The Economic Signi…cance of National Border E¤ects Carolyn L. Evans1 Federal Reserve Bank of New York2 March 3, 2000 1 Fortheir assistance and support, I wish to thank Richard Caves, Donald Davis, James Harrigan, Elhanan Helpman, Robert Lawrence, and Kei-mu Yi. I also thank James Harrigan, John Helliwell, James Rauch, and Phil Swagel for provision of data and participants at seminars at the Federal Reserve Bank of New York, the University of Chicago Graduate School of Business, and the 1999 EIIT Conference at Purdue University for helpful comments. I acknowledge …nancial assistance from the National Science Foundation, the MacArthur Foundation, and the Weatherhead Center for International A¤airs at Harvard University. Josh Green…eld and Carin Smith provided excellent research assistance. 2 The views expressed in this paper are those of the author, and not necessarily those of the Federal Reserve Bank of New York or the Federal Reserve System. Abstract National borders diminish trade volumes, in many cases by a factor of ten to twenty. Al- though this e¤ect has been widely documented, its economic signi…cance remains unclear. Speci…cally, no previous study has conclusively linked this empirical phenomenon to under- lying, economically meaningful issues, such as the presence of distortionary barriers to trade, the magnitude of any adverse welfare consequences, and the likelihood of a role for policy. In order to address the economic signi…cance of national border e¤ects, this paper pro- vides evidence on two questions which are fundamental to the issue: (1) Do large border e¤ects arise because of high perceived-price wedges between foreign and domestic products, or because imports and domestic goods are very close substitutes?; and (2) If price wedges are important, do they re‡ect distortionary barriers to trade or do they arise from nondis- tortionary factors, such as di¤erences in transactions costs or product characteristics? To address the …rst question, I show that border e¤ects are in fact lower for more dif- ferentiated (less elastic) products, so that the substitutability between domestic and foreign products is a large part of the reason why borders matter so much. Nevertheless, under the assumption of any reasonable elasticity of substitution, implied price wedges between imports and domestic goods are far higher than recognized tari¤ barriers. Regarding the reasons for price wedges, I …nd that goods with higher information costs do not have higher border e¤ects and that information costs are not a primary causal factor of the e¤ects of borders on trade volumes. I also examine the importance of national product characteristics and of actual border barriers to trade. While inherent nationality di¤erences are not an important reason for border e¤ects, borders per se hold substantial explanatory power. My results suggest that border e¤ects may imply distortionary barriers to trade, welfare costs, and a role for policy. I conclude that national border e¤ects have economic signi…cance. 1 Introduction National borders diminish trade volumes, in many cases by a factor of ten to twenty.1 This empirical …nding has proven robust to controlling for a number of alternative explanatory variables, including income, outside trading opportunities, and distance. Further, borders matter for country-pairs as closely linked as Canada and the U.S., as well as for less ap- parently integrated economies (see McCallum (1995) for the original U.S.-Canada result). Although an expanding literature has documented this e¤ect across space and time, the economic signi…cance of these …ndings remains unclear. Speci…cally, no previous study has conclusively linked this empirical phenomenon to underlying, economically meaningful is- sues, such as the presence of distortionary barriers to trade, the magnitude of any adverse welfare consequences, and the likelihood of a role for policy. This lack of evidence on interpretation of the “border e¤ect” rests in large part on its de…nition: the extent to which the volume of domestic trade exceeds the volume of inter- national trade.2 It is thus a comparison of two trade volumes. A number of factors could cause the volume of domestic trade to exceed that of international trade: tari¤s, regulatory di¤erences, information cost di¤erences, the “nationality” of a product, a high elasticity of substitution between imports and domestic goods (together with a small barrier to trade). The volume measure alone does not provide information on the relative importance of the myriad of possible causal factors. Consider, for example, a border e¤ect of 2.65, so that trade between two locations within a country is 2.65 times higher than trade between two locations in di¤erent countries, after controlling for distance, income, and alternative trading opportunities. What could cause such an e¤ect? Consider the following explanations, all of which generate a border e¤ect of 2.65: Scenario #1 The goods produced by the two di¤erent countries are virtually perfect substitutes, with an elasticity of substitution (¾) of 20. Imports are subject to a 5% tari¤. Scenario #2 Imports and domestic goods are imperfect substitutes (¾ = 2). The costs of gathering information about a foreign product are 63% higher than are the costs of doing so for a domestic product. Scenario #3 Imports and domestic goods are imperfect substitutes (¾ = 2). There is a 63% tari¤ on imports. Clearly, the economic implications of these three scenarios di¤er. In the …rst, the magni- tude of the border e¤ect is largely due to the high degree of similarity between imports and domestic goods. It implies neither unrecognized barriers to trade nor high welfare costs. In the second case, the barriers, welfare consequences and policy implications of border e¤ects 1 McCallum (1995) …rst showed that trade between two Canadian provinces was more than twenty times larger than trade between a province and a U.S. state, after controlling for size and distance. This result was particularly shocking given that the U.S. and Canada share a common language, a long border, and low o¢cial barriers to trade. Subsequent work by Helliwell (1996, 1997, 1998), Wei (1996), and Anderson and Smith (1999a, 1999b) has con…rmed and expanded upon such results, both for Canada and for OECD and non-OECD countries. 2 The term “border e¤ect” refers to the downward impact of national boundaries on the volume of trade, i.e. that two di¤erent countries trade far less with each other than do two locations in the same country, after controlling for factors such as income, alternative trading opportunities, and distance. In the context of estimating equation (2), it is the exponential of the coe¢cient on the LOCAL dummy variable, i.e. exp(°). 1 are minimal because the price wedge giving rise to border e¤ects re‡ects a transaction-cost di¤erential. Only in the third case do border e¤ects imply high distortionary barriers to trade, adverse welfare e¤ects, and a potential role for policy.3 This example illustrates that the economic signi…cance of border e¤ects hinges on two crucial issues: 1. Do large border e¤ects arise because of high perceived-price wedges between foreign and domestic products (Scenarios #2 or #3), or because imports and domestic goods are very close substitutes (Scenario #1)? 2. If price wedges are important, do they re‡ect distortionary barriers to trade (Scenario #3) or do they arise from nondistortionary factors, such as di¤erences in transactions costs or product characteristics (Scenario #2)?4 This paper explores the economic signi…cance of border e¤ects by completing four exercises that address these two issues. The exercises in Sections 3.1 and 3.2 directly address the …rst issue listed above: the relative importance of import-domestic price wedges, as opposed to a high elasticity of sub- stitution between the two types of goods. The …rst exercise asks whether di¤erentiated products (i.e., goods with lower elasticities) have higher border e¤ects; this observable rela- tionship between border e¤ects and elasticities allows me to infer the relative importance of unobservable price wedges, as opposed to elasticities. Border e¤ects are, in fact, lower for more di¤erentiated products. This suggests that the substitutability between domestic and foreign products is an important part of the reason why borders matter so much. Never- theless, as the second exercise shows in Section 3.2, under the assumption of any reasonable elasticity of substitution, implied price wedges between imports and domestic goods are far higher than recognized tari¤ barriers. The exercises in Sections 4.1 and 4.2 then address the second issue listed above: the reasons for these price wedges. I …rst examine the role of transaction-cost-based di¤erences between imports and domestic goods, focusing on information costs. Using industry char- acteristics, I …nd that goods with higher information costs do not have higher border e¤ects. I conclude that information costs are not a primary causal factor of the e¤ects of borders on trade volumes. Finally, Section 4.2 examines the importance of national product characteristics and of actual border barriers to imported products. I ask a very basic question: Does the 3 Helliwell (1998) discusses the “correct” magnitude of border e¤ects: “A typical economy with an “ap- propriate” amount of border e¤ect would probably have ... unrestricted rights to travel and trade across national borders. Its equilibrium patterns of travel and trade may continue to show border e¤ects of the sort in evidence among the OECD countries at the end of the 20th century. If so, it would be because the costs of information are such that local networks are more e¤ective than global ones for meeting and responding to local changes in tastes or circumstances.” 4 Throughout this paper, I will use the term “price wedge” or “price di¤erence” in order to refer generally to the di¤erences between the prices of imports and domestic goods, as perceived by the consumer. For only the case of purely distortionary, policy-imposed di¤erentials, however, I will use the term “barrier”. Transaction-cost or product-characteristic based di¤erences, on the other hand, will be incorporated in the broader category of price wedges. This distinction is not only semantic. Barriers may imply welfare costs and policy actions, while transaction-cost or product-characteristic based di¤erences imply neither. 2 border, in and of itself, create a barrier to trade (“location”), or do imports inherently di¤er from domestic products so that consumers purchase them less readily (“nationality”)? To isolate these two factors, I use data on the sales of U.S. multinationals producing in foreign countries, in addition to data on sales of domestic …rms and imports. Comparison of imports from the U.S. and sales of U.S. foreign a¢liates isolates the “location” e¤ect, controlling for “nationality,” while examination of U.S. foreign a¢liate sales versus those of domestic goods highlights “nationality,” controlling for “location”. In fact, I …nd that borders per se rather than an inherent nationality di¤erence hold more explanatory power. Thus, my results suggest that border e¤ects imply distortionary barriers to trade, welfare costs, and a role for policy. I conclude that national border e¤ects have economic signi…cance. 2 Theoretical and Empirical Framework The empirical analysis is based on a gravity-type model of trade. A large number of studies document the empirical explanatory power of this model, which predicts that the aggregate volume of trade between two places will be determined by the income of the two countries and the distance between them.5 The literature on the e¤ects of national borders on trade ‡ows has adopted this empirical model as a natural framework for exploring the relative volumes of internal and external trade.6 For this paper, I modify the gravity framework used in other studies by focusing on industry-level trade, rather than on aggregate ‡ows.7 I derive the estimating equation from a very standard model of trade in di¤erentiated products, in which goods are di¤erentiated by location of production.8 Trade between locations may encounter both transport costs and barriers to trade. Working through such a model, assuming that a good g has a price of 1 in its country of origin, and taking logs yield the following expression for shipments between a producer c` and a consumer c: g log SHIPcc0 = log Yc + log Ycg + log Rg0 + log Rc 0 c g (1) g g ¡¾ g log Dcc0 ¡ ¾ g log Tcc0 g where SHIPcc0 is total purchases of the g good by country c from a producer located in 0 c , Yc is the income of the importer, Ycg is the income of the producer derived from sales 0 g of good g, and Rc0 and Rc are measures of a country’s remoteness, which measure the g GDP-weighted distance of a given country from alternative trading partners.9 ¾ g is the 5 Some common references include Tinbergen (1962), Linneman (1966), and a large number of more recent papers. 6 For example, McCallum (1995), Helliwell (1996, 1997, 1998), Wei (1996), and Anderson and Smith (1999a, 1999b). 7 Helliwell (1998) does include some industry-level analysis. My speci…cation di¤ers from his, however, because I use production levels as explanatory variables, as dictated by the theoretical model described in Appendix B. 8 Details on deriving the estimating equation are provided in the Theory Appendix. For additional discussion, see Helpman and Krugman (1985), Deardor¤ (1995). 9 See appendix for exact functional form. 3 g g elasticity of substitution among varieties of good g, and Dcc0 and Tcc0 represent, respectively, transportation costs (either domestic or international) and other costs of international trade. These other costs of international trade could include a number of factors which would di¤erentiate intra- and inter-national trade, including governmental barriers or di¤erences g 0 in information costs. I assume that Tcc0 = 1 for trade within a country, i.e. if c = c. g This variable (Tcc0 ) will capture di¤erences between international and domestic trade. The border e¤ect is the di¤erence between local and international trade for two locations having g g identical values for all variables in the model other than Tcc0 , i.e. ¾ g log Tcc0 .10 This expression (Equation 1) is in the form of a gravity model of trade and will form the basis for the empirics. The estimating equation will be of the form: g g log SHIPcc0 = ®0 + ¯ 1 log GDPc + ¯ 2 log P RODc0 + ¯ 3 log DISTcc0 (2) +¯ 4 log Rc0 + ¯ 5 log Rc + °LOCAL g The dependent variable (SHIPcc0 ) will be goods consumption by a country c of goods pro- duced by producer c` in industry g. For the case of imports, I use bilateral industry-level import data. For a measure of “domestic trade”, i.e. how much a country consumes of its own goods, I use data on national production (gross output) within an industry less total g exports by that industry. For the case of foreign a¢liates of U.S. multinationals, SHIPcc0 is given by local sales of foreign a¢liates. g The independent variables include production levels in the producing location (P RODc0 ), which are given by data on industrial production (gross output for national production, total sales for foreign a¢liates of U.S. multinationals), income (GDPc ), and the distance between the location of production and the location of consumption (DISTcc0 ). The measure of bilateral distance is the great circle distance, generally from capital to capital, between the two trading countries. For trade within a country, own distances are calculated as 4 of the 1 distance to the nearest trading partner; for islands, own distance was calculated as 2 of the 1 minor radius of the country.11 Another independent variable is what has been termed the “remoteness” variable (Rc0 , Rc ). Introduced to this literature by Wei (1996), it is intended to capture the distance of a 0 given country (c, c ) from alternative trading partners. My empirical proxy for remoteness is based on the speci…cation of Helliwell (1997a): X DIST 0 cc Rc = GDPc0 c0 10 g Note that in levels, rather than logs, it will be exp(¾g log Tcc0 ). 11 This measure of domestic distance was …rst suggested by Wei (1996). Although not ideal in the sense that it may not capture the exact distance over which domestic trade occurs, this measure should depict variation across countries in domestic transport costs. In fact, Helliwell (1997b) notes that for trade within Canada, this proxy produces measures very close to those calculated from data on interprovincial trade distances and some assumed distances for intra-provincial distances. Note that almost all of the results in this paper do not depend on the domestic distance proxy. Changes in the proxy a¤ect only the scaling of the border e¤ect coe¢cient, and most of my results do not depend on the magnitude of border e¤ects. Nevertheless, all sections of this paper have been re-estimated using twice and one-half of the distance values used here. Although the magnitude of the e¤ects of borders does change somewhat, the qualitative results and my conclusions remain the same. 4 0 where the countries c are all of c’s trading partners in the sample. Although not a direct translation of the theoretical de…nition of this term, this form captures the essential elements of the remoteness variable, since distance and GDP enter as, respectively, numerator and denominator. Finally, the dummy variable LOCAL captures the di¤erences in consumption levels de- pending on the source of production. LOCAL takes the value of 1 when the consumer and the producer are in the same location (i.e. country), and zero otherwise. Thus, the magni- g tude of the border e¤ect (exp(¾ g log Tcc0 ) in the theoretical model) will be exp(°), where ° is the coe¢cient on the LOCAL dummy variable. The sources for these data are standard and are described in the Data Appendix, as are all other data used in the paper. The years and countries which I examine vary slightly across sections, depending on data relevance and availability, so Table 1 provides the details on the individual samples. Throughout the paper, I use instrumental variables techniques due to the endogeneity of production and GDP. Industry and year …xed e¤ects and distance- industry interaction terms are included where appropriate. More speci…c information is provided in the individual sections. 3 Price Wedges or Elasticities?12 An issue crucial to the economic signi…cance of border e¤ects is the degree to which high border e¤ects arise from high elasticities of substitution between imports and domestic goods, as opposed to from high price wedges between the two. It is impossible to address this issue by merely observing the magnitude of border e¤ects. This point is illustrated by the g g de…nition of the border e¤ect: exp(¾ g log Tcc0 ). It is the combination of a price wedge (Tcc0 ) and the elasticity of substitution (¾ g ); thus, the magnitude of a given border e¤ect in fact provides complete information on neither of these elements. Two simple partial-equilibrium diagrams further illustrate why elasticities are so impor- tant for interpreting past results. In Figures 1 and 2,13 T VF T ¡T VB is the di¤erence between the trade ‡ow predicted to occur in the absence of barriers to international trade and the ‡ow which actually does occur. This given reduction in trade volume is linked to both an “elastic” and an “inelastic” demand curve. In Figure 1, it is clear that a given reduction in trade volume could imply either a high (¿ i ) or a low (¿ e ) barrier, depending on the elasticity. Clearly, the implications of a given volume …nding are more dire the lower the elasticity of the underlying demand curve. The magnitude of the welfare e¤ect implied by a given vol- ume impact similarly depends on the elasticity of substitution, as shown in Figure 2 (where the welfare e¤ect is measured as the change in consumer surplus). The exercise in Section 3.1 addresses this issue by asking whether less elastic (more di¤erentiated) products have higher border e¤ects. Section 3.2 then analyzes the height of barriers implied by measured industry-level border e¤ects, in combination with a reasonable 12 In the context of the model presented in Section 2 and Appendix B, the term “elasticity” refers to ¾g , which in the model is both the elasticity of substitution between varieties of a particular good and the price elasticity of each individual variety, under common assumptions. See Helpman and Krugman (1985) for an explanation. 13 Note that both …gures do not incorporate the e¤ects of a domestic supply. 5 elasticity of substitution. It also provides calculations of the elasticities required to explain completely the magnitude of border e¤ects, if measured tari¤s are the only barriers to trade. Together, these two exercises will provide an indication of the relative roles of elasticities and of import-domestic price wedges. As I will show, both factors play some role, but the magnitude of unexplained price wedges appears far larger than measured barriers to trade. 3.1 The Relation Between Elasticities and the Magnitude of Bor- der E¤ects 3.1.1 Introduction14 In order to learn about the relative importance of elasticities and price wedges, it would be useful to know how elasticities are related to price wedges. More speci…cally, I would like to know whether a given aggregate border e¤ect is due to high price di¤erences on inelastic products or to lower di¤erences on very elastic products, or to some combination thereof. Unfortunately, this relation is not directly observable. Instead, I analyze the relationship between border e¤ects and elasticities, which is ob- servable. The border e¤ect-elasticity relation is informative because border e¤ects increase with a fall in the elasticity only when high price wedges on inelastic products are the primary cause of border e¤ects. Thus, if the data show that border e¤ects do not increase with a fall in elasticity, high price wedges on inelastic products are probably not important. To explain the method, …rst note that certain price wedges between domestic and inter- national trade may vary systematically with the degree of product di¤erentiation or elasticity of substitution among varieties, either increasing or decreasing with a fall in elasticity (see Table 2). As an example of the former case, some have argued that informational costs should be higher for di¤erentiated products.15 On the other hand, import-domestic wedges may not vary systematically with di¤erentiation, such as with tari¤s and national legal institutions. These three possible correlations are listed in the second column of Table 2. The relationship between border e¤ects and elasticities will allow me to distinguish among these three cases, as shown in the third column of Table 2. If price wedges which increase systematically with lower elasticities are important, such as may be the case with informa- tion costs, then more di¤erentiation (a lower elasticity) should be associated with higher border e¤ects (Case 1). On the other hand, if price wedges are fairly constant across goods, any small price di¤erence would have a much lower e¤ect for products that are more highly di¤erentiated (Case 2). Finally, if wedges are higher for more homogeneous (higher elas- ticity) products, more homogeneous products should have higher border e¤ects (Case 3). Thus, only if higher price wedges on di¤erentiated products have an e¤ect outweighing any elasticity e¤ect would higher border e¤ects on more di¤erentiated products be expected. In the contrary, if I do not observe this pattern, I should be able to reject that possibility.16 14 Evans (1999a) contains more extensive analysis of these issues. 15 For example, Rauch (1999). 16 Of course, the possibility exists that barriers increase slightly with a fall in elasticity, but that these e¤ects are masked by the e¤ects of barriers on the more elastic products. This case, however, would also indicate that barriers which increase with a decrease in the elasticity are not the major causal factor behind observed aggregate border e¤ects and would not alter our …nal interpretation of the empirical result. A 6 Thus, exploring the relationship between border e¤ects and elasticities is the primary goal of this section. This observable border e¤ect-elasticity relation will then be used to infer the unobservable price wedge-elasticity relation. 3.1.2 Results The data for this section are for eight OECD countries in 1990, across 12 industries. Several measures are used to proxy for di¤erences across industries in elasticities/degree of di¤erenti- ation among products. First, I consider intra-industry trade as a proportion of the total trade within an industry.17 Although the link between product di¤erentiation and the proportion of intra-industry trade does not follow directly from the standard theory of trade in di¤er- entiated products, several empirical studies have found the two to be positively associated with each other.18 Second, I utilize the measure of homogeneity of products proposed by Rauch (1999). I calculated for each of the industries the percent of trade in products which he characterizes as not having either an organized exchange or a reference price, i.e. the products likely to have lower elasticities. Thus, a higher “Rauch index” indicates a less homogeneous (more di¤erentiated) product. Third, I consider R&D spending as a share of sales, which may measure di¤erentiation and the importance of information. Finally, I also consider the ratio of advertising to sales within an industry. Within the industrial organization literature, the ratio of advertising expenditure to total sales is a standard approach to capturing the role of product di¤eren- tiation.19 Some of these individual measures may be open to question. Taken as a group, however, if I …nd consistent results across all measures, then it would be di¢cult to dispute the implied role of price wedges which fall systematically with a fall in product di¤erentiation, as opposed to elasticities. I use the estimating framework of equation 2, with the following potential trading rela- tionships: (1) purchase of domestic goods; (2) bilateral imports from a sample of partner countries. The LOCAL variable is de…ned as one for purchases of domestic goods and zero for bilateral imports. To gauge the e¤ect of the elasticity proxies on the impact of national borders, I include an interaction term, which is the LOCAL variable multiplied by the measure of interest. All measures vary between 0 and 1, with a higher index associated with a higher degree of di¤erentiation (a lower elasticity). Thus, for example, for an industry with 50% intra-industry trade, trade between two unrelated countries in the sample will have a border e¤ect of exp(¯ LOCAL + (¯ LOCAL¤IIT ¤ :5)), where ¯ LOCAL is the coe¢cient on the LOCAL similar conclusion would apply to the case in which barriers di¤er across products, but in no systematic relationship to elasticities. 17 The proportion of intra-industry trade is calculated using the Grubel-Lloyd index as described in Hum- mels and Levinsohn (1993). 18 For example, see Caves (1981), Loertscher and Wolter (1980), and Hansson (1991). Davis (1997) uses this index as a measure of product di¤erentiation. Hansson (1991) also provides a theoretical model in which this relationship holds. 19 See Scherer and Ross (1990), pp. 436-438. 7 dummy variable, and ¯ LOCAL¤IIT is the coe¢cient on the interaction term. A positive coe¢cient on the interaction term indicates that more di¤erentiated (less elastic) products have higher border e¤ects. As shown in Table 3, the results for all measures indicate that a higher degree of product di¤erentiation is actually associated with a lower border e¤ect.20 For example, the coe¢cient of -1.52 on the Rauch variable indicates that a perfectly homogeneous product (R=0) will have a border e¤ect of exp(2:27) = 9:68, while a product with R=.5 will have a border e¤ect of exp(2:27 ¡ (:5 ¤ 1:52)) = 4:53. The proportion of intra-industry trade, the Rauch index, and the R&D to sales ratio are signi…cant at the 0.5%, 0.3%, and 0.9% levels, respectively. The advertising to sales ratio is signi…cant at the 6.5% level. Thus, all variables yield consistent results suggesting that border e¤ects fall with a higher degree of product di¤erentiation. This suggests that high border e¤ects are partially at- tributable to the elasticity of substitution between domestic goods and imports, and that high border e¤ects do not necessarily indicate large price wedges between domestic and for- eign products. It also implies that price discrepancies that increase with a fall in elasticity are not likely to be an important explanation of aggregate border e¤ects. Thus, product- speci…c information, in the sense described by Rauch (1999), probably does not play a major role. If such wedges do exist, their e¤ects are outweighed by elasticity e¤ects. (I will return to this point in section 4.1.) 3.2 The Magnitude of Implied Barriers to Trade and Elasticities 3.2.1 Introduction The preceding results provide some support for the notion that border e¤ects may be readily explained simply by elasticities in combination with widely-acknowledged existing barriers to trade. However, the analysis leaves unanswered the question of the magnitude of the price wedges required to explain completely the magnitude of observed border e¤ects. In other words, I have shown that elasticities are a part of the explanation, but I have not yet provided an indication as to what extent elasticities alone are able to explain border e¤ects. This section provides estimation and calculations to answer this question. It shows the extent to which observed border e¤ects may be entirely explained by reasonable elasticities of substitution, together with acknowledged barriers to trade, i.e. tari¤s. In order to gauge to what degree elasticities and tari¤s do indeed explain border e¤ects, I …rst estimate industry border e¤ects. I then calculate implied barriers to trade across the industries in the sample, based on the measured border e¤ect and the assumption of a “reasonable” elasticity of substitution. Comparing these overall implied wedges to actual tari¤ data yields an indication of the “unexplained” price wedge.21 I also calculate the 20 The estimation technique in this section is two-stage-least-squares with population and population-based remoteness measures as instruments for GDP and GDP-based remoteness measures. Following Harrigan (1995, 1996), several endowment measures are used as instruments for production levels. Measures of the log of the number of workers, the log of the capital stock, and the log of agricultural land are interacted with industry dummy variables to create a set of 12 ¤ 3 = 36 instruments for production. Industry-speci…c regressions of the log of production on the instruments yields R2 s ranging between 0.91 and 0.99. Industry …xed e¤ects and distance-industry interaction terms are included. 21 ¯g Based on the de…nition of the border e¤ect, the implied barrriers may be calculated as Tcc0 = exp[ ¾H ]¡1. m g 8 elasticities required to explain the magnitude of border e¤ects, if the only barriers to trade are measured tari¤s. 3.2.2 Results Table 4 provides the results for implied barriers to trade. In the second column is the border e¤ect for non intra-E.U. trade for two non-adjacent countries for each industry in the sample, calculated as exp(¯ g ), where ¯ g is the coe¢cient on an industry–dummy-variable-LOCAL L L interaction term.22 The third column provides the average tari¤ for these countries in each industry. The subsequent columns provide the calculated ad valorem tari¤ equivalent price wedges for a range of elasticities, in order to provide a rough order of magnitude of implied and unexplained price wedges. Thus, if all goods have an elasticity of 2, total implied price wedges range between 96.66% (Other Manufacturing) and 553.99% (Petroleum Re…neries and Products). To approximate a slightly more realistic case with variation in elasticity across industries, I calculate a range of elasticities based on the Rauch index. The derived elasticities range between 2 and 5. Total implied price di¤erences range between 67.22% (Basic Metal Industries) and 227.67% (Non-metallic Mineral Products).23 Rather than focusing on speci…c magnitudes, I am interested in providing general ranges to compare to the actual tari¤ data in the third column. Clearly, these implied price di¤erences are far larger. For example, while the average tari¤ in Wood Products and Furniture is 4.65%, the implied wedges range between 76.36% and 313.06%. More generally, with the elasticity ranging between 2 and 5, unexplained price wedges range between 62.33% and 221.12%. In other words, even after eliminating the e¤ects of tari¤s, the import-domestic price wedge may be as high as 221.12%.24 An alternative calculation is the elasticity implied by measured border e¤ects and tari¤s.25 This is the elasticity which would be required to explain entirely the magnitude of border e¤ects, if the only barriers to international trade were measured tari¤s. As shown in Table 5, this exercise yields elasticities which range between 24 and 76, with a simple average of 43. I subtract the tari¤ levels from the overall implied price wedge. See Lee and Swagel (1997) for a description of the tari¤ data. 22 Full results for the regression are in Table A1. Dummy variables for trade between two European Union countries and for trade between two adjacent countries are included in order to calculate border e¤ects for non-intra-EU trade for two non-adjacent countries. 23 See data appendix for a description of calculation of these elasticities. The issue of an appropriate value for the elasticity of substitution between imports and domestic goods is a contentious one. Several older studies (such as Stern et al. (1976)) suggest that the aggregate number is between 1 and 2. At a disaggregated level, others have found that it ranges between 1 and 12, for imports from all countries to the U.S. (Shiells et al. (1986)). Feenstra (1994) suggests that these numbers may be too low for a number of methodological reasons. He …nds disaggregated values that range between 2.96 (typewriters) and 42.9 (silver bullion). Hummels (1999) …nds, for the U.S., New Zealand, Argentina, Brazil, Chile, and Paraguay, elaticities at the 2-digit SITC level largely between 3 and 8. Obviously, any single number conclusion on the magnitudes of implied di¤erentials would rest on more precise estimation of these elasticities. In any case, my results may be interpreted as an indication of the order of magnitude of these di¤erentials. 24 One possible concern is that tari¤s may be related to non-tari¤ barriers in such a way that industries with high tari¤s tend to have high protection in general, leading to higher border e¤ects than would be indicated by the tari¤ data alone. In fact, this does not appear to be the case. 25 ¯g The implied elasticity is calculated as ¾ g = ln(T m +1) . H 0 cc 9 As detailed in footnote 23, these numbers far exceed most estimated elasticities, providing additional evidence that border e¤ects result from price wedges which exceed the magnitude of tari¤s. Thus, although Section 3.1 showed that elasticities are indeed part of the story, perceived- price wedges not explained by tari¤s must still be non-negligible. In the next sections, I consider possible reasons for these price di¤erences. 4 The Reasons for Price Wedges As illustrated above, implied price wedges between foreign and domestic products are far larger than would be suggested by recognized barriers to trade. This …nding leads to the second crucial issue listed in the introduction: if price wedges are important, do they re‡ect distortionary barriers to trade or are they due to nondistortionary factors, such as di¤erences in transaction costs or product characteristics? This section completes two exercises which will explore the importance of several possible causal factors. Section 4.1 focuses on information costs as a transaction-cost based explanation. Section 4.2 examines the roles of both nationality as a product characteristic and of borders per se as a barrier to trade. 4.1 Information Costs Within and Across Borders 4.1.1 Introduction Di¤erences in the costs of obtaining information about the quality or existence of a foreign versus a domestic product could partially explain border e¤ects.26 If it is indeed more costly to learn about a foreign product, this higher transaction cost could tend to reduce the quantities of foreign goods purchased. The relevance of information-cost di¤erentials is important for drawing conclusions about the welfare consequences of border e¤ects. If information costs are important, border e¤ects do not necessarily arise from distortionary barriers, and may instead simply re‡ect di¤erent transaction costs. The ease or di¢culty of information transfer may be a¤ected by industry-speci…c char- acteristics. For example, information transfer should be more di¢cult for complex products composed of a bundle of attributes than for simple, standardized goods. Other industry characteristics which should be associated with an important role for information transfer in- clude the importance of information in purchasing and consuming a product and the degree of di¤erentiation among varieties of a product. For products possessing these characteristics, domestic varieties may be purchased more readily than foreign ones if international infor- mation transfer is indeed more costly than domestic. Thus, if information cost di¤erences between foreign and domestic products are a primary explanation for border e¤ects, borders should matter more for industries where information transfer is more costly, di¢cult, and important.27 26 Helliwell (1997, 1998), Rauch (1999). Also see the Journal of International Economics (1999) for a number of papers on the related topic of the role of networks in international trade. 27 This argument makes an implicit assumption about the link between industry characteristics and the foreign-domestic information cost di¤erential. More speci…cally, I assume that (1) there is some di¤erence 10 To capture variation in the need for information across industries, I …rst use a set of indicators on the importance and di¢culty of information transfer, which comes from the results of a Conference Board survey of marketing activities (Bailey (1975)). As a …rst indicator, I use data on the frequency of sales or technical service. High frequency should indicate that information about the use of a product must be transferred subsequent to the original purchase. Second, a variable indicating whether a good is made to order, as opposed to being …lled from stock, should indicate a need for information transfer regarding the buyer’s particular needs and preferences. Third, I consider whether a particular product tends to be a major purchase for a buyer; as Bailey (1975) notes, major purchases often entail more negotiation and information-gathering. Finally, a variable describing infrequency of purchase should also provide some indication of the absence of experiential knowledge and thus the need for transfer of information. All of these indicators should be positively associated with the di¢culty and/or importance of information transfer, so that border e¤ects should be higher, for example, in industries where products are made to order and require frequent technical service if information costs are the major explanation of border e¤ects. I also ask how the degree of di¤erentiation a¤ects border e¤ects, using the results of the previous section. If information costs are indeed important, border e¤ects should be higher for more di¤erentiated products. 4.1.2 Results I continue to utilize the estimating framework of equation 2, with the following potential trading relationships: (1) purchase of domestic goods; (2) bilateral imports from a sample of countries. The data are for 8 OECD countries in 1990, across 12 industries. The LOCAL dummy variable is 1 for purchases of domestic goods and zero for bilateral imports. The results for the Conference Board Data are in Table 6. To examine the e¤ects of these industry characteristics on the border e¤ect, I incorporate a variable composed of the interaction between the variable of interest and the HOM E dummy variable. This coe¢cient describes how variation in the variable a¤ects the di¤erence between national and international trade. Thus, a negative sign suggests that an increase in that variable tends to reduce the impact of borders. Due to data availability, I am forced to limit analysis to …ve of the twelve industries, so column (i) also provides a benchmark for this smaller sample. Column (ii) shows that border e¤ects are in fact lower where products tend to be made to order, as opposed to …lled from stock. They also tend to be lower where sales/technical service is more frequent (column (iii)). When purchases within an industry tend to be major between the information costs for imports and domestic goods, per unit of information; and (2) that infor- mation costs are higher in certain industries. Thus, in industries where information costs are higher, the di¤erence in the per unit cost of information will lead to a higher overall di¤erence between foreign and domestic information costs. For example, a very complex product requires me to learn about its many attributes, for each of which there is a foreign-domestic di¤erence in the costs of information. Thus, if information costs are an important explanation for border e¤ects, border e¤ects should be higher in these high-information-cost industries. 11 purchases, border e¤ects are also lower (column (iv)). The coe¢cient on the frequency of purchase is not signi…cant. Three of the four indicators of the importance of information transfer thus suggest that border e¤ects are lower where information transfer is likely to be more di¢cult and/or important. The results for di¤erentiation and information costs are indicated in Table 3, as discussed in Section 3.1: border e¤ects are in fact lower for more di¤erentiated products. Thus, product-speci…c information does not play a major role in the creation of border e¤ects. The e¤ects of such di¤erentials are outweighed by elasticity e¤ects. Thus, the evidence on both product di¤erentiation and industry characteristics suggests that product-speci…c information-transfer di¢culties are not likely to be a major cause of border e¤ects. 4.2 Nationality and Location To continue the analysis of possible explanations of import-domestic price wedges, this sec- tion asks how much of the border e¤ect is explained by borders per se and by nationality di¤erences.28 4.2.1 Introduction In the most general sense, when domestic goods are purchased more readily than imports, two types of di¤erences between the two could exist: (1) Barriers at the border, i.e. tari¤s, non-tari¤ barriers, regulatory di¤erences, etc. (“location”). (2) An inherent di¤erence between domestic and foreign products (“nationality”). g In terms of equation 1, the Tcc0 term may be composed of two elements, “border barriers” g g (¿ cc` ) and di¤erentials due to di¤erences between domestic and foreign products (Ccc` ), such g that Tcc0 = ¿ g ` ¤ Ccc` .29 This distinction is important for reasons related to both the welfare cc g implications and the policy relevance of border e¤ects. If borders per se in fact create border e¤ects, distortionary barriers, governments, and policy may play an important role. On the other hand, if domestic and foreign products are indeed “di¤erent,” border e¤ects could re‡ect simply an optimal response to these perceived di¤erences. Welfare consequences of border e¤ects should not be large, and the role of policy in the creation and alleviation of border e¤ects may be minimal. To address this issue, I utilize data on sales of U.S. multinationals producing in foreign countries, in addition to data on sales of domestic …rms and imports.30 Using this infor- 28 More extensive analysis of this issue is contained in Evans (1999b). 29 An alternative interpretation may be derived from a model (such as the one in Appendix B) in which ¯ g di¤ers across trading partners. In this interpretation, di¤erences between domestic and foreign products c` arise from di¤erences in this parameter of the utility function. The way that the di¤erential enters into the equation is actually quite similar for the two cases. For an extensive discussion of this alternative approach, see Hummels (1999). 30 There is one potential problem with the data used for this approach. The data on domestic production include activities for all multinationals producing within that country, so that the measure of domestic trade is not as “pure” as one might desire. In a separate Data Appendix, available from the author upon request, I provide analysis which addresses this issue by partially adjusting the data for the presence of multinationals. 12 mation allows me to isolate the two broad potential sources of di¤erences between imports and domestic goods. Comparison of imports from the U.S. and sales of U.S. foreign af- …liates isolates the “location” e¤ect, controlling for “nationality.” In terms of the model, this examination of consumption levels for products produced by foreign a¢liates versus levels for imports will measure the ¿ g ` element, while eliminating any di¤erences due to the cc g Ccc` element. Thus, if nationality e¤ects alone explain border e¤ects, borders should not matter at all when comparing consumption levels from these two sources. On the other hand, examination of U.S. foreign a¢liate sales versus those of domestic goods highlights “nationality,” controlling for “location.” In this case, if borders alone explain border e¤ects, and nationality plays no role whatsoever, borders should matter to the same degree when comparing imports from the U.S. to either sales of foreign a¢liates or to domestic sales. 4.2.2 Results The form of the test will again be a standard gravity-type model, as in equation 2, with three potential trading relationships: (1) a country purchasing goods from its own domestic …rms; (2) a country consuming the goods of U.S. foreign a¢liates producing within its borders; and (3) a country importing goods from the U.S. The variable LOCAL is a dummy variable which takes on the value of 1 when the producer and consumer are located in the same country, and 0 otherwise. Using imports as the benchmark, I examine the e¤ects of consuming a domestic product and of consuming a good produced by a foreign a¢liate located within national borders, with the variable LOCAL measuring these e¤ects for both cases. Thus, the test consists of running two separate equations simultaneously in a three- stage-least-squares framework. The …rst equation contains data on imports from the U.S. and domestic sales; it measures the extent to which the volume of domestic trade exceeds international trade. The second contains data on imports from the U.S. and on the sales of the foreign a¢liates of U.S. multinationals producing within the foreign country. It measures the degree to which foreign a¢liates sales volume exceeds (or falls short of) the volume of imports from the U.S. This second equation thus measures the “location” e¤ect. I then compare the results of the two di¤erent equations to derive the “nationality” e¤ect. Table 7 presents the results for these tests, for nine OECD countries in seven manu- facturing industries between 1989 and 1994.31 Column (i) shows that the coe¢cients for nearly all of the variables are quite similar whether I compare imports to domestic sales or to foreign a¢liate sales. In particular, the coe¢cient on the LOCAL variable is 2.38 for the domestic sales case and 2.35 for the foreign a¢liate sales case. As shown by the p-value, I Such adjustment does not substantially a¤ect the results. 31 In this section, I use a three-stage-least-squares estimation technique, with population and population- based remoteness measures as instruments for GDP and GDP-based remoteness measures. Following Har- rigan (1995, 1996), several endowment measures are used as instruments for production levels. Measures of the log of the number of workers, the log of the capital stock, and the log of agricultural land are interacted with industry dummy variables to create a set of 7 ¤ 3 = 21 instruments for production. Industry-speci…c regressions of the log of production on the instruments yields R2 s ranging between 0.82 and 0.98. Industry and year …xed e¤ects and distance-industry interaction terms are included. 13 am unable to reject the hypothesis that the two are equal. Thus, the e¤ects of borders are very similar whether I compare imports to domestic producers or to a¢liates of U.S. multi- nationals located in the country. In the table, I also provide a “border e¤ect”. This is the exponential of the coe¢cient on the LOCAL dummy variable, and indicates by how many times domestic or foreign a¢liate sales would exceed imports. Based on this estimation, we would expect foreign a¢liates located overseas to sell about 10 times as much as would be imported, after controlling for the other variables in the model.32 Column (ii) provides the breakdown by industry. Again, the coe¢cients for the LOCAL dummy variable are quite similar across all industries, and I am unable to reject the hypothesis that they are equal for six of the seven industries. For the industry where foreign a¢liate and domestic e¤ects do indeed di¤er (Chemicals and Allied Products), the e¤ects of borders per se remain larger than any di¤erence between foreign a¢liate and domestic e¤ects. The OLI theory of foreign direct investment presents one potential criticism of this anal- ysis. It posits that multinationals develop a¢liates overseas in order to exploit some pro- prietary asset of the parent company,33 which would suggest that foreign a¢liates may be fundamentally di¤erent from the other sources of production. If foreign a¢liates do di¤er systematically from U.S. exporters through possession of some proprietary asset, then my analysis picks up the e¤ects of such characteristics, in addition to those of borders per se. To explore this possibility, I utilize data on the level of proprietary assets for both U.S. exporters and for foreign a¢liates. Possession of a proprietary asset may a¤ect the price perceived by consumers, either because of di¤erent costs of production or because of some ability to overcome barriers inherent to foreign products. To account directly for these e¤ects, I …rst incorporate the level of a proprietary asset as a regressor, using a functional form of ln(1 + P A),34 where P A is the proxy for the level of proprietary assets, thus allowing me to ask about the e¤ects of borders, holding constant this factor which could a¤ect perceived prices. Second, I also indirectly examine the price e¤ects by measuring the impact of changes in the level of proprietary assets on border e¤ects. If foreign a¢liates indeed leverage proprietary assets in order to create advantages over exporters, we would expect that border e¤ects are higher in the presence of proprietary assets. To test this hypothesis, I incorporate a term which is an interaction between the proprietary-asset proxy and the LOCAL dummy variable. This coe¢cient represents the e¤ects of changes in the level of the proprietary asset on the border e¤ect. Table 7 (Cont.) provides the results when controlling for proprietary assets as proxied by the R&D to sales ratio for foreign a¢liates and for U.S. exporters (columns (iii) and (iv)). The ratio for the foreign a¢liates captures the R&D performed for foreign a¢liates, by both 32 Note that, as illustrated in equation 1 and discussed in Section 3, the magnitude of the border e¤ect arises from the combination of a barriers and an elasticity, so that a given border e¤ect cannot be linked to a given barrier without an assumption about the elasticity of substitution. 33 “OLI”: Ownership, Location, Internalization. For discussion of this theory, see Dunning (1988, 1993). 34 The given functional form follows from a price speci…ed as (for example) pm = pm ¤ ¿ m ¤ (½m )¡1 , sr rr sr rr where pm represents the price at the factory, ¿ m are barriers to trade, and ½m represents the e¤ects of the rr sr rr proprietary asset. Since a zero level of proprietary assets should leave the perceived price unchanged, I use the functional form of 1 + P:A: Pr oxy . The logarithmic form follows from the transformation of the import Sales demand equation into logs. 14 the foreign a¢liate itself and by other entities. The ratio for exporters is represented by data for the U.S. as a whole. Controlling for this factor has little e¤ect on the results, as illustrated by a comparison between Columns (i) and (iii).35 With the interaction term included (column (iv)), I am comparing the border e¤ect for imports and domestic goods to that for imports and foreign-a¢liate goods where there is a zero level of proprietary assets present. By doing so, I eliminate any portion of the foreign-a¢liate location e¤ect due to the possession of proprietary assets. Again, the results do not change substantially. Thus, for the most part, the “border e¤ect” for foreign a¢liates producing within a host country di¤ers little from that for domestic …rms. This result suggests that borders per se, rather than a nationality di¤erence, are a more important part of the reason why borders matter. 5 Summary and Conclusions The fact that national borders sharply reduce trade ‡ows has received a great deal of at- tention, in large part because recent empirical …ndings could imply large barriers to trade, adverse welfare consequences, and a role for policy. As I discussed at the outset, however, no previous work has linked this empirical phenomenon to underlying, economically meaningful issues. To address the economic signi…cance of national border e¤ects, this paper provides evidence on two fundamental questions: 1. Do large border e¤ects arise because of high perceived-price wedges between foreign and domestic products, or because imports and domestic goods are very close substitutes? 2. If price wedges are important, do they re‡ect distortionary barriers to trade or do they arise from nondistortionary factors, such as di¤erences in transactions costs or product characteristics? To address the …rst question, I showed that border e¤ects are in fact lower for more dif- ferentiated (less elastic) products. This suggests that the substitutability between domestic and foreign products is a large part of the reason why borders matter so much. Neverthe- less, under the assumption of any reasonable elasticity of substitution, implied price wedges between imports and domestic goods are far higher than recognized tari¤ barriers. Regarding the reasons for price wedges, I found that goods with higher information costs do not have higher border e¤ects and that information costs are not a primary causal factor of the e¤ects of borders on trade volumes. I also examined the importance of national product characteristics and of actual border barriers to trade. While inherent nationality di¤erences are not an important reason for border e¤ects, borders per se hold substantial explanatory power. This …nding, in particular, supports the proposition that distortionary 35 I have performed similar analysis using data on the royalties and license fees paid by foreign a¢liates to their parent companies. These payments are an alternative indicator of the level of proprietary assets leveraged by a foreign a¢liate. This link has also been highlighted by Caves and More (1994). I have also used advertising to sales in order to account for the possible advantages of a local presence in producing heavily branded products. The results do not change and are available from the author upon request. 15 barriers, governments, and policy play an important role in the existence and magnitude of border e¤ects. As an additional indicator of the economic signi…cance of national border e¤ects, Table 8 provides some very rough, back-of-the-envelope estimates of the welfare impact of national border e¤ects.36 The numbers in the table provide the percentage change in indirect utility caused by the elimination of the barriers implied by national border e¤ects, for elasticities of substitution equal to 2, 5, or 10. I provide estimates for reducing barriers to zero, as well as for reducing barriers to the level of average tari¤s for each country. The percentage change in indirect utility ranges between 8.5% and 17.6%, if the elasticity of substitution between imports and domestic goods is 5 and barriers are completely eliminated. If barriers instead drop to the level of average tari¤s, gains in indirect utility range between 6.1% and 16.2%. Together with the results detailed elsewhere in this paper, these rough estimates support the notion that border e¤ects are indeed economically signi…cant. My conclusions must be tempered by several caveats. First, some of the variables I use are at best proxies for the concepts I aim to capture and, thus, present certain limitations. Second, the actual magnitude of barriers to trade and welfare e¤ects will ultimately be determined only after consensus is reached on an appropriate value for the elasticity of substitution. Finally, although I have provided evidence on some of the reasons that would downplay the economic signi…cance of border e¤ects, other similar explanations could also be important and deserve further investigation. Finally, recall the three scenarios presented at the outset. Scenario #1, in which high border e¤ects arise almost entirely from high elasticities of substitution, provides at best a partial explanation of this phenomenon. In Scenario #2, primarily transaction-cost based di¤erences between foreign and domestic products create border e¤ects; the data do not support this proposition. Thus, Scenario #3 remains; in this case, borders matter because of distortionary barriers at the border, and border e¤ects indeed imply potentially high barriers, welfare costs, and a role for policy. The …ndings indicate that this case is an important part of the explanation of border e¤ects. Further, my rough calculations suggest that the actual welfare impact of national border e¤ects may not be small. I conclude that national border e¤ects have economic signi…cance. References [1] Anderson, Michael A. and Smith, Stephen S. (1999a) “Canadian Provinces in World Trade: Engagement and Detachment.” Canadian Journal of Economics 32: 22-38. 36 These calculations are made based on the expression for indirect utility in the theoretical model presented in Appendix B. They are based on country-level border e¤ects for an aggregate national good, rather than on industry-level border e¤ects. I hold constant income and the nontraded sector of the economy. The percentage change in indirect utility is the proportional change implied by a lowering of barriers from the level calculated from border e¤ects to either 0 or to actual tari¤ barriers. Data sources and assumptions for the calculations are provided in Appendix A. 16 [2] Anderson, Michael A. and Smith, Stephen S. (1999b) “Do National Borders Really Mat- ter? Canada-U.S. Regional Trade Reconsidered.” Review of International Economics 7(2): 219-227. [3] Bailey, Earl L. (1975) “Marketing Cost Ratios of U.S. Manufacturers: A Technical Anal- ysis.” Conference Board Report No. 662. New York: Conference Board. [4] Bureau Europeen des Unions de Consommateurs (1989) “Car Prices and Progress To- wards 1992.” Brussels, October 15. [5] Caves, Richard E. (1981) “Intra-industry Trade and Market Structure in the Industri- alized Countries.” Oxford Economic Papers 33 (July): 203-223. [6] Caves, Richard E. and Bradburd, Ralph M. (1988) “The Empirical Determinants of Vertical Integration.” Journal of Economic Behavior and Organization. 9: 265-280. [7] Caves, Richard E. and More, Anand (1994) “Intra…rm Royalties in the Process of Ex- pansion of U.S. Multinational Enterprises.” in V.N. Balasubramanyam and D. Sapsford, eds. The Economics of International Investment. Aldershot: Edward Elgar. [8] COMPUSTAT Database (Various years). [9] Davis, Donald. (1997) “The Home Market, Trade, and Industrial Structure.” National Bureau of Economic Research Working Paper #6076. [10] Deardor¤, Alan (1995) “Determinants of Bilateral Trade: Does Gravity Work in a Neo- classical World.” National Bureau of Economic Research Working Paper #5377. [11] Dunning, John H. (1988) Explaining International Production. London: Unwin Hyman.. [12] Dunning, John H. (1993) The Globalization of Business: The Challenge of the 1990s. New York: Routledge. [13] Engel, C., and J.H. Rogers (1996) “How Wide is the Border?” American Economic Review 86: 1112-1125. [14] Evans,-Carolyn-L. (1999a) National Borders and International Trade. Harvard Univer- sity, Ph.D. Dissertation. [15] Evans, Carolyn L. (1999b) “The Sources of Border E¤ects: Nationality or Location?” Working paper. [16] Feenstra, Robert C. (1994) “New Product Varieties and the Measurement of Interna- tional Prices.” American Economic Review 84(1): 157-177. [17] Feenstra, R., (1997) “U.S. Exports, 1972-1994: With State Exports and Other U.S. Data.” National Bureau of Economic Research Working Paper #5990. [18] Feenstra, R., Lipsey, R., and Bowen, H. (1997) “World Trade Flows, 1970-1992, with Production and Tari¤ Data.” National Bureau of Economic Research Working Paper #5910. 17 [19] Fitzpatrick (1986) Direct Line Distance. Metuchen, New Jersey: Scarecrow Press. [20] Flam, Harry (1992) “Product Markets and 1992: Full Integration, Large Gains?” The Journal of Economic Perspectives. 6(4): 7-30. [21] Grubel, H and Lloyd, P. (1975) Intra-Industry Trade: The Theory and Measurement of International Trade in Di¤erentiated Products. London: MacMillan. [22] Hansson, Par. (1991) “Determinants of Intra-Industry Specialization in Swedish Foreign Trade.” Scandinavian Journal of Economics 93(3), 391-405. [23] Harrigan, James (1995) “Factor Endowments and the International Location of Pro- duction: Econometric Evidence for the OECD, 1970-1985.” Journal of International Economics 39: 123-141. [24] Harrigan, James (1996) “Openness to Trade in Manufactures in the OECD.” Journal of International Economics 40: 23-39. [25] Helliwell, John F. (1996) “Convergence and Migration Among Canadian Provinces.” Canadian Journal of Economics 29 (Proceedings Issue: April), S324-S330. [26] Helliwell, John F. (1997) “National Borders, Trade, and Migration.” National Bureau of Economic Research Working Paper #6027. [27] Helliwell, John F. (1998) How Much Do National Borders Matter? Washington, D.C.: The Brookings Institution. [28] Helpman, Elhanan and Krugman, Paul (1985) Market Structure and Foreign Trade. Cambridge, MA: MIT Press. [29] Hummels, D. (1999) “Toward a Geography of Trade Costs.” mimeo, University of Chicago. [30] Hummels, D. and Levinsohn, J. (1993) “Monopolistic Competition and International Trade: Reconsidering the Evidence.”National Bureau of Economic Research Working Paper #4389. [31] Feenstra, Robert C. and Rauch, James E., eds. (1999) “Symposium on Business and Social Networks in International Trade.” Journal of International Economics 48(1): 1-150. [32] Lee, J. and Swagel, P. (1997) “Trade Barriers and Trade Flows across Countries and Industries.” Review of Economics & Statistics. 79 (3): 372-82. [33] Linneman, Hans. (1966) An econometric study of international trade ‡ows. Amsterdam: North-Holland. [34] Loertscher, R. and Wolter, F. (1980) “Determinants of intra-industry trade: Among countries and across countries.” Weltwirtshaftliches Archiv 116:280-293. 18 [35] McCallum, John (1995) “National Borders Matter: Canada-U.S. Regional Trade Pat- terns.” American Economic Review 85 (June): 615-23. [36] Maskus, K. (1991) “Comparing International Trade Data and National Characteristics Data for the Analysis of Trade Models,” in Peter Hooper and J. David Richardson, International Economic Transactions, NBER Studies in Income and Wealth Volume 55, Chicago: University of Chicago Press. [37] National Science Foundation, Division of Science Resources Studies (1997) Research and Development in Industry: 1994, Detailed Statistical Tables, NSF 97-331, by Raymond M. Wolfe (Arlington, VA, 1997). [38] OECD (1995) The OECD Input-Output Database. [39] OECD (Various Years) The OECD STAN Database. [40] OECD (1998) Labour Force Statistics 1977/1997: 1998 Edition. [41] Penn World Tables. Version 5.6 [42] Rauch, James E. (1999) “Networks Versus Markets in International Trade.” Journal of International Economics 48(1): 7-36. . [43] Scherer, F.M. and Ross, D. (1990) Industrial Market Structure and Economic Perfor- mance, Third Edition. Boston: Houghton-Mi-in Company. [44] Shiells, Clinton R., Stern, Robert M., and Deardor¤, Alan V. (1986) ”Estimates of the Elasticities of Substitution between Imports and Home Goods for the United States.” Weltwirtschaftliches Archiv 122(3): 497-519. [45] Tinbergen, Jan. (1962) Shaping the world economy - Suggestions for an international economic policy. New York: Twentieth Century Fund. [46] UNCTAD. (1991) Micro TCM System. [47] U.S. Bureau of the Census. (1978-1988) U.S. Exports, Schedule B, Commodity by Coun- try. FT446, Washington, D.C.: The Bureau. [48] U.S. Bureau of the Census. (1989-1994) U.S. Exports, Harmonized System, Commodity by Country. FT447, Washington, D.C.: The Bureau. [49] U.S. Bureau of Economic Analysis. (1985-1994) U.S. Direct Investment Abroad: Oper- ations of U.S. Parent Companies and Their Foreign A¢liates. Washington, D.C.: U.S. Bureau of Economic Analysis. [50] Wei, Shang-Jin (1996) “Intra-national Versus Inter-national Trade: How Stubborn are Nations in Global Integration.” National Bureau of Economic Research Working Paper #5531. [51] World Bank. World Development Indicators on CD-ROM. 19 A Data Appendix A.1 Countries and Industries Included in Data Set Countries Sections 3.1, 3.2, 4.1:Australia, Canada, Denmark, France, Germany, Japan, United Kingdom, United States Section 4.2: Australia, Canada, France, Germany, Italy, Japan, the Netherlands, Spain, United Kingdom Industries Sections 3.1, 3.2: Non-manufactured products; Food, Beverages, and Tobacco; Textiles, Apparel, and Leather; Wood Products and Furniture; Paper Products and Print- ing; Chemicals and Drugs; Petroleum Re…neries and Products; Rubber and Plastic Products; Non-metallic Mineral Products; Basic Metal Industries; Fabricated Metal Products, Machin- ery, and Equipment; Other Manufacturing; These categories are based on the ISIC sectoral division in the OECD Input-Output Database. Section 4.1: Food, Beverages, and Tobacco; Textiles, Apparel, and Leather; Chemicals and Drugs; Rubber and Plastic Products; Basic Metal Industries; These categories are based on the ISIC sectoral division in the OECD Input-Output Database, but are limited by data availability. Section 4.2: Chemicals and Allied Products; Electric and Electronic Equipment; Non- electric Machinery; Food and Kindred Products; Primary and Fabricated Metals; Trans- portation Equipment; and Other Manufacturing; These are the categories provided in the BEA publicly-available data. A.2 Trade and Production Data Sections 3.1, 3.2, 4.1 Data on OECD bilateral trade ‡ows are taken from Feenstra, Lipsey, and Bowen (1997), with the original source as the Statistics Canada World Trade Database. The data are provided on an SITC basis; they were concorded to ISIC based on Maskus (1991). Feenstra et al. also provide the trade data according to the WBEA classi…cation. These data were also concorded to an ISIC basis in order to con…rm the original SITC-ISIC concordance. Results did not change substantially. Domestic trade for manufactured goods is production (gross output) within each industry less exports from that industry. Non-manufacturing domestic trade is gross goods production less manufacturing production less non-manufacturing exports. The gross goods production data were provided by John Helliwell, with the original source as United Nations (1996) National Accounts Statistics: Main Aggregates and Detailed Tables. Non-manufacturing exports are calculated as total exports less manufacturing exports. Production data are from the OECD Statistical Analysis Database. The production data were converted to U.S. dollars using the annual exchange rate in the Database. Section 4.2 Data on exports from the U.S. to the countries in the sample are taken from Feenstra (1997), with the original sources as Bureau of the Census (1978-1988 and 1989- 1994). They were converted from the SITC classi…cation to the BEA classi…cation using 20 the concordance in Feenstra, Lipsey, and Bowen (1997). Production (gross output) data within industries for the U.S. are from the OECD Statistical Analysis Database. Domestic trade for manufactured goods is production (gross output) within each industry less exports from that industry. Domestic production, employment, and total export data are from the OECD Statistical Analysis Database. The production data were converted to U.S. dollars using the annual exchange rate in the Database. Data on the activities of foreign a¢liates of U.S. multinationals are from BEA (1985- 1994), as provided in Feenstra (1997). Total sales and local sales by foreign a¢liates are used as, respectively, production and consumption. A.3 GDP, Population, Distance variables The distance data were provided by John Helliwell. DISTcc0 is the distance from exporter k to importer j. It is generally measured from capital to capital and calculated using Great Circle Distances from Latitude and Longitude given in Direct Line Distances, by Fitzpatrick (1986). Own distances are calculated as 4 of the distance to its nearest trading partner. For 1 islands or countries with no trading partner in the sample group own distance was calculated as 2 of the minor radius of the country. These internal distances are consistent with the 1 formulation used by Wei (1996). For Sections 3.1, 3.2, 4.1, GDP and population data are taken from the PENN World Tables. For Section 4.2, GDP data are taken from the OECD National Accounts Statistics; population data are from the PENN World Tables and the U.N. Demographic Yearbook. A.4 Product Di¤erentiation and Information Cost Variables Advertising to sales ratio is from the COMPUSTAT database, as is the R&D to sales ratio. The original characterization of goods was provided by James Rauch and is further described in Rauch (1999). The method of aggregation is described in the text. Intra-industry trade as a proportion of total trade was calculated according to the Grubel-Lloyd index, as interpreted in Hummels and Levinsohn (1993). The data on information transfer across industries are from Bailey (1975) and are based on a survey of U.S. manufacturers. The data in the report are concorded to a U.S. SIC basis, aggregated using U.S. manufacturing output data for 1990, and then concorded to the ISIC categories. Data availability and concordance issues limit the data to …ve industries. Within these aggregated industries, the more disaggregated industries covered by the survey include between 70% and 100% of the aggregated total, based on 1990 U.S. production levels. It is assumed that the covered proportions are representative of the aggregate category. A.5 Calculation of elasticities from Rauch index The Rauch index values for the industries vary between 0.03 and 1.00. I set the two endpoints of the elasticity range (2 to 5) to the minimum and maximum Rauch index values, and used linear interpolation to assign elasticities to the intervening industries, based on their Rauch index values. 21 A.6 Other Variables Data on research and development expenditures in Section 4.2 are from BEA(various issues) for foreign a¢liates and for parent companies. For the U.S., the data are private expenditures on research and development as provided in National Science Foundation (1997). P DIST 0 0 Remoteness indices are calculated as Rc = c0 GDPc0 c , where the summation over c is c 0 over country c’s trading partners c within the sample. Population is used instead of GDP for some of the analyses. Endowment data used as instruments in Sections 3.1, 3.2, and 4.1 were provided by James Harrigan, with the original source as Penn World Tables (workers, capital stock) and World Bank World Development Indicators (agricultural land). Instruments for Section 4.2 are from OECD (1998) (labor force), World Bank World Development Indicators (agricultural land), and the Penn World Tables (capital stock). For the capital stock data, the last two years were extrapolated from a 1985 to 1992 series. A.7 Variables for Calculation of Indirect Utility The expression for indirect utility is in the theory appendix. For the calculations of the e¤ects of a large change, as described in the text, I examine the proportional change in indirect utility associated with reducing the barriers implied by measured border e¤ects. The details for the two scenarios are in the text. Each country’s change in indirect utility is based on an estimated country-level border e¤ect. The proportion of the traded goods in consumption (¹g ) are taken from the OECD Input-Output table. The measure of transport costs is based on Hummels (1999), Table 3, Panel 1. I used the estimated coe¢cients for the e¤ects of distance on transport costs for the U.S. in order to calculate distance-based measures of transport costs for the bilateral pairs in my sample. I used the average value to weight for the U.S.. The estimates are quite reasonable, with the ad valorem transport cost factor ranging between 2% and 12%. ¯ g and pg ` are assumed to be 1. Yc is the GDP cc of country c. B Theory Appendix The model is a very standard representation of trade in di¤erentiated products.37 All countries c0 produce and trade g di¤erentiated products, which are distinguished by their country of origin. Assume identical technologies across countries and a two-tier utility function in country c with Cobb-Douglas upper-level and Spence-Dixit-Stiglitz lower-level utility. g U c = ¦g (Xc )¹ g X ¹g = 1 g 37 Some references are Deardor¤ (1995) and Helpman and Krugman (1985). 22 X g g 1 g Xc = [ ¯ c0 (xcc0 )²g ] ²g c0 ¾g ¡ 1 ²g = ¾g where ¾ g is the elasticity of substitution between varieties of the good, ug represent Cobb- Douglas shares, and Xc is an aggregator over available varieties. g The representative consumer optimizes subject to a budget constraint: XX g g [ pcc0 xcc0 ] = Yc g c0 I solve the model using a two-stage budgeting procedure. Solving the lower-level optimization problem, I …nd that: (xg 0 )²g ¡1 cc pg 0 ¯ g c = cc g (xg )²g ¡1 cc g pcc ¯ c0 so ¡1 1 1 ¡1 xg = xg 0 (pg ) 1¡²g (pg 0 ) 1¡²g (¯ g ) 1¡"g (¯ g0 ) 1¡"g cc cc cc cc c c Substituting into the aggregator Xc and solving for xg 0 , I have: g cc ¡1 1 (pg 0 ) 1¡²g (¯ g0 ) 1¡"g xg 0 cc = P cc 1 c ²g g 1 Xc g 1¡"g g [ c0 (¯ c0 ) (pcc0 ) ²g ¡1 ] ²g De…ne a price index: X 1 ²g ²g ¡1 Gg = [ (¯ g0 ) 1¡"g (pg 0 ) ²g ¡1 ] ²g c c cc c0 X 1 = [ (¯ g0 )¾g (pg 0 )1¡¾g ] 1¡¾g c cc c0 So, I have that: pg 0 ¡¾g g xg 0 cc = ( gcc g ) Xc Gc ¯ c0 substituting into total expenditure on manufactures, I …nd that this total expenditure is Gc Xg . Solving the upper-level utility maximization problem, I …nd: g c ¹g Y c (¯ g0 )¾g xg 0 cc = g ¾g cg 1¡¾g (pcc0 ) (Gc ) Trade is characterized by both transport costs and other barriers to trade: pg 0 = pg0 c0 Dcc0 Tcc0 cc c g g 23 0 where pg 0 is the price of good g produced in c and imported by c, pg0 c0 represents the price of cc c g g good g at the factory in the country of production, and Dcc0 and Tcc0 represent, respectively, transportation costs (either domestic or international) and other costs of international trade. g 0 I assume that Tcc0 = 1 for trade within a country, i.e. if c = c. I now have: X 1 c c c g g Gg = [ (¯ g0 )¾g (pg0 c0 Dcc0 Tcc0 )1¡¾g ] 1¡¾g 0 c 0 Thus, consumption in location c of a variety produced in c is: ¹g Y c (¯ g0 )¾g xg 0 = c cc c c g g (Gg )1¡¾g (pg0 c0 Dcc0 Tcc0 )¾g Since total shipments must account for transport costs and ad valorem barriers, I …nd that 0 total shipments from a location c to a location c in industry g are: g ¹g Yc (¯ g0 )¾g c EXcc0 = (Gg )1¡¾g (pg0 c0 )¾g (Dcc0 Tcc0 )¾g ¡1 c c g g 0 Thus, I may de…ne the portion of national income for country c obtained from sales of good g produced in c`, as: X Yc Ycg = pg0 c0 ¤ (¹g (¯ g0 )¾g ( 0 )) c c c (Gg )1¡¾g (pg0 c0 )¾g (Dcc0 Tcc0 )¾g ¡1 c c g g Solving for ¯ g 0 , substituting in the expression for consumption of an individual variety cc 0 shipped to country c from country c , I …nd: g Yc Ycg 0 SHIPcc0 = P (Gg )1¡¾g (pg0 c0 Dcc0 Tcc0 ) c c g g ¾g Yc c (pg0 0 Dg 0 T g 0 )¾ g ¡1 (Gg )1¡¾ g c c c cc cc g where SHIPcc0 is total purchases of the g good by country c from a producer located in 0 country c , Yc is the income of the importer, Ycg is the income of the producer derived from 0 sales of good g, and Gc is a price index, as noted above. g g De…ne Rc0 and Rc as measures of a country’s remoteness, which measure the GDP- g weighted distance of a given country from alternative trading partners, where: X Rc0 = [ (pg0 c0 Tcc0 Dcc0 )(1¡¾g ) Yc G(¾g ¡1) ]¡1 g c g g c c Rc = G¾g ¡1 g c Assuming that a good g has a price of 1 in its country of origin, I thus have: log SHIPcc0 = log Yc + log Ycg + log Rg0 + log Rc g 0 c g g g ¡¾ g log Dcc0 ¡ ¾ g log Tcc0 24 which is the same as equation 1 in the text. Indirect utility in this model is given by:: ¹g ¹g Uc = Yc ¦g ( ) Gg c The calculations in the text use a percentage change in indirect utility between two alternative scenarios. 25 Figure 1 P Inelastic PB i τi PB e τe PFT i = PFT e Elastic Q TVB TVFT Figure 2 P Inelastic i PB ∆CS i ∆CS e e PB PFT i = PFT e Elastic TVB TVFT Q 26 Table 1 Section Countries Years Industries Sources of Production • Domestic 3.1 8 OECD 1990 11 manufacturing • All bilateral-pair imports 3.2 countries industries, non- manufactured products 5 manufacturing • Domestic 4.1 8 OECD 1990 industries* • All bilateral-pair imports 27 countries 7 manufacturing • Domestic 4.2 9 OECD 1989 to 1994 industries • Imports from U.S. countries • U.S. foreign affiliates * Only five manufacturing industries are used due to data availability. Table 2 Border Effects, Price Wedges (PWs), and Elasticities (1) Unobservable (2) Observable Implications for Economic Significance of Elasticity-Price Wedge Elasticity-Border Effect Aggregate Border Effects Relation Relation PWs Welfare Case 1 • Low elasticity goods have • Low elasticity goods have high High aggregate border effects High aggregate border high PWs border effects imply very high PWs on low effects probably imply 28 • High elasticity goods have • High elasticity goods have low elasticity products large adverse welfare low PWs border effects consequences, depending on the origins of PWs Case 2 • No relationship • Low elasticity goods have low Likely that high aggregate High aggregate border border effects border effects imply low PWs effects probably imply • High elasticity goods have high on high elasticity goods smaller adverse welfare border effects consequences, depending Or on origins of PWs • No Relationship Case 3 • Low elasticity goods have • Low elasticity goods have low High aggregate border effects High aggregate border low PWs border effects imply low PWs on high effects imply only small • High elasticity goods have • High elasticity goods have high elasticity goods adverse welfare high PWs border effects consequences (1) Relation between (a) elasticity of substitution between imports and domestic goods, and (b) import-domestic good perceived price wedges (unobservable) (2) Relation between (a) elasticity of substitution between imports and domestic goods, and (b) border effects (observable) Table 3 Product Differentiation/ Elasticity Effects Equation (i) (ii) (iii) (iv) Dependent Variable: ln(SHIPgcc') ln(Production Exporter) 0.79 * 0.79 * 0.79 * 0.79 * (0.04) (0.04) (0.04) (0.04) ln(GDP Importer) 0.85 * 0.85 * 0.85 * 0.85 * (0.04) (0.04) (0.04) (0.04) ln(Distance) -0.80 * -0.75 * -0.79 * -0.77 * (0.08) (0.08) (0.08) (0.08) ln(Remoteness Exporter) 0.37 * 0.37 * 0.37 * 0.37 * (0.16) (0.16) (0.16) (0.16) ln(Remoteness Importer) 1.39 * 1.39 * 1.39 * 1.39 * (0.15) (0.15) (0.15) (0.15) LOCAL 3.27 * 2.27 * 1.69 * 1.72 * (0.70) (0.33) (0.24) (0.23) Intra-Industry Trade/Total Trade -2.86 * LOCAL*IIT (1.01) Rauch Index -1.52 * LOCAL*Rauch (0.52) Industry R&D/Sales -17.45 * LOCAL*RDS (6.71) Industry Advertising/Sales -16.21 LOCAL*ADS (8.77) No. Obs. 768 768 768 768 2 R 0.85 0.85 0.85 0.85 SER 1.16 1.16 1.16 1.16 Note: Non-manufactured goods is the excluded industry. Industry-specific intercepts are included in all equations, but are not reported here. Industry-specific distance interaction terms are also included, but are not included here. IV with heteroscedasticity consistent standard errors. * Significant at the 5% level. 29 Table 4 Border Effects, Tariffs (%), and Price Wedges (%) Elasticities and Price Wedges (%) Elasticity=2 Range (2 to 5) Elasticity=5 Border Effect Un- Un- Un- (non-intra- Average Assumed Implied explained Assumed Implied explained Assumed Implied explained EU, non- Tariff Price Price Price Elasticity Price Elasticity* Price Elasticity Price adjacent (%) Wedge Wedge Wedge Wedge Wedge Wedge trade) OECD Industry Non-manufactured Products 20.21 2.00 349.50 4.22 103.93 5.00 82.43 Food, Beverages, Tobacco 25.44 8.59 2.00 404.42 395.83 4.63 101.14 92.55 5.00 91.04 82.45 Textiles, Apparel, Leather 12.65 12.20 2.00 255.60 243.41 2.76 150.53 138.33 5.00 66.11 53.91 30 Wood Products, Furniture 17.06 4.65 2.00 313.06 308.40 2.70 185.42 180.77 5.00 76.36 71.71 Paper Products and Printing 19.96 4.90 2.00 346.73 341.84 3.89 115.85 110.96 5.00 81.98 77.08 Chemicals and Drugs 10.51 6.90 2.00 224.23 217.33 4.04 79.06 72.16 5.00 60.08 53.19 Petroleum Refineries and Products 42.77 5.09 2.00 553.99 548.89 5.00 111.94 106.85 5.00 111.95 106.86 Rubber and Plastic Products 10.71 8.68 2.00 227.33 218.65 3.28 106.24 97.56 5.00 60.69 52.01 Non-metallic Mineral Products 24.76 6.55 2.00 397.62 391.07 2.70 227.67 221.12 5.00 90.00 83.46 Basic Metal Industries 4.89 4.89 2.00 121.10 116.21 3.09 67.22 62.33 5.00 37.35 32.46 Fabricated Metal Pdcts., Mach. and 8.26 4.93 2.00 187.37 182.44 2.05 180.73 175.81 5.00 52.54 47.61 Equip. Other Manufacturing 3.87 5.88 2.00 96.66 90.79 2.00 96.66 90.78 5.00 31.07 25.19 *Note: Elasticities calculated based on Rauch index, which varies between 0 and 1. Higher values indicate more differentiated products. Rauch Index: Max Min 1.00 0.03 Table 5 Border Effects, Tariffs, and Implied Elasticities What elasticities are implied by measured border effects and average tariffs? Border Effect (non-intra- Average Tariff (%) Implied Elasticity EU, non-adjacent trade) OECD Industry Non-manufactured Products 20.21 -- -- Food, Beverages, Tobacco 25.44 8.59 39.29 Textiles, Apparel, Leather 12.65 12.20 22.05 Wood Products, Furniture 17.06 4.65 62.36 Paper Products and Printing 19.96 4.90 62.64 Chemicals and Drugs 10.51 6.90 35.28 Petroleum Refineries and Products 42.77 5.09 75.62 Rubber and Plastic Products 10.71 8.68 28.49 Non-metallic Mineral Products 24.76 6.55 50.61 Basic Metal Industries 4.89 4.89 33.23 Fabricated Metal Pdcts., Mach. and Equip. 8.26 4.93 43.91 Other Manufacturing 3.87 5.88 23.69 31 Table 6 The Costs of Information Equation (i) (ii) (iii) (iv) (v) Dependent Variable: ln(SHIPgcc') ln(Production Exporter) 0.72 * 0.72 * 0.72 * 0.72 * 0.72 * (0.04) (0.04) (0.04) (0.04) (0.04) ln(GDP Importer) 0.86 * 0.86 * 0.86 * 0.86 * 0.86 * (0.05) (0.05) (0.05) (0.05) (0.05) ln(Distance) -1.17 * -1.12 * -1.10 * -1.09 * -1.17 * (0.09) (0.09) (0.09) (0.09) (0.09) ln(Remoteness Exporter) 0.26 0.26 0.26 0.26 0.26 (0.20) (0.19) (0.19) (0.19) (0.20) ln(Remoteness Importer) 1.50 * 1.50 * 1.50 * 1.50 * 1.50 * (0.20) (0.20) (0.20) (0.20) (0.20) LOCAL 1.23 * 1.61 * 3.55 * 1.86 * 1.22 * (0.24) (0.27) (1.03) (0.34) (0.31) Industry Variables: Products made to order -0.01 * LOCAL*ORDER (0.01) Frequent Sales/Technical Service -0.06 * LOCAL*SERVICE 0.03 Major purchase -0.01 * LOCAL*MAJOR 0.01 Infrequent purchase 0.00 LOCAL*INFREQUENT (0.01) No. Obs. 320 320 320 320 320 2 R 0.88 0.88 0.89 0.89 0.88 SER 0.92 0.92 0.91 0.91 0.92 Industry fixed effects and distance-industry interaction terms included. IV with heteroscedasticity consistent standard errors. * Significant at the 5% level. 32 Table 7 Location and Nationality g Dependent Variable: ln(SHIP cc') (i) (ii) Border Effect Border Effect Foreign Foreign Domestic Affiliate Domestic Foreign Domestic Affiliate Domestic Foreign 1 1 Sales Sales Sales Affiliates P-Value Sales Sales Sales Affiliates P-Value ln(Production) 0.99 * 0.98 * 0.98 * 0.97 * (0.02) (0.02) (0.02) (0.02) ln(GDP Consumer) 0.39 * 0.36 * 0.39 * 0.35 * (0.03) (0.03) (0.03) (0.03) ln(Distance) -0.83 * -0.80 * -0.65 * -0.60 * (0.05) (0.05) (0.09) (0.09) ln(Remoteness Producer) 0.75 * 0.73 * 0.73 * 0.70 * (0.16) (0.16) (0.15) (0.16) ln(Remoteness Consumer) 0.97 * 0.96 * 0.97 * 0.96 * (0.10) (0.10) (0.09) (0.10) Location Effects: Local 2.38 * 2.35 * 10.77 10.52 0.81 (1 for dom. or foreign-affil. sales, 0 (0.16) (0.17) otherwise) CHEM*Local 3.12 * 2.67 * 22.59 14.45 0.00 (0.35) (0.37) ELEC*Local 1.69 * 1.63 * 5.41 5.12 0.72 (0.35) (0.36) FOOD*Local 3.88 * 4.02 * 48.44 55.73 0.37 (0.35) (0.37) MACH*Local 1.47 * 1.56 * 4.34 4.74 0.60 (0.37) (0.38) OTHER*Local 3.15 * 3.17 * 23.41 23.86 0.91 (0.36) (0.38) PFMET*Local 2.25 * 2.11 * 9.53 8.25 0.37 (0.35) (0.37) TRANS*Local 0.85 * 0.98 * 2.35 2.67 0.41 (0.36) (0.37) Number of Observations 692 692 692 692 Estimation 3SLS 3SLS Industry Fixed Effects Yes Yes Distance-Industry Interaction Yes Yes Time Period 1989-1994 1989-1994 Year Fixed Effects Yes Yes 1. P-value is the probability associated with the hypothesis that domestic sales and foreign affiliate sales effects are equal. A low value indicates that we are able to reject that hypothesis, I.e. that the two differ from each other significantly. * Significant at the 5% level. 33 Table 7 (Cont.) Location and Nationality Dependent Variable: ln(SHIPgcc') (iii) (iv) Border Effect Border Effect Foreign Foreign Domestic Affiliate Domestic Foreign Domestic Affiliate Domestic Foreign Sales Sales Sales Affiliates P-Value1 Sales Sales Sales Affiliates P-Value1 ln(Production) 0.98 * 0.97 * 0.97 * 0.97 * (0.02) (0.02) (0.02) (0.02) ln(GDP Consumer) 0.40 * 0.37 * 0.41 * 0.38 * (0.03) (0.03) (0.03) (0.03) ln(Distance) -0.89 * -0.87 * -0.89 * -0.86 * (0.05) (0.05) (0.05) (0.05) ln(Remoteness Producer) 0.81 * 0.75 * 0.80 * 0.75 * (0.18) (0.18) (0.18) (0.18) ln(Remoteness Consumer) 1.00 * 0.99 * 1.01 * 1.00 * (0.10) (0.10) (0.10) (0.10) R&D/Sales, U.S. as ex., Direct -2.00 (1.26) R&D/Sales, Interaction -5.80 * (1.69) Location Effects: Local 2.16 * 2.06 * 8.66 7.84 0.32 2.13 * 2.13 * 8.44 8.44 1.00 (1 for dom. or foreign-affil. sales, 0 otherwise) (0.17) (0.18) (0.17) (0.18) Number of Observations 627 627 627 627 Estimation 3SLS 3SLS Industry Fixed Effects Yes Yes Distance-Industry Interaction Yes Yes Time Period 1989-1994 1989-1994 Year Fixed Effects Yes Yes 1. P-value is the probability associated with the hypothesis that domestic sales and foreign affiliate sales effects are equal. A low value indicates that we are able to reject that hypothesis, I.e. that the two differ from each other significantly. * Significant at the 5% level. 34 Table 8 Welfare Effects of Elimination of Border Effects (% Change in Indirect Utility) Suppose barriers to imports drop to 0. Sigma 1 = 2 Sigma 1 =5 Sigma 1 = 10 Importing Country % Change in IU % Change in IU % Change in IU Australia 61.16 15.74 6.01 Canada 61.26 16.45 6.68 Denmark 70.41 17.55 7.05 France 37.57 11.64 4.97 Germany 37.61 11.66 4.97 Japan 36.17 11.65 4.97 UK 40.01 12.22 5.22 US 25.08 8.46 3.74 Suppose barriers to imports drop to actual tariff levels. Sigma 1 = 2 Sigma 1 =5 Sigma 1 = 10 Importing Country % Change in IU % Change in IU % Change in IU Australia -- -- -- Canada 57.51 13.84 4.48 Denmark 68.10 16.17 6.08 France 35.63 10.29 4.02 Germany 35.25 10.03 3.83 Japan 32.66 8.90 2.64 UK 38.24 10.99 4.36 US 22.21 6.07 1.65 1. Sigma is the elasticity of substitution between domestic goods and imports. 35 Table A.1 Border Effects Across Industries Dependent Variable: ln(SHIPgcc') Implied Border Effects Non Intra-EU, EU-EU, Non-Adjacent Non-Adjacent trade trade ln(Production Exporter) 0.81 * (0.04) ln(GDP Importer) 0.87 * (0.04) ln(Distance) -0.46 * (0.11) ln(Remoteness Exporter) 0.23 (0.17) ln(Remoteness Importer) 1.25 * (0.16) Adjacency 0.62 * (0.12) EU members 0.55 * (0.21) Non-manufactured products 3.01 * 20.21 11.65 (0.57) Food, Beverages, Tobacco 3.24 * 25.44 14.67 (0.54) Textiles, Apparel, Leather 2.54 * 12.65 7.29 (0.50) Wood Products, Furniture 2.84 * 17.06 9.84 (0.60) Paper Products and Printing 2.99 * 19.96 11.50 (0.54) Chemicals and Drugs 2.35 * 10.51 6.06 (0.44) Petroleum Refineries and Products 3.76 * 42.77 24.65 (0.80) Rubber and Plastic Products 2.37 * 10.71 6.18 (0.49) Non-metallic Mineral Products 3.21 * 24.76 14.27 (0.59) Basic Metal Industries 1.59 * 4.89 2.82 (0.55) Fabricated Metal Products, Machinery, and Equip. 2.11 * 8.26 4.76 (0.47) Other Manufacturing 1.35 3.87 2.23 (0.75) No. Obs. 768 2 R 0.85 SER 1.16 Note: Non-manufactured goods is the excluded industry. Industry-specific intercepts are included in all equations, but are not reported here. Industry-specific distance interaction terms are also included, but are not included here. IV with heteroscedasticity consistent standard errors. * Significant at the 5% level. 36