regionalism in a Multilateral World. Journal of Political Economy 106(6): Frankel, Jeffrey A Regional Trading Blocs. Washington, DC: Institute for International Economics. Frankel, Jeffrey A., Ernesto Stein, and Shang Jin Wei Trading Blocs and the Americas: The Natural, the Unnatural, and the Super Natural. Journal of Development Economics 47(1): Frankel, Jeffrey A., Ernesto Stein, and Shang Jin Wei Regional Trading Arrangements: Natural or Supernatural? American Economic Review 86(2): Frankel, Jeffrey A., Ernesto Stein, and Shang Jin Wei Continental Trading Blocs: Are They Natural or Supernatural? In The Regionalization of the World Economy, ed. Jeffrey A. Frankel, Chicago: University of Chicago Press. Furusawa, Taiji and Hideo Konishi Free Trade Networks. Journal of International Economics 72(2):

28 Goodreau, Steven M., James A. Kitts, and Martina Morris Birds of A Feather, or Friend of a Friend? Using Exponential Random Graph Models to Investigate Adolescent Social Networks. Demography 46 (1): Grossman, Gene M. and Elhanan Helpman "The Politics of Free Trade Agreements." The American Economic Review 85(4): Haggard, Stephen Regionalism in Asia and the Americas. In The Political Economy of Regionalism, eds. Edward D. Mansfield and Helen V. Milner, New York: Columbia University Press. Handcock, Mark S., David R. Hunter, Carter T. Butts, Steven M. Goodreau, and Martina Morris Statnet: Software Tools for the Representation, Visualization, Analysis and Simulation of Network Data. Journal of Statistical Software 24(1). Available at Haveman, Jon and David Hummels Trade Creation and Trade Diversion: New Empirical Results. Journal of Transnational Management and Development 3(2): Hunter, David R., Mark S. Handcock, and Cater T. Butts ergm: A Package to Fit, Simulate and Diagnose Expoential Family Model for Networks. Journal of Statistical Software 24 (3): Katzenstein, Peter J Introduction: Asian Regionalism in Comparative Perspective. In Network Power: Japan and Asia, eds. Peter J. Katzenstein and Takashi Shiraishi, Ithca, N.Y.: Cornell University Press. Kennan, John and Raymond Riezman Optimal Tariff Equilibria with Customs Union. Canadian Journal of Economics 113(1): Koehly, Laura M., Steven M. Goodreau, and Martina Morris Exponential Family Models for Sampled and Census Network Data. Sociological Methodology 34: Kim, Sangmoon and Eui Hang Shin "A Longitudinal Analysis of Globalization and Regionalization in International Trade: A Social Network Approach." Social Forces 81(2): Krishna, Pravin Regionalism and Multilateralism: A Political Economy Approach. Quarterly Journal of Economics 113(1): Krueger, Anne Are Preferential Trading Agreements Trade Liberalizing or Protectionist? Journal of Economic Perspectives 13(2): Krugman, Paul The Move Toward Free Trade Zones. In Policy Implications of Trade and Currency Zones: A Symposium, sponsored by the Federal Reserve Bank of Kansas City, Kansas City, Mo.: Federal Reserve Bank. Krugman, Paul Regionalism versus Multilateralism: Analytical Notes. In New Dimensions in Regional Integration, eds. Jaime de Melo and Arvind Panagariya, New York: Cambridge University Press. Krugman, Paul and Maurice Obstfeld International Economics: Theory and Politics. Reading, MA: Addison Wesley. Lawrence, Robert Z Regionalism, Multilateralism, and Deeper Integration. Washington, D.C.: Brookings Institution. Levy, Philip A Political Economic Analysis of Free Trade Agreements. American 28