National borders and international trade: evidence from the European Union

National borders and international trade: evidence from the European Union Volker Nitsch Bankgesellschaft Berlin Abstract.In this paper the impact of national borders on international trade within the European Union is considered. Using a gravity model, I find that, averaged over all EU countries, intranational trade is about ten times as high as international trade with an EU partner country of similar size and distance. This relatively strong home bias suggests that even within the European Union national borders still have a decisive impact on trade patterns. JEL Classification: F02, F14, F15, O52 Frontières nationales et commerce international: le cas de l Union européenne. Ce mémoire examine l impact des frontières nationales sur le commerce international à l intérieur de l Union européenne. On montre que, en moyenne pour tous les pays de l Union européenne, le commerce intra-national est à peu près dix fois plus important que le commerce international avec un pays partenaire de l Union européenne de même taille et à même distance. Ce degré relativement important de préférence nationale suggère que, même dans le cadre de l Union européenne, les frontières nationales ont encore un impact déterminant sur les patterns de commerce. 1. Introduction John McCallum ~1995! has recently presented an empirical puzzle. Using data from the 1988 input-output tables for Canada, a data set that is apparently unique in allowing the analysis of trade patterns between regions of different countries, he finds that Canadian provinces trade about twenty times more with each other than An earlier version of this paper was presented at the University of Munich and the 1998 meeting of the European Economic Association in Berlin. I am thankful to Karsten Junius and two anonymous referees for valuable comments and to John Helliwell, Dalia Marin, and Holger Wolf for helpful discussions. This paper was partly written while I was visiting the Kiel Institute of World Economics. Any opinions expressed do not necessarily represent the views of the Bankgesellschaft Berlin. Email: volker.nitsch@bankgesellschaft.de Canadian Journal of Economics 0 Revue canadienne d Economique, Vol. 33, No. 4 November 0 novembre 2000. Printed in Canada 0 Imprimé au Canada 0008-4085 0 00 0 1091 1105 0 Canadian Economics Association

1092 V. Nitsch they do with U.S. states of a similar economic size and proximity. This result suggests a surprisingly large home bias in international trade, since the national border between Canada and the United States is commonly assumed to be one of the most easily passable lines in the world and, therefore, to have relatively little effect on trade. 1 Recent attempts to confirm McCallum s finding, however, have delivered mixed results. On the one hand, John Helliwell ~1996, 1998! has extended McCallum s basic sample to cover 1988 96 and also has performed a number of robustness checks, finding only slight variations in the estimated border effects. On the other hand, Shang-jin Wei ~1996! applies an interesting method to generate measures of intranational trade volumes and distances for OECD countries and finds a home bias of only two and one-half, that is, a bias substantially smaller than McCallum s and Helliwell s estimates for Canada. This paper contributes to the literature along several lines. First, I examine directly the impact of national borders on international trade in another region of the world that is assumed to be highly integrated, the European Union. A first attempt to estimate the border effect in EU member countries has already been made by Wei ~1996!, finding a very small home bias of about 1.7. This result, however, may suffer from specification problems ~see Helliwell 1997! and missing data. 2 Second, I propose a new measure for estimating the average of intranational distances. Instead of using ad hoc measures based on the distances between particular cities ~as proposed by Wei 1996 and Holger Wolf 1997!, I suggest making the average distance within a country a function of country size. Third, a new set of production data is utilized. While Wei s ~1996! analysis was largely based on OECD statistics, providing accurate data for only a few countries, an almost complete data set of total goods production is directly available from the statistics agency of the European Union, Eurostat. The set-up of the paper is as follows. In section 2 some methodological thoughts are presented. The data are described in section 3. The results are presented in section 4, and the paper is concluded in section 5. 2. Methodology The basic framework for estimating the home country bias in the goods market is the long-established and empirically highly successful gravity equation, wherein the volume of trade between any two countries is related to the economic size of these countries and the geographic distance between them: ln~x ij! a b 1 ln~y i! b 2 ln~y j! b 3 ln~d ij! e ij, ~1! 1 John Helliwell ~1996!, for example, reports survey results where most respondents estimate the Canadian home country bias to be in the order of 0.6 to 1.4, that is, on average, a Canadian province should export about the same amount to another province as to a U.S. state, after one controls for sizes and distances. 2 Wei s OECD sample does not include Belgium0Luxembourg, Greece, and Ireland.

National borders and international trade 1093 where X ij are exports from country i to country j; Y i and Y j are the GDP of countries i and j, respectively; and D ij denotes the distance between the two countries. This basic specification can be augmented by other variables that are assumed to be related to the bilateral volume of trade. Authors of previous studies, for example, have found statistically significant coefficients on dummies for country pairs that speak a common language or share a common border. Moreover, it has recently been argued ~see, e.g., Deardorff 1998! that the relative distances of trading partners have an impact on the volume of trade. In particular, remote countries ~e.g., Australia and New Zealand! can be expected to trade more with each other than two countries that are separated by the same absolute distance but are geographically well positioned near other markets ~e.g., Germany and Portugal!. Following Wei s ~1996! theoretical discussion, then, which itself is based on Deardorff ~1998!, the remoteness of a country i is defined as the reciprocal of country k s GDP divided by the bilateral distance between country i and country k summed over all trading partners of country i ~in the sample!: 3 R i ~S k @Y k 0D ik #! 1. ~2! The measured remoteness of countries averaged over the period from 1980 to 1990 is plotted in figure 1. As expected, the small countries, Belgium0Luxembourg and the Netherlands, are the least remote countries in the sample. It is somewhat surprising, however, that the United Kingdom also is, according to our procedure of calculating remoteness, geographically well positioned in the European Union, largely owing to the closeness of London to other European capitals. A similarly interesting observation is that Portugal is only slightly more remote than Italy, illustrating the property of our measure that if nearby countries are also big countries ~Spain, in the case of Portugal!, then the existence of distant countries is less relevant. The respective country will have a low remoteness index. Given this set-up, the home bias is estimated by adding a dummy that takes the value of one for trade flows within countries and zero otherwise. Hence, the following benchmark specification is adopted: ln~x ij! a ghome b 1 ln~y i! b 2 ln~y j! b 3 ln~d ij! b 4 ln~r i! b 5 ln~r j! b 6 Language b 7 Adjacency e ij. ~3! 2.1. How to measure intra-national distances Taking data on trade flows between subnational units ~i.e., provinces, states, or regions!, this regression can be estimated directly. As intranational trade statistics 3 Departing from his theoretical derivation, Shang-jin Wei ~1996! has measured the remoteness of a country as the mean distance to all trading partners weighted by their respective output shares. Helliwell ~1997, 170!, however, has noted that this functional form, in which both third-country GDP and bilateral distance enter with the same sign, is inconsistent with the theoretical requirements of the gravity approach. It is interesting to note that Wei ~1996, appendix A2! also provides a sensitivity analysis illustrating that his results depend to a large degree on the ~inappropriate! definition of his empirical remoteness measure.

1094 V. Nitsch FIGURE 1 Remoteness of countries averaged over 1980 90 NOTE: Remoteness is calculated as R i ~S k ~Y k 0D ik!! 1. are only rarely available, 4 however, measures for trade volumes within countries as well as for internal distances have to be constructed. In fact, one of the contributions of Wei ~1996! is the provision of highly imaginative methods to approximate these two missing data series. Following Wei s assumption, then, that a country s exports to itself are simply the difference between its total production and its total exports to the rest of the world, data for intranational trade volumes are generated by subtracting merchandise exports from the value of production. Given this intuitive procedure to obtain data for domestic sales, however, the main difficulty is to determine the average distance of the transportation of goods within each country. 5 In fact, it is quite crucial to find a way that allows us to specify intranational distances correctly, since the size of the estimated home country bias will be proportional to the value of the average internal distance. In view of this apparent sensitivity of the results of the estimated home bias to the assumption about internal distances, it is surprising that, until recently, there 4 Helliwell ~1996!, for example, has argued that the surprising finding of a large home country bias in the goods market became evident only after the publication of the Statistics Canada interprovincial trade data. Recently, Holger Wolf ~1997! has used a comparable data set of trade flows between U.S. states to estimate the home bias on the subnational level. 5 If we follow conventional gravity models, the distance between countries is approximated as the direct distance between the economic centers ~usually the capitals! of the respective countries.

National borders and international trade 1095 has been relatively little effort to determine the average intranational distance. Basically, two ad hoc procedures are widely used in the literature. On the one hand, Wei ~1996! has proposed measuring the average distance within a country as one-half of the distance from the economic centre of that country to the border of the nearest neighbour. On the other hand, Wolf ~1997! has approximated the distance for intrastate trade as the distance between the largest and the second-largest city within that state. Both measures, however, suffer from the problem of possible geographical inconsistencies. In particular, it is very questionable that the location of two particular cities can deliver an indicator for average internal distances that is congruent across countries. To illustrate some of the problems, consider Wei s ~1996! one-half to the border rule. In the actual implementation of that rule, Wei calculates internal distances as one-quarter of the distance between the capital of a country to the capital of the nearest neighbour. This approach, however, implicitly assumes that the cities are equally close to the border, a highly unrealistic proposition. For example, take Denmark s internal distance, which is, according to Wei s procedure, one-quarter of the distance between Copenhagen and Bonn ~i.e., @0.25 660 km # 165 km!, although Bonn s distance from the Danish border is about 510 km and Copenhagen s distance from the German border is only about 150 km. Another problem becomes apparent if the distance between the capitals of two neighbouring countries is taken as an approximation for the average intranational distance in both countries. Wei ~1996!, for example, assumes that Germany and France have the same average of internal distances ~namely, one-quarter of the distance between Bonn and Paris! although France has a land area that is more than double than that of Germany. In order to avoid the problems connected with Wei s ~1996! and Wolf s ~1997! ad hoc measures, it is suggested that the average distance within a particular country be made a function of the size of the country. Although this procedure does not take into account different shapes, internal structures, or trading patterns and therefore also has obvious shortcomings ~e.g., in the case of large and narrow countries like Chile or Norway!, it delivers at least a measure that is consistent across countries. A first step, then, is simply to approximate the average intranational distance with the square root of the land area. As it is questionable, however, whether this intuitive approach automatically delivers correct measures for internal distances, it may be useful to model a specific geographic shape and to calculate the mean distance among a number of particular points within that economy. In particular, the aim is to examine whether it is necessary to multiply the square root of the land area by a specific factor in order to get consistency with average internal distances in hypothetical spatial distributions; and if so, to find a formulation that is fairly robust for different shapes. To illustrate the basic idea, consider a simple circular economy with three equalsized cities, one in the centre and the other two located on opposite sides of the circle. If one allows for trade within cities, the average internal distance in that country is about 0.89 times the radius of the circle. Extending this simulation, then,

1096 V. Nitsch and introducing additional locations or varying the shape of the economy apparently affects the result only marginally. In fact, after some experimentation with more complex structures ~using grids of equally spaced points! it turns out that the radius of a circle or 10!p~ 0.56! times the square root of the area may indeed be a good approximation for the average internal distance and is therefore used in the empirical analysis. 6 In interpreting the empirical results, two points are particularly noteworthy. First, the assumption of an equal distribution of economic activity in calculating the mean distance ignores the effects of urban areas in which economic activity is highly concentrated in space. Therefore, a scaling factor of 0.56 probably provides an upward-biased estimate of the average internal distance and, accordingly, yields overestimates of the home bias. However, cross-checking with available data for real trading distances suggests that this problem may be of only minor importance. In fact, Helliwell and Verdier ~1999! have proposed a very promising approach to compute the average trading distance within a country. Using the standard gravity framework, they divide a country s population among cities and rural areas, each of which has a reasonably homogeneous population density within it, and then calculate the weighted distance within and between these sub-units. Applying this detailed technique to Canadian provinces, they find that the average value for intraprovincial trade distances is about 0.31 times the square root of the area and thus is somewhat smaller than the proposed factor of 0.56. If one applies their method to Canada as a whole, however, and ignores highly skewed trade within provinces, the average distance is about 1240 km or 0.50 times the square root of the Canadian area of 6,060,840 km 2, a scaling factor that is surprisingly close to the result from hypothetical distributions. In sum, 0.56 times the square root of the area appears to be a reasonable proxy for the average trading distance within a country. In future work, the aim will be to extend Helliwell and Verdier s ~1999! method, which takes a far more detailed account of a country s internal geographic structure to a larger sample of countries. Second, it should be noted that regressions based on the total volume of trade within a country and using 0.56 times the square root of the area or any other approximation for the average intranational distance will not deliver results that are compatible with McCallum s ~1995! and Helliwell s ~1996! initial calculations of the Canadian home bias. Remember that the finding of a Canadian border effect of factor 20 is based on a comparison of interprovincial trade, on the one hand, and trade between Canadian provinces and U.S. states, on the other hand, with no assumptions about trade within Canadian provinces. In contrast, Wei s ~1996! empirical 6 In an earlier version of this paper, I approximated the scaling factor by calculating the tradeweighted distance within and between Canadian provinces ~yielding values of 0.2 for total Canadian trade and 0.6 for interprovincial trade only!. Instead of measuring the average distance for potential trade, however, this approach already incorporates the trade-reducing effects of distance, so that the calculated values are too low. I am thankful to John Helliwell and a referee for making this point.

National borders and international trade 1097 approach additionally includes intraprovincial trade flows ~owing to the lack of data for volumes of trade between subnational units!, thereby introducing a problem that has the potential to seriously distort the results. In fact, it is impossible to determine how much trade actually takes place within urban agglomerations. For trade over very small distances when goods are hardly moved at all, however, it is questionable whether the distance-cost formulation of the gravity framework holds. 7 3. Data In previous studies on the home bias in European trade production data from the OECD have been applied. As the OECD s Industrial Structure Statistics provides accurate data for only a few countries ~Finland, France, Norway, and Sweden!, however, usually a complicated procedure was applied to construct measures for most of the countries in the sample. Moreover, no data apparently were available for Belgium0Luxembourg, Greece, and Ireland. In this paper a different and more detailed set of production data is used, compiled by the statistics agency of the European Union, Eurostat. In particular, Eurostat s yearly editions of Structure and Activity of Industry allow the construction of an almost complete data set of total goods production ~measured in ECU! for all EU member countries from 1979 to 1990. Missing data, then, are approximated by multiplying a country s GDP by the average of the production-to-gdp ratio of the last and the next year for which production data are directly available. Eurostat also provides data on bilateral and total exports which are taken from different editions of Eurostat s Statistische Grundzahlen der Gemeinschaft. GDP and population data are obtained from a statistics compilation by the European Commission ~European Economy No. 60!. Finally, the distance between countries is measured as the great circle distance between the national capitals, with the exception of Germany, for which Frankfurt was chosen as the economic centre. In sum, then, the data set comprises information from ten EU countries 8 ~with pooled data for Belgium and Luxembourg! for the period from 1979 to 1990. After 1982 Eurostat also reports data for Portugal and Spain, which permits enlargement of the data set to twelve countries, so that there are for each year a total of ~9 9! 81 and ~11 11! 121 data points, respectively. 7 It is surely debatable which formulation of a country s home bias appears to be preferable. At a first look, Wei s ~1996! approach of defining home bias as the ratio of total domestic sales to exports relative to some baseline gravity model seems to be more convincing, since it comprises a country s total intranational trade, while McCallum s ~1995! estimate ignores most of a country s internal trade and is probably also influenced by the actual definition of provinces. McCallum s approach, however, also has the invaluable advantage of comparing trade flows over similar distances and thus on an equivalent basis, in which only the border effects differentiate between national and international trade flows. 8 The countries are Belgium0Luxembourg, Denmark, France, West Germany, Greece, Ireland, Italy, the Netherlands, and the United Kingdom.

1098 V. Nitsch TABLE 1 Home country bias in the European Union, 1979 90 Home 1.92** 2.38** 2.18** 2.43** 2.43** 2.10** ~0.20! ~0.21! ~0.20! ~0.21! ~0.24! ~0.19! ln~distance ij! 1.07** 0.86** 0.96** 0.85** 0.79** 1.04** ~0.08! ~0.08! ~0.08! ~0.08! ~0.12! ~0.07! ln~gdp i! 0.67** 0.67** 0.69** 0.68** 0.68** 0.77** ~0.04! ~0.04! ~0.04! ~0.04! ~0.04! ~0.04! ln~gdp j! 0.71** 0.69** 0.72** 0.70** 0.71** 0.86** ~0.04! ~0.04! ~0.04! ~0.04! ~0.04! ~0.04! Adjacency 0.68** 0.57** 0.60** 0.57** ~0.15! ~0.16! ~0.16! ~0.15! Language 0.68** 0.29 0.25 0.15 ~0.21! ~0.22! ~0.22! ~0.20! ln~remote i! 0.24*** ~0.13! ln~remote j! 0.09 ~0.13! # Observations 81 12 81 12 81 12 81 12 81 12 81 12 S.E.R..60.62.68.70.58.60.60.60.61.60.66.63.54.57.63.65.53.55.56.55.56.55.62.58.57.60.66.68.56.58.59.58.59.58.65.61.54.57.63.65.53.55.56.55.56.55.62.58.54.56.62.64.51.53.53.53.54.54.60.56 Adj. R 2.92.91.89.88.92.91.91.91.91.91.89.90 Estimation method.93.93.91.90.93.93.92.93.92.93.91.92.92.92.90.89.93.92.91.92.91.92.90.91.93.93.91.90.93.93.92.93.92.93.91.92.93.93.91.90.94.93.93.93.93.93.91.92.53.58.63.65.54.54..55.57.57.56.61.59.94.93.91.90.93.93.92.92.92.92.91.92 SUR SUR SUR SUR SUR IV-SUR NOTES: Standard errors are in parentheses; **, *, *** denote significance at 1, 5, and 10 per cent levels, respectively; ln~distance ii! is defined as 0.56 sqrt~area!. 4. Results Following Wei ~1996! and Helliwell ~1996, 1997!, I estimate the gravity equations for separate years mostly simultaneously as a system by employing the method of seemingly unrelated regression ~SUR!. In particular, while year-specific intercepts are allowed for, the coefficients on the variables are restricted to be the same in all years. The basic results are reported in tables 1 and 2. As expected, the overall empirical fit of the gravity approach is excellent. About 90 per cent of the variation of trade flows is explained by the model. Moreover, all the estimated coefficients on the standard gravity variables have the expected sign and are statistically ~.99 per cent level! and economically significant. The GDP coefficients range from 0.67 to 0.76, indicating that when the GDP of one of the trading partners is higher by 1 per cent, the trade volume increases ~less than proportionately! by about 0.7 per cent. Similarly, a 1 per cent increase in distance decreases trade by about 0.7 to 1.1 per cent.

National borders and international trade 1099 TABLE 2 Home country bias in the European Union, 1983 90 Home 2.16** 2.52** 2.38** 2.59** 2.80** 1.83** ~0.21! ~0.23! ~0.21! ~0.23! ~0.26! ~0.25! ln~distance ij! 1.02** 0.84** 0.91** 0.82** 0.67** 1.16** ~0.08! ~0.09! ~0.08! ~0.09! ~0.12! ~0.10! ln~gdp i! 0.74** 0.74** 0.75** 0.74** 0.73** 0.85** ~0.04! ~0.04! ~0.04! ~0.04! ~0.04! ~0.05! ln~gdp j! 0.76** 0.74** 0.76** 0.75** 0.76** 0.92** ~0.04! ~0.04! ~0.04! ~0.04! ~0.04! ~0.05! Adjacency 0.54** 0.42** 0.51** 0.34*** ~0.15! ~0.16! ~0.17! ~0.19! Language 0.73** 0.48*** 0.41*** 0.40 ~0.24! ~0.25! ~0.25! ~0.28! ln~remote i! 0.35** ~0.12! ln~remote j! 0.02 ~0.12! # Observations 121 8 121 8 121 8 121 8 121 8 121 8 S.E.R..59.60.62.59.58.59.63.62.58.59.61.57.55.57.61.60.57.59.61.58.57.58.63.61.57.58.61.57.55.57.61.60.56.58.60.56.54.56.59.57.72.71.74.70.65.64.65.64 Adj. R 2.92.92.91.92.92.92.90.91.93.92.92.92.93.92.91.91.93.92.92.92.92.92.90.91.93.92.92.93.93.92.91.91.93.92.92.93.93.93.91.92.88.89.88.88.90.90.90.90 Estimation method SUR SUR SUR SUR SUR IV-SUR NOTES: See notes to table 1. If there were nothing to the notion of home country bias ~and average intranational distances, as well as the internal trade volumes, were approximated correctly!, these basic variables would soak up all the explanatory power. There would be nothing left to attribute to a dummy variable representing intranational trade. In column 1 of table 1, however, which presents the result of the simplest gravity model, the dummy variable for intranational trade is 1.92 and statistically highly significant ~.99 per cent level!, indicating that a country exports about 6.8 ~ exp@1.92#! times as much to itself as it exports to a foreign country of similar economic size and distance. This finding suggests that the home country bias in the European Union is substantially lower ~by more than one-half! than the home bias found for Canada by McCallum ~1995! and Helliwell ~1996, 1997!, but is considerably larger than Wei s ~1996! estimates. In columns 2 to 4, language and adjacency dummies are added and found to be statistically significant ~with the correct signs!. It is interesting that when these variables are included into the gravity equation, the home bias increases to 2.43; that is, intranational trade is about eleven times as high as international trade after direct distance, economic sizes, common language, and a common land border are accounted for. The increase of the home bias follows straightforwardly from the

1100 V. Nitsch definition of these dummy variables. In particular, I follow Helliwell s ~1997! method of assigning a value of one to the language and border variables only for trade flows between two countries sharing a common language or border, not for intranational trade flows. 9 In column 5 the results of adding remoteness variables are reported. Contrary to Wei s ~1996! claim, the inclusion of remoteness measures does not affect the basic results. Specifically, the coefficient on the remoteness of the importer is statistically not significantly different from zero, and the coefficient on the exporter s remoteness, which is significant at the 10 per cent level, even has the wrong sign. In sum, with this augmentation, the estimated home bias in the European Union remains unchanged at factor 11. As a robustness check, I have followed the standard procedure of using the log of population size as an instrument for the log of GDP. The results are reported in column 6, which shows that this modification has no effect on the explanatory power of the regression. The estimated home country bias is now about 8.2 ~ exp@2.10#!, and therefore is somewhat lower than it is in a comparable specification using the GDP as size variable. As shown in table 2, the results are little affected by the inclusion of Portugal and Spain. In general, the estimated home bias coefficients are slightly larger than they are in the small sample. For example, in the basic gravity specification ~reported in column 1!, the border effect increases to 8.7 ~ exp@2.16#!. The rise is an indication of a large home bias in the two additional countries, but it also might be in part attributable to the different time period. As Eurostat started reporting production data for Portugal and Spain only three years before their EU accession in 1986, the regressions for the large sample relate to the period from 1983 to 1990. Another interesting point to note is that the inclusion of remoteness variables now increases the estimated home bias in the European Union to factor 16, a result comparable to McCallum s ~1995! and Helliwell s ~1996, 1997, 1998! findings for Canada. Moreover, the coefficient on the remoteness of the exporting country still has the wrong sign, but it is now statistically highly significant at the 1 per cent level. 4.1. Evolution of the home bias The panel also allows analysis of the evolution of the home bias in the European Union. A first visual summary of the basic results is provided in figure 2, where the estimated home bias based on OLS regressions of equation ~3! is plotted for separate years. A sharp decline of the point estimates in the period from 1979 to 1982 is illustrated. Since then, the average home bias in the European Union has dropped only gradually from about factor 9 to factor 7 in the small sample and from factor 12 to factor 10 in the sample that includes Portugal and Spain. 9 Helliwell ~1997, 174! provides an extensive discussion motivating this approach.

National borders and international trade 1101 FIGURE 2 Evolution of the home bias in the European Union NOTES: The home bias is calculated as the anti-log of the estimated coefficient on the home dummy in year-specific OLS regressions of equation ~3!. The small sample comprises Belgium0Luxembourg, Denmark, France, West Germany, Greece, Ireland, Italy, the Netherlands, and the United Kingdom. In the large sample, Portugal and Spain are added. Another approach to examining the trend in the home bias is to take the first difference of the gravity model. Then, all fixed effects are dropped, and the equation to be estimated is d ln~x ij! a ghome b 1 d ln~y i! b 2 d ln~y j! b 3 d ln~r i! b 4 d ln~r j! e ij, ~4! where d denotes first difference and now measures the changes in the home bias. The results are reported in table 3. Although the explanatory power of the regressions is weak, the coefficients on the changes in the home bias are statistically significant. Moreover, all coefficients on the home dummy are negative, confirming the previous finding of a declining home bias in the sample. Columns 1 3 are the results of OLS regressions for changes during separate subperiods of three to four years. Consistent with the visual impression of figure 2, the coefficient on the change in the border effect is declining across time. In columns 5 6, where results for the large sample are reported, the estimates for the trend in the home bias are numerically larger and constant across subperiods. As a

TABLE 3 Evolution of home country bias in the European Union, 1979 90 Direct regression IV regression 1979 83 1983 86 1986 90 1979 90 1983 86 1986 90 1983 90 1979 83 1983 86 1986 90 1979 90 1983 86 1986 90 1983 90 DHome 0.18* 0.15** 0.11* 0.14** 0.21** 0.22** 0.21** 0.18* 0.15** 0.11* 0.14** 0.21** 0.22** 0.21** ~0.08! ~0.05! ~0.05! ~0.03! ~0.05! ~0.05! ~0.04! ~0.08! ~0.05! ~0.05! ~0.03! ~0.06! ~0.08! ~0.05! D ln~gdp i! 0.02 0.22 0.20 0.16 1.31** 0.36*** 0.70** 2.88 5.40* 3.31** 1.11 7.60** 3.15*** 4.32** ~0.19! ~0.28! ~0.33! ~0.14! ~0.23! ~0.19! ~0.14! ~2.21! ~2.26! ~1.23! ~0.86! ~2.68! ~1.79! ~1.40! D ln~gdp j! 0.18 0.72* 1.57** 0.52** 0.40*** 1.92** 1.36** 0.92 2.84 0.75 0.22 0.44 0.02 0.99 ~0.19! ~0.28! ~0.33! ~0.14! ~0.23! ~0.19! ~0.14! ~2.21! ~2.26! ~1.23! ~0.86! ~2.68! ~1.79! ~1.40! D ln~remote i! 0.03 3.95* 3.17* 1.78* 1.70 0.46 0.25 1.10 5.12** 3.70* 2.06** 3.03* 0.01 1.61* ~1.13! ~1.67! ~1.47! ~0.82! ~1.38! ~1.09! ~0.81! ~1.36! ~1.12! ~1.59! ~0.74! ~1.30! ~1.55! ~0.73! D ln~remote j! 0.08 3.46* 2.99* 1.70* 1.86 4.17** 5.10** 0.61 0.14 2.04 0.20 0.58 7.81** 3.71** ~1.13! ~1.67! ~1.47! ~0.82! ~1.38! ~1.09! ~0.81! ~1.36! ~1.12! ~1.59! ~0.74! ~1.30! ~1.55! ~0.73! # Observations 81 81 81 81 3 121 121 121 2 81 81 81 81 3 121 121 121 2 S.E.R. 0.22 0.13 0.13.24.23.14 0.17 0.17 0.18, 0.18 0.22 0.13 0.14.24.14.15 0.18 0.24 0.19, 0.25 Adj. R 2 0.01 0.29 0.27.10.14.16 0.30 0.60 0.15, 0.55 0.03 0.29 0.14.09.18.07 0.14 0.24 0.08, 0.18 Estim. method OLS OLS OLS SUR OLS OLS SUR OLS OLS OLS SUR OLS OLS SUR NOTES: See notes to table 1.

National borders and international trade 1103 TABLE 4 Do richer countries have a smaller home bias? Deviation from unweighted average Deviation from weighted average 1979 90 1983 90 1979 90 1983 90 Home 2.48** 2.87** 2.44** 2.79** ~0.22! ~0.23! ~0.22! ~0.23! Home ln~gdp0pop! 0.74** 1.19** 0.75** 1.21** ~0.23! ~0.23! ~0.24! ~0.23! ln~distance ij! 0.75** 0.57** 0.75** 0.56** ~0.11! ~0.11! ~0.11! ~0.11! ln~gdp i! 0.71** 0.77** 0.71** 0.77** ~0.03! ~0.03! ~0.03! ~0.03! ln~gdp j! 0.74** 0.80** 0.74** 0.80** ~0.03! ~0.03! ~0.03! ~0.03! Adjacency 0.56** 0.49** 0.56** 0.49** ~0.15! ~0.15! ~0.15! ~0.15! Language 0.29 0.43# 0.29 0.43# ~0.20! ~0.22! ~0.20! ~0.22! ln~remote i! 0.28* 0.47** 0.28* 0.47** ~0.13! ~0.11! ~0.12! ~0.11! ln~remote j! 0.04 0.18 0.03 0.18# ~0.12! ~0.11! ~0.12! ~0.11! # Observations 81 12 121 8 81 12 121 8 S.E.R..51.53.60.62.49.51.52.51.53.52.59.54.50.53.55.51.51.52.57.55.51.53.60.62.49.51.52.51.53.52.59.54 Adj. R 2.94.94.92.91.94.94.93.94.93.93.92.93.95.94.93.94.94.94.92.92.94.94.92.91.94.94.93.94.93.93.92.93 Estimation method SUR SUR SUR SUR.50.53.55.51.51.52.57.55.95.94.93.94.94.94.92.93 NOTES: See notes to table 1; ln~gdp0pop! is calculated as the difference, in log form, between a country s GDP per capita and the weighted ~columns 1 2! and unweighted average ~columns 3 4! for the entire sample of countries. robustness check, the regressions are re-estimated in columns 8 14 with output replaced by population as the scale variable. The results on the change of the home bias are unaffected. 4.2. Per capita income and home bias In earlier work by Wei ~1996! and Helliwell ~1997, 1998! it was suggested that richer countries have, on average, a smaller home bias than poorer countries. In table4iexamine whether this effect is also evident in the European Union. In particular, following Helliwell, a variable is constructed that reflects the deviation of a country s GDP per capita from the sample average and therefore has positive ~negative! values for countries that are richer ~poorer! than the average.

1104 V. Nitsch Accordingly, the negative coefficient on that variable, which is statistically significant at the 1 per cent level, reflects the fact that richer countries in the European Union have, indeed, a smaller home bias. The results for the small sample, as shown in columns 1 and 3 of table 4, imply an elasticity of the home bias of about 0.75 in response to increases in average per capita incomes, which is fairly similar to Helliwell s ~1998! OECD results. 5. Concluding remarks This paper adds to the attempt made by Helliwell ~1997! to make sense of the apparently diverging evidence of a home country bias in the goods market. While McCallum ~1995! and Helliwell ~1996! found that trade between two Canadian provinces are about twenty times as large as their trade with U.S. states, after using a gravity model to control for the effects of size and distance, Wei ~1996! argued that in a typical OECD country the home bias is lower by an order of magnitude. In particular, using a more detailed specification of the gravity equation that also adjusts for common language, common border, and geographic location relative to the rest of the world, Wei showed that an average OECD country imports only two and one-half as much from itself as from an otherwise identical foreign country. Besides applying a new set of production data, in this paper I provide two major contributions. First, I suggest a conceptually more sophisticated method to approximate average distances within countries. However, focusing on the land area of a country is just a first crude step in incorporating a country s geography in the determination of the average national distance. Future research should be directed at taking an analytically more complete account of the internal geographic structure of a country. Second, I explicitly analyse the degree of home bias in the European Union. In fact, it is now conventional wisdom that the European Union is one of the most fully integrated regions of the world. Over the past forty years, trade barriers have been removed and trade links have intensified. Therefore, it is a surprising finding that an average EU country still exports about seven to ten times more to itself than to a partner country, after adjustment is made for sizes, distance, common language, common border, and remoteness. In sum, it is suggested in this paper that national borders still matter, even within the European Union. References Cyrus, Teresa L. ~1997! Why do national borders matter? Industry-level evidence, unpublished paper, UC Berkeley Deardorff, Alan V. ~1998! Determinants of bilateral trade: does gravity work in a neoclassical world? in The Regionalization of the World Economy, ed. J. Frankel ~Chicago: University of Chicago Press! Helliwell, John F. ~1996! Do national borders matter for Quebec s trade? Canadian Journal of Economics 29, 507 22

National borders and international trade 1105 ~1997! National borders, trade and migration, Pacific Economic Review 2, 165 85 ~1998! How Much Do National Borders Matter? ~Washington, DC: Brookings Institution! Helliwell, John F., and Genevieve Verdier ~1999! Comparing interprovincial and intraprovincial trade densities, unpublished paper, University of British Columbia McCallum, John ~1995! National borders matter: Canada-U.S. regional trade patterns, American Economic Review 85, 615 23 Wei, Shang-jin ~1996! Intra-national versus international trade: how stubborn are nations in global integration? NBER Working Paper No. 5531 Wolf, Holger C. ~1997! Patterns of intra- and inter-state trade, NBER Working Paper No. 5939