National Borders Matter... Where One Draws the Lines Too

The fact that crossing a political border dramatically reduces trade flows has been widely documented in the literature. The increasing number of borders has surprisingly attracted much less attention. The number of independent countries has indeed risen from 72 in 1948 to 192 today. This paper estimates the effect of political disintegration since World War II on the measured growth in world trade. We first show that trade statistics should be considered carefully when assessing globalization over time, since the definition of trade partners varies over time. We document a sizeable resulting accounting artefact, which accounts for 17% of the growth in world trade since 1948. Second, we estimate that political disintegration alone since World War II has raised measured international trade flows by 9% but decreased actual trade flows (including inter-regional trade) by 4%.


I Introduction
International trade has grown almost twice as fast as world income since World War II, leading to an increase in the trade to GDP ratio from 24% in 1960 to 48% in 2003. The causes of economic globalization remain surprisingly disputed (Krugman, 1995). The usual suspects put forward in the literature are decreasing transport costs and the removal of tariffs and non-tariff barriers to trade. Hummels (2007) nevertheless finds little systematic evidence documenting the decline in transport costs. He shows that, whereas air shipping experienced a sharp cost reduction between 1955 and 2004, ocean shipping, which accounts for 99% of world trade by weight, did not register such a decline. 1 These explanations relate to the dissolution of national borders, either because of technological progress in transport and communication or of the worldwide implementation of free trade policies. A third explanation relates to the drawing of political borders. As stated by Krugman (1995, p.340), "if international trade only includes shipments that cross the borders, it is clear that the volume of trade depends quite a lot on where one draws the line". In this respect, an outstanding feature of the past five decades is the increasing number of sovereign nations. The number of countries has indeed increased from 72 in 1948 to 192 today, thanks in particular to the decolonization process and the break-up of the Soviet Union. Figure 1 clearly suggests that the number of independent countries and global trade openness measured by export plus imports over GDP are correlated. 2 The aim of this paper is to investigate how and to what extent political disintegration has affected the volume and geography of world trade since World War II.
The break-up of nations affects international trade through three channels. On the one hand, former intra-national trade flows become international trade flows. Since these shipments existed before independence, recording them as international flows creates an accounting artefact in the measurement of international trade over time. Moreover, pairs of countries ever in a colonial relationship or former members of the same country share linkages and economic, cultural and institutional characteristics that make them trade more than average pairs of countries (Rose, 2000;de Sousa and Lamotte, 2007;Head et al., 2007). On the other hand, the creation of political borders change relative trade costs. It creates impediments to trade -tariffs as well as non-tariffs barriers to international flows such as independent currencies or different standards and competition policies (Anderson and van Wincoop, 2004) -that dramatically reduce international trade in comparison to  1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 (McCallum, 1995;Anderson and van Wincoop, 2003). Finally, since bilateral trade flows depend on relative trade costs (Anderson and van Wincoop, 2003), creating border barriers with some partners should increase trade with other partners.
For instance, the Czech Republic would be expected to trade less with Slovakia after the break-up of Czechoslovakia, but more with any third country.
The contribution of this paper is threefold. We first provide an assessment of the growth in international trade since World War II due to the variation over time in the sample of countries recorded in the international statistics. We show that this accounting artefact is sizeable and accounts for one sixth in the growth of measured international trade since World War II. Part of the accounting artefact comes from the fact that the trade flows of some former colonies were not recorded at all in international statistics (neither independently nor as part of their colonizer's trade) up to their independence. Second, we investigate the impact of the creation of new political borders alone on the volume and geography of world trade since World War II. Based on a theoretically motivated gravity model of trade, we estimate world trade flows in 2005 and in a counterfactual world in 2005 with 1945 borders. We show that newly independent states trade less with former members of the same country (or their former colonizer) but trade more with the rest of the world. Overall, our empirical results suggest that political disintegration has increased measured international trade flows by 9% but decreased actual (including inter-regional) trade flows by 4%. Third, we show that the distribution of GDP abroad affects a country's trade to GDP ratio; a country's trade openness increases when GDP abroad is distributed among a larger number of trade partners.
Several papers investigate the causes of the growth in world trade as a share of GDP. Estevadeordal et al. (2003) study the causes of the rise and fall of the trade to output ratio during the 1870-1939 period. They show that the driving forces of trade globalization were a fall in transport costs and the rise of the gold standard. Rose (1991) and Baier and Bergstrand (2001) focus on the second wave of globalization after World War II. They both work on a sample of OECD countries. Baier and Bergstrand (2001) show that two thirds of the growth in trade between the late 1960s and the late 1980s can be explained by income growth, 25% by a reduction in tariffs and 8% by a decline in transport costs. 3 All these papers investigate the determinants of international trade growth on a subset of countries, i.e. they investigate the intensive margin of the growth in international trade.
They cannot address the effect of the change in underlying features of the data such as the distribution of GDP among an increasing number of potential trading partners, i.e. the extensive margin of the growth in international trade. Yet world trade has increased by a factor of 30 between 1948 and 2007, while over the same period trade between OECD countries 4 has increased by a factor of only 20. At the same time, the international trading system has expanded from 72 × 71/2 = 2556 potential bilateral trade relationships in 1948 to 192 × 191/2 = 18336 today. Assessing the contribution of this extensive margin to the growth in international trade requires a structural estimation of a theoretically motivated gravity model. The paper proceeds as follows. The next section provides evidence and estimates of the accounting artefact related to variations over time in countries reported in international trade statistics. The third section presents a theoretically based gravity framework allowing the estimation of the impact of border changes on the geography and volume of world trade flows. Section 4 reports the estimation results.

II An assessment of the accounting artefact
The assessment of globalization through measured world trade depends crucially on the number of trade partners considered:

Independent states and trade partners in the DOTS
The DOTS database reports trade flows between "entities" that are not necessarily independent countries as defined by international law and practice. As put forward by Russett et al. (1968) and Small and Singer (1982, p38-46), in order to be registered as a state member of the international system, "the entity must be a member of the United Nations or League of Nations, or have population greater than 500,000 and receive diplomatic missions from two major powers". 5 Trade partners in the DOTS database include some territorial entities that are not independent states, and some independent states are not considered as trade partners. the DOTS. This discrepancy is due to the fact that some independent countries (Taiwan, Monaco, Andorra, Liechtenstein, Botswana, Lesotho, Namibia and Swaziland 6 ) were not taken into account whereas some territorial entities, that were not sovereign states, were. 7 In addition, Figure 2 underlines that the DOTS matrix is not squared, since the number of declared trade partners varies by trade origin and increases over time.   Figure 2: Sovereign states, trade origins and trade partners size of the DOTS matrix has accordingly been multiplied by more than two.
The size of the DOTS matrix therefore depends to a large extent on changes in political borders over time. New independent countries, however, did not necessarily enter the DOTS in the years following their independence. Some colonies were already taken into account in the DOTS in 1948 (see Table 1). Subsequently, in half of the cases, a new independent country was recorded as a trade origin before its independence. Since 1948, the number of effective (positive) bilateral trade flows has been continuously growing and has increased more than six-fold over the period. New trade flows (i.e. involving a new trade origin and/or partner) represent an increasing share of bilateral five French overseas departments (Guadeloupe, Guyana, Martinique, Reunion and French Guiana) were considered as distinct trade origins even if they were (and still are) part of France. On the contrary, Belgium and Luxembourg were recorded as a single trade partner from 1948 to 1996. 5 trade flows, from 4% in 1950 to 55% in 2007 (see Figure 3 and Table 2). 9 In comparison, the share of trade flows between countries that used to belong to the same statistical entity made up 1% of world trade transactions in 2007 (see table 2). The value of trade flows involving new entities has also increased dramatically, to stand at 28% in the 2007.  In the general case of a country break-up, former inter-regional trade flows become international flows because former regions of the same country become sovereign nations. 10 For instance, the break-up of Czechoslovakia in 1993 resulted in the creation of two new countries: the Czech Republic and Slovakia. Czechoslovakian trade, however, included the trade of all its regions with the rest of the world. In this case, we thus define trade flows between the Czech Republic and Slovakia (i.e. USD 4882 millions in 1993) as a statistical artefact since they existed before independence but were not recorded in the DOTS.
A country's independence may nevertheless have more severe consequences on the measurement of international trade. The trade of some independent states, territories or colonies was simply not recorded at all in international trade statistics before their independence or their creation as a trade origin. In this case, all trade flows of these countries contribute to the accounting artefact since they existed but were not recorded in 1948 and appeared in international statistics between 1948 and 2007. To illustrate the relevance of this second source of accounting artefact, let's consider the case of the former French West and Equatorial African colonies, whose trade flows were simply not included in the DOTS database even as part of the French imports or exports. 11 Table 3 shows that in 1962 these countries represented 4.25% of French foreign trade and that, from 1959 to 1962, their independence and creation as a trade origin accounted for 13.4% of the increase in French trade. In addition, although they represented less than one percent of world trade in 1962, their creation accounted for almost 4% of the increase in the value of world trade flows between 1959 and 1962.
The accounting artefact is sizeable also at the world level. In order to assess the bias introduced in world trade statistics by the creation of new trade origins and trade partners, we compute the share of the growth in world trade due to flows that were not recorded as international trade flows in 1948. More precisely, we subtract from the total value of imports recorded in the DOTS since 1948 the total value of trade between country pairs 10 As mentioned previously trade origins in the DOTS are not necessarily independent states, and their definition has changed over time. Such changes are not neutral in term of measured international trade. For instance, from 1948 to 1996 Belgium and Luxembourg were considered as a single trade partner. In 1997, these two countries were for the first time considered as distinct trade partners, and their bilateral trade recorded in the DOTS. 11 The French Customs and Excise Department (1959) defines the French statistical territory as continental France (including the free zones of pays de Gex and Haute-Savoie), Corsica, Monaco and Sarre.  Table 4 confirm that the accounting artefact is sizeable: it accounts for more than one sixth of the growth in world trade since World War II. The accounting artefact was particularly significant in the 1950s and 1960s due to the decolonization process and in the 1990s due to the break-up of the Soviet Union. 13 Overall, 17% of world trade growth since 1948 is related to the inclusion of new trade origins in the DOTS database.

Results in
This section shows that variations over time in the number and definition of trade origins and partners in international trade statistics inflate the measured growth in international trade. As underlined above, the creation of a new trade origin does not necessarily reflect the recognition of a new sovereign state. Moreover, changing political borders may 12 Gleditch (2002) argues that many zeros in the DOTS are problematic and should be treated as missing observations.
13 Note that the German reunification has had no effect on the accounting artefact since trade flows between the Federal Republic of Germany and the German Democratic Republic was null before reunification.
8 affect trade with third countries. Fully quantifying the contribution of changing political borders on the growth in world trade therefore requires to go beyond simple descriptive statistics and to structurally estimate a trade model. In the next section, we investigate the impact of changing political borders on the volume and geography of world trade based on a micro-founded gravity model of tradeà la Anderson and van Wincoop (2003).

III The gravity model and borders
The gravity model of trade is based on two building blocks (Anderson and van Wincoop, 2003). Consider first N countries/regions each specialized in the production of one good differentiated by place of origin. We assume the supply of each good to be fixed. Let then add identical, homothetic preferences represented by the following CES utility function: where c ij is consumption of goods produced in country i by consumer of country j and σ is the elasticity of substitution between all goods. Country j's consumers maximise (2) subject to the budget constraint: where p ij is the price of goods produced in country i for country j's consumers and y j is country j's nominal income. 9 Let t ij be the variable trade costs on exports from i to j. Then p ij = p i t ij where p i is country i's supply price. The value of exports from country i to country j is x ij = p ij c ij and total income of country i is y i = j∈N x ij .
Maximization of (2) subject to (3) yields: where P j is given by: Market clearance implies: Following Anderson and van Wincoop (2003), using (6) to solve for the unknown prices p i together with (4) and (5), we obtain: 1 World openness under free trade Under free trade, i.e. without transport costs or tariffs so that prices are identical for all countries, bilateral trade is determined by the product of the two countries' GDPs. The analysis of world trade is then straightforward: global trade openness depends only on the size distribution of countries. Since prices are identical for all countries, we normalize them to unity: p ij = 1∀i, j. We thus have: In this framework, Helpman (1987) shows that, for a group of countries A, the fraction 10 of GDP traded inside the region is defined by: where T A is the volume of intra-regional trade, Y A is the regional GDP, s A is the share of region A in world GDP and s iA are the share of region A's GDP of each country i in region A. At the world level, Helpman (1987)'s equation yields: where T W is the volume of world trade, N is the total number of countries in the world and s i is i's share of world GDP. The last term on the right hand side of (12) is a "size dispersion index". The latter is minimized for N identical countries whose share of world GDP is 1 N . Hence, under free trade, the relationship between global trade openness and the number of independent countries is straightforward: an increase in the number of countries increases the share of world production traded internationally by decreasing the "size dispersion index". in 1950 was with an outside world slightly larger than itself but is now five times larger.

Gravity with trade barriers
With transport costs, the analysis is less straightforward since bilateral trade depends on relative trade costs (Anderson and van Wincoop, 2003). Any change in bilateral trade costs between two countries therefore affects the trade flows of each country with its other partners. A country break-up creates border barriers between former regions of the same country. This has two effects on international trade. First, it reduces actual trade flows between the two new independent countries but increases recorded international trade flows since trade between regions is not recorded as international trade before independence.
Second, it increases the average trade barriers that new countries face with all their trading partners. A new independent country should thus trade less with the former regions of the same country but trade more with other partners.
As underlined by Feenstra (2004) and Anderson and van Wincoop (2003), equation (7) 11 can be properly estimated using country fixed effects that account for inward and outward multilateral resistance terms Π i and P j . We can therefore consistently estimate the effect of national independence on trade between two former members of the same country from (7). However, to assess the effect of country break-ups on trade with other partners, we need to compute the multilateral resistance terms from (8) and (9).
Assuming symmetric trade barriers, t ij = t ji , we have Π i = P i , and: We assume the following trade cost function: The vector of observable bilateral linkages affecting trade costs includes variables measuring geographical, cultural and historical proximity as well as trade policy variables: where Contig ij , Lang ij and Comcur ij are dummies for countries sharing a common border, a common official language and a common currency respectively. RT A ij and W T O ij measure common membership of a regional trade agreement and the World Trade Organization, and Smctry ij , Colony ij and ComCol ij are dummies equal to 1 for, respectively, regions of the same country, countries ever in a colonial relationship and sharing a common colonizer since World War II.
So we have: The β's coefficients in (16) may be consistently estimated from (7). We may then solve the vector of P 1−σ j using the system of N goods market equilibrium conditions (16), estimated coefficients from (5) and GDP shares. We are thus able to estimate the impact of the establishment of a political border on trade with former regions of the same country and with the rest of the world.

IV Empirics
Our empirical strategy consists of three steps. First, we need to estimate the impact of the components of the trade cost vector using a gravity equation with country fixed effects. Indicators (WDI). The data on regional trade agreements originates from Vicard (2009) and data on currency unions are an extended version of Glick and Rose (2002)  countries that used to belong to the same country (Czech Republic and Slovakia) as a single category (Colony ij ). We define a second category for countries sharing a common colonizer (ComCol ij ). 16

Same country effect
In this section, we take advantage of the discrepancy between sovereign nations and entities in the DOTS to estimate the impact on bilateral trade of belonging to the same country.
In 2005, the DOTS database reports trade flows for 20 entities that are not sovereign states. 17 Substituting trade costs (14) into (7) and log-linearizing, we obtain the following econometric model: (17) where I i and E i are importer's and exporter's fixed effects and ε ij is the error term.
The estimation of (17) by OLS may indeed yield biased estimates: because of Jensen's inequality (E(lnε) = lnE(ε)). Santos Silva and Tenreyro (2006) show that estimating gravity equations in the standard log-linear form yields biased parameter estimates. They suggest using a Poisson quasi maximum likelihood (PQML) estimator, which yields consistent estimates in the presence of heteroscedasticity. The PQML technique also enables us to avoid dropping zero trade values, which represent 31% of observations in our dataset.
We thus estimate (17) in exponential form on a cross-section in 2005 using a PQML estimator.
Results are presented in column (1) of Table 5. The coefficients on standard gravity variables are in line with the literature. The coefficient on the common colonizer dummy is however not significant. Since we work on a cross-section in 2005, colonial business linkages may have vanished since independence (Head et al., 2007). However, when we include a dummy for countries sharing a common colonizer and still in an ongoing colonial 16 These definitions slightly differ from those adopted in Rose (2000) or Head et al. (2007). For instance, Mali and Côte d'Ivoire were part of French West Africa before their independence. Each of them had a direct colonial link with France, but they were also part of the same territorial entity. Mali and Côte d'Ivoire are accordingly included in Colony ij instead of ComCol ij .
17 These entities are small and remote from the rest of the country they belong to. Our results should accordingly not be regarded as an estimation of the border effect put forward by McCallum (1995) and Anderson and van Wincoop (2003) using data on Canadian provinces and US States. They are however much more representative of the "border effect" between colonies and colonizer, since most colonies were also small and remote from their colonizer. relationship, its coefficient is also insignificant, confirming the insignificance of common colonial history for trade relationships after independence. The inclusion of country fixed effects controlling for legal origin may however partially explain this result. Finally, the insignificant coefficient on the common currency dummy may reflect the fact that we consider the euro together with other common currencies without a long panel dataset (Frankel, 2008).

Multilateral resistance terms
In this section, we compute the multilateral resistance terms using the system of goods market-equilibrium conditions (16) and the trade costs estimated from equation (17 We have complete data for 181 countries, of which 448 pairs have been part of the same country or in a colonial relationship since 1945. For each country, computing (16) requires data on all trade partners, including the country itself. A difficulty here is to measure internal distance (Head and Mayer, 2002;Redding and Venables, 2004). We use 3 different measures of internal distance. First, we consider intra-national distance as the distance to the nearest neighbor divided by four (Wei, 1996;Anderson and van Wincoop, 2003). Second, we compute distance as dist ii = 0.67 area/Π, which measures the average distance between producers and consumers in a country considered as a circle. Finally, we compute distances based on bilateral distances between the largest cities weighted by 19 Note that the difference in the definition of the dummy for current and past colonial relationships does not drive this result. 20 The selection bias due to the exclusion of zero trade flows when using OLS is not significant (column (2) of Table 5). 21 Computationally, we solve (16) for P 1−σ i so that we do not need estimates of elasticity of substitution σ. their population (Head and Mayer, 2002). 22 The measurement of internal and international distances is consistent with this weighted measure of distance. The common border, common language, and the same country and colonial relationship dummies are also set to unity for intranational trade costs. We compute trade costs from (14) for the year 2005 and for our counterfactual world for each of the internal distance measures. 23 The level of multilateral resistance P σ−1 is, as expected, lower for large countries such as the United States and Japan, and countries surrounded by other rich countries, such as EU countries (see appendix) or small and rich countries with large neighbors (Singapore, Hong-Kong). Conversely, small countries and remote islands face the largest multilateral resistances. The level of P σ−1 depends on the measure of internal distance used. Wei (1996)'s measure of internal distance places more weight on domestic GDP and reduces P σ−1 for rich countries and increases it for poor countries. The opposite is true for the measure of internal distance based on domestic area. Weighted distance increases P σ−1 for all countries with respect to Wei (1996)'s measure, but more for small and remote countries. The ranking of P σ−1 however changes only at the margin. Large countries, like the United States, have relatively larger internal distance with area based and weighted distance measures than with Wei (1996)'s measure, and have a relatively larger multilateral resistance compared to smaller countries like Japan.
The effect of our counterfactual exercise on multilateral resistance depends on the kind of countries considered (see appendix). Switching to 1945 borders decreases multilateral resistance in former colonies and former regions of a large country. The decrease is large for countries close to their former colonizer (Papua New Guinea, Namibia) or the former regions of the same country (former members of the USSR). Conversely, OECD countries exhibit no change in P σ−1 . France and the UK are exceptions since they experience a small decrease in their multilateral resistance, due to the removal of border barriers with their former colonies. Since former colonies are small and remote, removing border barriers with them has a limited impact on French and English average trade costs. Russia also exhibits an increase in its multilateral resistance. These results are basically similar for all three measures of internal distance. 22 The general formula is dist ij = k∈i (pop k /pop i ) l∈j (pop l /pop j )dist kl where pop k is the population of agglomeration k belonging to country i and dist kl is the geodesic distance between agglomeration k and l.
23 When weighted distance is used to measure internal distance, we also re-estimate (17) using weighted international distance and use the estimated coefficients on trade costs. Results are presented in Table 7 in Appendix A.

17
In this final section, we estimate trade flows for our sample of 181 countries in 2005 and for our counterfactual world. 24 More specifically, we compute: (18) where h = a stands for the actual world in 2005 and h = c for our counterfactual world.
x h ij is the predicted trade flow from i to j and Smctry h ij and Colony h ij are respectively regions of the same country and countries ever in a colonial relationship since World War II in each state of the world.P t i have been computed in the preceding section. Results using each of the three measures of internal distance are presented in table 6. The results are qualitatively similar for all three measures of internal distance. We thus focus our discussion on the simplest measure, the Wei (1996)  As regards the geography of world trade, trade between former regions of the same country and between colonizers and their colonies in 2005 is half its value in our counterfac-24 German reunification is not taken into account here since Germany was divided into the Federal Republic of Germany and the German Democratic Republic in 1949. 25 We calculate R-squared as the squared correlation between the observed and the predicted bilateral exports. tual world. The presence of new border barriers between these countries also affects their exchanges with the rest of the world because it increases their average trade barriers. In the aggregate, former regions of the same country and colonizers and their colonies trade 1.9% more with third countries. A country's break-up therefore affects third countries' trade.
Finally, trade between OECD countries is only marginally affected by new political borders. France and the UK are exceptions since their bilateral trade is estimated to have increased by 2% thanks to the decolonization process. In the aggregate, French and English trade with the rest of OECD countries is 1% larger in 2005 than in our counterfactual world. Notes: Computed from specification (1) in Table 5 for the Wei and area distance measures, and from specification (1) in Table 7 for the weighted distance measure. *Trade between colonizers and their colonies and regions of the same country.

V Conclusion
This Second, we lay out an empirical strategy to investigate the impact of country breakups on world trade based on a structural estimation of a gravity model of tradeà la Anderson and van Wincoop (2003). We find that the rise in the number of independent countries since WWII has increased international trade by 9%, by raising Appendix A