Robert M. La Follette School of Public Affairs at the University of Wisconsin-Madison Working Paper Series La Follette School Working Paper No. 2013-009 http://www.lafollette.wisc.edu/publications/workingpapers A Consideration of Patterns of Intra-Industry Trade: Why Do Countries Export and Import More/Less in the Same Industry? Isao Kamata La Follette School of Public Affairs, University of Wisconsin–Madison ikamata@lafollette.wisc.edu 1225 Observatory Drive, Madison, Wisconsin 53706 608-262-3581 / www.lafollette.wisc.edu The La Follette School takes no stand on policy issues; opinions expressed in this paper reflect the views of individual researchers and authors. A Consideration of Patterns of Intra-Industry Trade: Why Do Countries Export and Import More/Less in the Same Industry? Isao Kamata § University of Wisconsin–Madison This version: March 15, 2013 Abstract It is observed in data that in most manufacturing industries a country’s exports and imports, measured in values per GDP within an industry, are positively correlated. This indicates that the countries exporting more (less) in an industry tend to import more (less) in the same industry, relative to the size of their economy. In this paper I employ the model of intra-industry trade that has been widely used in the literature, also allowing transport costs and price asymmetry across countries, to examine whether and how the model can explain such positive correlations between a country’s export and import in the same industry. The theoretical prediction of the model is that a country’s export and import per GDP could both be smaller (greater) in the same industry if the country‘s production share in the world total in that industry and its GDP share in the world are both greater (smaller). The result of the data analysis shows, however, that the model can predict only smaller or negative correlations in many industries, while countries’ production sizes and GDPs are almost perfectly positively correlated. This result seems to imply that an alternative mechanism missed in the conventional model is needed to explain the pattern of intra-industry trade. Keywords: Intra-industry trade; Export-import correlation; New Trade Theory JEL classification: F10, F12, F14, I am particularly grateful to Alan Deardorff and Juan Carlos Hallak for extensive discussions and also thank Gary Saxonhouse for valuable and helpful comments. I am solely responsible for all remaining errors. § 1. Introduction The law of comparative advantage is the oldest and the most prevailing proposition in the theory of international trade. The law states that a country will specialize in production and export of the goods in which it has an advantage in relative production costs due to difference in technology (in the Ricardian sense) or difference in relative factor abundance (in the sense of Heckscher-Ohlin (H-O)); and will import the other goods.1 This provides a sharp prediction of patterns of international trade in the case in which goods are homogeneous: a country will not import the goods that it exports, or vice versa, and thus there is no intra-industry trade. The “new theory” of international trade such as the work by Krugman (1979, 1980) introduced the market structure of monopolistic competition to explain the intra-industry trade that was observed in reality. Helpman and Krugman (1985) incorporated the monopolistic competition model with the traditional H-O framework, but the original prediction by the law of comparative advantage is still valid in their model, in a slightly weaker manner: that is, a country is expected to be a net exporter of the industry in which it has comparative advantage. The expected cross-country pattern of trade within an industry, then, would be such that, if a country exports more in an industry, the country should import less in that industry. What do data tell about the pattern? Table 1 shows the correlations between the values of exports and imports across countries within each of the manufacturing industries classified by the 3-digit ISIC (Revision 2).2 The values of exports and imports are divided by GDP of each country in order to adjust for the size of the economy. The result is surprising: in most of the manufacturing industries, exports and imports of the countries are positively correlated, and the correlations are fairly strong in some industries. This implies that the countries that export more (relative to the size of the economy) in an industry tend to import more in the same industry, and vice versa. This seems to be the opposite to what is expected according to the law of comparative advantage.3 The model of intra-industry trade with monopolistic competition of differentiated products has been widely used in the literature, such as Romalis (2004). The purpose of this 1 Indeed in some cases countries will specialize incompletely and the law of comparative advantage cannot predict the exact patterns of countries’ specialization. Deardorff (1980) has a great summary on this. 2 Description of the data is in Section 5 of this paper. 3 The correlation cannot be positive if products are homogeneous and thus there is no intra-industry trade. That is, when a country exports in an industry (Xci>0), it does not import in that industry at all (Mci=0), or vice versa. Thus, the covariance Cov(Xci,Mci) ≡ E[XciMci]-E[Xci]·E[Mci] is definitely non-positive since the first term is zero but E[Xci]≥0 and E[Mci]≥0 so that the second term is negative. Note that the signs of covariance and correlation coefficient are the same. 1 1. Introduction The law of comparative advantage is the oldest and the most prevailing proposition in the theory of international trade. The law states that a country will specialize in production and export of the goods in which it has an advantage in relative production costs due to difference in technology (in the Ricardian sense) or difference in relative factor abundance (in the sense of Heckscher-Ohlin (H-O)); and will import the other goods.1 This provides a sharp prediction of patterns of international trade in the case in which goods are homogeneous: a country will not import the goods that it exports, or vice versa, and thus there is no intra-industry trade. The “new theory” of international trade such as the work by Krugman (1979, 1980) introduced the market structure of monopolistic competition to explain the intra-industry trade that was observed in reality. Helpman and Krugman (1985) incorporated the monopolistic competition model with the traditional H-O framework, but the original prediction by the law of comparative advantage is still valid in their model, in a slightly weaker manner: that is, a country is expected to be a net exporter of the industry in which it has comparative advantage. The expected cross-country pattern of trade within an industry, then, would be such that, if a country exports more in an industry, the country should import less in that industry. What do data tell about the pattern? Table 1 shows the correlations between the values of exports and imports across countries within each of the manufacturing industries classified by the 3-digit ISIC (Revision 2).2 The values of exports and imports are divided by GDP of each country in order to adjust for the size of the economy. The result is surprising: in most of the manufacturing industries, exports and imports of the countries are positively correlated, and the correlations are fairly strong in some industries. This implies that the countries that export more (relative to the size of the economy) in an industry tend to import more in the same industry, and vice versa. This seems to be the opposite to what is expected according to the law of comparative advantage.3 The model of intra-industry trade with monopolistic competition of differentiated products has been widely used in the literature, such as Romalis (2004). The purpose of this 1 Indeed in some cases countries will specialize incompletely and the law of comparative advantage cannot predict the exact patterns of countries’ specialization. Deardorff (1980) has a great summary on this. 2 Description of the data is in Section 5 of this paper. 3 The correlation cannot be positive if products are homogeneous and thus there is no intra-industry trade. That is, when a country exports in an industry (Xci>0), it does not import in that industry at all (Mci=0), or vice versa. Thus, the covariance Cov(Xci,Mci) ≡ E[XciMci]-E[Xci]∙E[Mci] is definitely non-positive since the first term is zero but E[Xci]≥0 and E[Mci]≥0 so that the second term is negative. Note that the signs of covariance and correlation coefficient are the same. 1 paper is to examine whether this conventional model can explain the positive correlation between a country’s exports and imports in the same industry that is observed in data. I start with the model under a conventional (but restrictive) set of assumptions such as free trade and symmetric prices, and later introduce a more general form by allowing transport costs and price differences across countries in order to see whether such relaxing affects the prediction of the model. The restrictive version of the model is then tested using cross-country data on manufacturing production, trade, and GDP by comparing the predicted correlations to the actual. The relaxed version is not tested because it involves variables that are not observable in the data. Instead, a possible procedure for testing the relaxed version of the model without having those unobservable variables is proposed. The key finding of this paper is the following: The theoretical results from the comparative statics of the model indicate that a country’s exports and imports per GDP could both be smaller (greater) in the same industry if the country‘s production share in the world total in that industry and the country’s GDP share in the world are both greater (smaller). The data analysis shows, however, that the model can predict only smaller or negative correlations in many manufacturing industries, while the sizes of the countries’ production and their GDPs are almost perfectly positively correlated. This result seems to imply that an alternative mechanism that is missed in the conventional model is needed to explain the pattern of intra-industry trade. The organization of the paper is as follows: In Section 2 a piece of evidence is presented to show that the positive correlation between exports and imports is robust regardless the level of industry aggregation. Section 3 presents the model and derives expressions of key variables. Section 4 shows comparative statics of the key variables to show how the model predicts ‘co-movement’ of a country’s exports and imports. Section 5 performs a data analysis for the restrictive version of the model, together with description of the data employed. Concluding Section 6 discusses the results. 2. Is It Due to Aggregation? Finger (1975) claimed that intra-industry trade would disappear if commodities were better-classified at a more disaggregated level. Likewise, one might ask whether the positive cross-country correlation between exports per GDP and imports per GDP within the same industry shown in Table 1 is simply due to the aggregation of industries, and whether, at a less aggregated level, countries do indeed import different goods from those 2 they export? In order to check whether this is the case or not, let us see the correlations at as disaggregated level as possible. Although my analysis overall in this paper is based on the industry classification (the International Standard Industrial Classification: ISIC), the data at a disaggregated-level ISIC are not available for multiple countries. Therefore, in this section I employ United Nations data on world trade classified by the Standard International Trade Classification (SITC: Revision 2)4 and calculate the correlation coefficient between exports per GDP and imports per GDP within each commodity by 3-, 4-, and 5-digit SITC.5 In addition, I draw on the data for the 6-digit Harmonized System (HS: 1992) from the same database for further disaggregating. The number of commodities included in the data for SITC at each digit level is 239, 786, and 1,464 respectively, and the number of countries is 97, 95, and 94, respectively. For the 6-digit HS, the number of commodities is 5,035 and the number of countries is 31.6 Figure 1 presents the histograms showing the distribution of the number of commodities over the level of correlations between exports and imports per GDP for each level of the aggregation. As shown in the figure, the distributions of the correlation coefficients do not change much by the level of aggregation, and the correlations are positive in most commodities. This shows that, at least for any of the aggregation levels that the data allow, the positive correlation between exports and imports (both per GDP) is robust. The aggregation may not matter for this pattern. 3. The Model The model that I present here is a version of the monopolistic competition model of intra-industry trade employed by Romalis (2004). I first present the common assumptions that the model is based on throughout this paper, and derive common elements of the model. I then consider three cases, ranging from the most restrictive assumptions to the least restrictive assumptions, to derive three versions of the model. (1) Common Assumptions The assumptions that are common to all the versions of the model are the 4 5 6 The source of the data is United Nations Commodity Trade Statistics Database, or UN Comtrade (on-line). 5-digit is the most disaggregated level for SITC (Rev.2) available in UN Comtrade. The data for HS are available only for a limited number of countries, which are mainly the OECD members. 3 following: 1. There are M (>1) countries (c = 1, 2, …, M) in the world (the multi-country setting). (Note, however, that in the following parts of this paper I also apply the two-country setting (country c and “the rest of the world”) in order to examine the model in a simpler manner.) 2. All consumers in all countries have identical Cobb-Douglas preferences over industries, so that consumers’ expenditure shares on each industry are constant for all prices and incomes. bi shows the expenditure share for industry i. That is; U bi ln Q i ; (1) b (2) i i 1. i 3. Each industry is monopolistically competitive. That is, each industry consists of a number of varieties that are imperfect substitutes for each other, and each firm produces one variety. nci denotes the number of firms in industry i in country c, and Ni denotes the world total number of firms in the industry: that is; M N i nci . (3) c 4. Consumers have identical preferences with a constant elasticity of substitution (CES) over varieties within an industry. By interpreting Qi in Equation (1) above as consumers’ sub-utility, this is shown by: Ni Q [ (qcD ( ))i ]1/ i ; i (4) i where qcD(ω) is the quantity of product (variety) ω in industry i demanded by consumers, θi = 1 - 1/σi where σi is the elasticity of substitution among varieties in the industry (σi > 1). 5. There may be transport costs for international trade. For modeling convenience, I use the “iceberg” form of transport costs that may vary across industries, according to which, for each industry i, τi ≥1 units of output must be shipped from one country for one unit to reach any other country. 4 Under these assumptions, the quantity of variety ω in industry i that is demanded by consumers in country c is: qcD ( ) pˆ c ( ) i Eci ; 1 i ˆ p ( ') c (5) 'i where pˆ c ( ) is the price of variety ω (in industry i) paid by consumers in country c. Eci denotes the total expenditure of consumers in country c on products in industry i: that is; Eci b iYc ; (6) where Yc is total income, or GDP, of country c. It is convenient to define the ideal price index for industry i in country c: 1 Gci [ pˆ c ( )1 i ]1 i . (7) i Note that, letting p(ω) be the price charged by the firm producing variety ω, the consumers’ price is pˆ ( ) = τip(ω) for foreign consumers due to the trade cost, while pˆ ( ) = p(ω) for domestic consumers. Hence, the ideal price index defined by Equation (7) becomes: nci G [ ( p( ) i c 1 i nci ' ) { ( p( ')) i c ' c 1 i 1 1 i }] . (8) ' In equilibrium, the firm producing variety ω must supply the quantity that is demanded (qS = qD) to domestic consumers, but must supply and ship τi times the quantity demanded (qS = τqD) to consumers in any other country due to the “iceberg” trade cost. Hence, the value of output produced by firm ω (located in country c), or the revenue of the firm, is: p ( )q S ( ) p( )[biYc biYc ( p ( ) i ( i p ( )) i i i { b Y ] c' (Gci )1 i c ' c (Gci ' )1 i p ( ) 1 i i p( ) 1 i i ) { b Y ( ) }. c' Gci Gci ' c ' c (2) Case I In this first case, I add the following set of assumptions to the common assumptions presented above: 5 (9) I-6. Production technologies of all varieties in an industry are identical in the world. I-7. Free and frictionless international trade: countries costlessly trade their products each other. That is, τi = 1 for any industry i. I-8. In equilibrium, the producers’ prices of all varieties in an industry are equal to each other. That is, p( ) p i i . Note that, under these assumptions, the output of each variety (or the size of each firm) in an industry is identical: qS(ω) = qiS i . With these additional assumptions, Equations (8) and (9) above are modified to the following version for this Case I (from now on, for simplicity, the output quantity qS is denoted by q without the script S): i 1 i G G [n ( p ) i c i i c p i q i bi 1 i 1 i 1 i n ( p ) c ' c i c' 1 M ( Yc ) . Ni c ] 1 i 1 i p (N ) i c ; (8-I) (9-I) That is, all firms in an industry have the same size and revenue. Now I derive the expressions for the values of production, exports, and imports of a country in an industry for this Case I. Since my interest is in comparison of values of exports and imports in an industry, I suppress the script i denoting an industry from the expressions below. Let VQc, VXc, and VMc be the values of production, exports, and imports of country c in an industry, respectively; and VQw, VXw, and VMw be those of the world (total). Then; nc M ( Yc ) ; N c VQc nc pq b VX c b (10-I) nc ( Yc ' ) ; N c ' c VM c bYc n c ' c N c' bYc (11-I) ( N nc ) . N (12-I) Equation (11-I) is so because VXc is the total value of the varieties produced in and shipped from country c to all other countries. Equation (12-I) is because VMc is the total value of the foreign varieties (for country c) bought by consumers in country c. 6 Therefore, for each country c, the values of exports and imports in an industry, as ratios to the country’s GDP, are, respectively:7 VX c VQw 1 yc (13-I) xc ; Yc Yw yc VM c VQw (1 xc ) ; Yc Yw (14-I) where xc ≡ VQc/VQw is country c’s share of world production in the industry, and yc ≡ Yc/Yw is country c’s share of the world income (GDP). Note that, in this Case I, since the values of output (or revenues) of all firms in an industry are identical, xc is equal to country c’s share of the number of firms in the industry to the world total number of firms: xc = ncpq/Npq = nc/N. Correlation Coefficient In this Case I, the correlation coefficient between the values of exports and imports per GDP within an industry, ρi(VXci/Yci, VMci/Yci), is derived as follows:8 VX i VM ci 1 yc i 1 yc i i i ( c , ) i( xc ,1 xci ) i ( xc , xc ) . (15-I) Yc Yc yc yc (3) Case II In this second case, I keep the same set of assumptions as in Case I above, except one: I introduce a positive trade cost in the model rather than free trade. That is, assumption I-7 in Case I is replaced with the following: II-7. There is a transport cost for international trade in industry i: i.e., τi > 1. The other two assumptions I-6 and I-8 are kept in this Case II.9 With this second set of assumptions added to the common assumptions 1 through 5, Equations (8) and (9) are modified into the following version for this Case II (suppressing the script i for the industry):10 1 Gc p[nc T ( N nc )]1 where T 1 ; 7 See Appendix C-1 for derivation of Equations (13-I) and (14-I). See Appendix C-2 for the details of the derivation of the correlation coefficient. 9 Note that the values of output (pq) are identical for all firms in an industry also in this Case II. 10 See Appendix C-1 for derivation of the following equations (8-II) through (14-II). 8 7 (8-II) pq b[Yc ( b[ p 1 p ) {Yc ' ( )1 }] Gc Gc ' c ' c Yc Yc ' T { }] . nc T ( N nc ) c ' c nc ' T ( N nc ' ) (9-II) Then, using the same notations as in Case I, the value of production in country c in the industry is: Yc Yc ' (10-II) VQc nc pq ncb[ T { }] . nc T ( N nc ) c ' c nc ' T ( N nc ' ) The first term in the equation shows how much value is consumed domestically in country c, and the second shows how much is shipped to other countries. Hence, the value of exports from country c in an industry is: Yc ' (11-II) VX c ncbT { }. c ' c nc ' T ( N nc ' ) On the other hand, the value of country c’s imports in an industry is: ( N nc )bTYc . VM c nc T ( N nc ) (12-II) Thus, for each country c, the values of exports and imports in the industry as ratios to the country’s GDP are, respectively; VX c VQw xc yc ' (13-II) T { }; Yc Yw yc c 'c (1 T ) xc ' T VM c VQw T Txc { }; Yc Yw T (1 T ) xc (14-II) where xc ≡ VQc/VQw and yc ≡ Yc/Yw as defined in Case I. Since also in this Case II the values of output (or revenues) of all firms in an industry are identical, it holds that xc = nc/N . On the other hand, in order for both prices and quantities to be the same for all varieties/firms in one industry in equilibrium, as assumed in this Case II, the following condition must be satisfied for any pair of countries:11 yc x T (1 xc ) c ' c c yc ' xc ' T (1 xc ' ) 11 (II-A) See Appendix C-3 for details. 8 yc ' yc . xc ' T (1 xc ' ) xc T (1 xc ) Substituting this into Equation (13-II) yields the following expression of a country’s export value per GDP: VX c VQw xc yc VQw xc . (13-II-2) T { } T ( M 1) Yc Yw yc c 'c xc T (1 xc ) Yw xc T (1 xc ) Correlation Coefficient The correlation coefficient ρi(VXci/Yci, VMci/Yci) in this Case II is the following:12 i ( VX ci VM ci xci xci , ) i ( i ,1 ) 1 . Yc Yc xc T (1 xci ) xci T (1 xci ) (15-II) Therefore, in this Case II (with all the assumptions holding strictly), exports and imports per GDP within an industry must be perfectly negatively correlated. The correlation cannot be positive at all, which means that this Case II model cannot explain the positive correlations that are observed in the data. I thus omit this Case II for further analysis in the rest of this paper. (4) Case III In Case III, I also relax the other two assumptions that have been kept in Cases I and II. Now instead of the assumptions I-6 and I-8 in Case I,13 the following assumptions are introduced: III-6. Production technologies of varieties (or firms) in an industry are identical within a country, but not necessarily identical across countries. III-8. In equilibrium, the prices of varieties charged by firms (excluding trade cost) in an industry are the same within a country, but different in general across countries.14 That is, p(ω) = pci if ω is a variety in industry i produced in country c, but pc’i ≠ pci for c’ ≠ c. Thus, in this Case III, the quantities produced (or the sizes of firms) are the same for 12 See Appendix C-2 for the details of the derivation of the correlation coefficient. Remember that these two assumptions were kept also in Case II. 14 This may be because of technology differences across countries by the assumption III-6. Or, this can be interpreted as the case in which factor prices are different across countries, while production technologies are identical for all firms in the world. 13 9 varieties in a country (q(ω) = qci if ω is a variety in industry i produced in country c), but not across countries in general (qc’i ≠ qci for c’ ≠ c). For the trade cost, the assumption II-7 holds here as in Case II: that is, a transport cost τi >1 is incurred for trade of products in industry i between any pair of two countries. With this third set of assumptions, the expressions of the ideal price index and the value of output of a single variety/firm are modified to the following version for Case III (suppressing the script i for an industry):15 1 Gc [nc pc1 T (nc ' pc1' )]1 where T 1 ; (8-III) c ' c pc qc pc1 b[ Yc Yc ' T { }] . 1 1 1 T (nc ' pc ' ) c ' c nc ' pc ' T (nc " pc " ) 1 c nc p c ' c (9-III) c " c ' Accordingly, country c’s values of production, exports, and imports in an industry are as follows: VQc nc pc1 b[ 1 c nc p VX c nc pc1 bT { c ' c VM c Yc Yc ' T { }] ; (10-III) 1 1 1 T (nc ' pc ' ) c ' c nc ' pc ' T (nc " pc " ) c ' c 1 c' nc ' p c " c ' Yc ' }; T (nc" pc1" ) (11-III) c " c ' bYcT (nc ' pc1' ) 1 c nc p c ' c T (nc ' pc1' ) . (12-III) c ' c Hence, for each country c, the values of exports and imports in an industry, as ratios to the country’s GDP, become as follows: VX c VQw nc pc1 yc ' [ T { }] ; 1 Yc Yw yc T (nc " pc1" ) c ' c nc ' pc ' (13-III) c " c ' T (nc ' pc1' ) VM c VQw c ' c [ ]. 1 Yc Yw nc pc T (nc ' pc1' ) (14-III) c ' c yc ≡ Yc/Yw as defined. Note, however, that in this Case III VXc/Yc and VMc/Yc can no longer 15 See Appendix C-1 for derivation of the following equations (8-III) through (14-III). 10 M usefully be expressed using xc VQc / VQw nc pc qc / nc pc qc since the values of output of c firms in an industry are different across countries (pc’qc’ ≠ pcqc for c’ ≠ c), so that xc is not equal to the share in the number of firms, nc/N, in this case. Correlation Coefficient The correlation coefficient ρi(VXci/Yci, VMci/Yci) in this Case III is as follows:16 VX i VM ci ni p i1 i ( c , ) i ( c c Yc Yc yc { n c ' c i c' (n i c' yc ' pci1' ) c ' c }, ) pci1' T i (nci " pci1" ) nci pci1 T i (nci ' pci1' ) c " c ' c ' c …… (15-III) Two-country setting Here I also consider the two-country version (i.e., country c and ‘the rest of the world’) of the model, in addition to the multi-country version that is shown above. In the two-country setting, since xc’ = 1-xc and yc’ = 1-yc for all c’≠c, Equations (13-III) and (14-III) are modified into the following expressions, respectively:17, 18 VX c VQw (1 yc ) Tnc pc1 [ ]; (13’-III) Yc Yw yc {( N nc ) Tnc pc1 } VM c VQw T ( N nc ) [ ]. 1 Yc Yw nc pc T ( N nc ) (14’-III) The corresponding correlation coefficient is:19 VX i VM ci (1 yc ) nci pci1 N i nci i ( c , ) i ( , ) Yc Yc yc {N i (1 T i pci1 )nci } T i N i ( pci1 T i )nci i ( (1 yc ) T i nci pci1 nci pci1 , ) yc {N i (1 T i pci1 )nci } T i N i ( pci1 T i )nci …… 16 (15’-III) See Appendix C-2 for the details of the derivation of the correlation coefficient. In these expressions for the two-country version, the price of foreign varieties is normalized to 1 (p-c = 1), so pc denotes the relative price of home (country c’s) varieties to that of foreign varieties. 18 See Appendix C-1 also for derivation of these two equations (13’-III) and (14’-III). 19 See Appendix C-2 for derivation. 17 11 4. How Does The Model Predict ‘Co-movement’ of Exports and Imports?: Comparative Statics The expressions for the correlation coefficients between exports and imports per GDP in each case, which are (15-I), (15-II), (15-III) or (15’-III) derived in the previous section, are not directly informative by themselves in order to examine in what circumstances the correlation becomes positive or negative. Therefore, in this section, I perform comparative-static analyses on the expressions for the values of exports and imports per GDP, VXc/Yc and VMc/Yc, of a country in an industry in order to give some intuition about how VXc/Yc and VMc/Yc within an industry will ‘co-move’ according to the theoretical model.20 Each of Cases I and III presented in the previous section are analyzed below in order.21 (1) Case I I first examine the effects of change in a country’s share of production value in an industry, xc = VQc/VQw, on the country’s values of exports and imports per GDP in that industry. From Equations (13-I) and (14-I) above; (VX c / Yc ) VQw (1 yc ) (16-I) 0; xc Yw yc (VM c / Yc ) VQ w 0. xc Yw (17-I) That is, if the shares of countries’ GDPs are fixed, the country that has a larger production share in an industry has a larger value of exports per GDP in that industry, and on the other hand a smaller value of imports per GDP in the same industry. Next I examine the effects of change in a country’s GDP share, yc = Yc/Yw. Again, from Equations (13-I) and (14-I); (VX c / Yc ) VQ 1 w 2 xc 0 ; (18-I) y c Yw y c (VM c / Yc ) 0. yc (19-I) 20 The drawback of comparative statics should be noted, however: it treats the variables in the expressions of VXc/Yc and VMc/Yc (i.e., production share (xc) and the GDP share (yc) for Case I; the price of varieties (pc), the number of varieties (nc), and yc for Case III) as all exogenous, so that it disregards the possibility of endogeneity among these variables. A direct analysis of the derived correlation coefficient is more desirable, and I also perform it though only for Case I. See Appendix A. 21 As shown in the previous section, the Case II model cannot explain the positive correlation between exports and imports at all, so Case II is not examined here. 12 That is, if the sizes of the production of countries are all equal in an industry, the country that has a larger GDP share in the world has a smaller value of exports per GDP in an industry, but the values of imports per GDP do not vary across countries. The mechanism generating these results is as follows. Under the assumptions for this Case I, each firm in an industry has an identical value of production (pq), and, out of the production value of each variety, the value exported to (the value consumed by consumers in) all other countries (or the rest of the world) is equal for each variety, which is determined by the size of incomes of all other countries. Thus, after adjusting by a country’s income size (GDP), the value of exports of the country in an industry is larger when (i) the country has a larger number of varieties/firms in that industry, and thus a higher share of production in the industry; or (ii) the income size of all other countries relative to that country ((1-yc)/yc) is larger: i.e., that country’s income share is smaller. On the other hand, the value of imports of the country in an industry is determined by (i) the number of varieties produced in all other countries; and (ii) that country’s income size. The latter does not matter for the value adjusted by GDP of the country, so the value of imports depends only on the number of varieties produced in the rest of the world: i.e., a country’s import value is larger when more varieties are produced out of the country, which means that the country has a smaller production share in that industry. The following table summarizes the prediction of the change in exports and imports per GDP according to the Case I model. xc ↑ yc ↑ VXc/Yc ↑ ↓ VMc/Yc ↓ -- Therefore, the Case I model provides the following prediction: if a country’s share in the production value in an industry and its GDP share are both smaller (greater),22 then that country’s exports and imports in values per GDP in the same industry could both be higher (lower). Otherwise, the export value (per GDP) in an industry is higher for the country whose import value (per GDP) in that industry is lower. (2) Case III Next, I perform comparative-static analysis for Case III. Note that, as shown in the 22 Note that this is a necessary condition for the values of exports and imports to be both greater or both smaller. 13 previous section, the expressions for VXc/Yc and VMc/Yc derived from the Case III model do not directly include the production share xc, unlike Case I. Therefore, the comparative statics here are with respect to three variables: (i) the number of varieties/firms in country c (nc); (ii) the (common) price of varieties produced in country c (pc); and (iii) the GDP share of country c in the world (yc = Yc/Yw). I first examine the effects of change in the number of varieties in a country in the industry, nc. In the multi-country setting, from Equations (13-III) and (14-III) above; (VX c / Yc ) VQw Tpc1 nc Yw yc [ (n T (n p yc '{nc ' pc1' T 1 c' {nc ' p c ' c c " c , c ' c " c ' c" pc1" )} 1 c" c" )}2 ]0; Tp (nc ' p ) (VM c / Yc ) VQw c ' c 0. 1 nc Yw {nc pc T (nc ' pc1' )}2 1 c (16-III) 1 c' (17-III) c ' c 23 In the two-country setting, from Equations (13’-III) and (14’-III); (VX c / Yc ) VQw (1 yc ) TNpc1 0; nc Yw yc {( N nc ) Tnc pc1 }2 (16’-III) (VM c / Yc ) VQ TNpc1 w 0. nc Yw {nc pc1 T ( N nc )}2 (17’-III) That is, if the prices of products in an industry and GDPs are the same across countries, the country that produces more varieties in an industry has a larger value of exports per GDP in that industry, and on the other hand a smaller value of imports per GDP in the same industry. Next I examine the effects of change in the (common) price of varieties produced in a country, pc. In the multi-country setting, from Equations (13-III) and (14-III); (VX c / Yc ) VQw (1 )Tnc pc pc Yw yc (n T (n p yc '{nc ' pc1' T [ 1 c' {nc ' p c ' c T { (nc ' p 1 c' c " c , c ' c " c ' c" c" pc1" )} 1 c" )}2 ]0; (18-III) )}(1 )nc pc (VM c / Yc ) VQ w c 'c 1 0. pc Yw {nc pc T (nc ' pc1' )}2 (19-III) c ' c ( 1 1 0) 23 In the two-country version of the model, pc denotes the (common) price of varieties produced in country c relative to the price of varieties produced in the rest of the world (the price in the rest of the world is normalized to 1). 14 In the two-country setting, from Equations (13’-III) and (14’-III); (VX c / Yc ) VQw (1 yc ) (1 )Tnc ( N nc ) pc 0; pc Yw yc {( N nc ) Tnc pc1 }2 (18’-III) (VM c / Yc ) VQ (1 )Tnc ( N nc ) pc w 0. pc Yw {nc pc1 T ( N nc )}2 (19’-III) These imply that, if the numbers of product varieties in an industry and GDPs are the same across countries, the country that produces more expensive varieties (in the two-country version, relative to varieties produced in the rest of the world) in an industry has a smaller value of exports per GDP in that industry, and a larger value of imports per GDP in the same industry. I lastly examine the effects of change in a country’s GDP share in the world, yc. In the multi-country setting, from Equations (13-III) and (14-III); (VX c / Yc ) VQ Tn p1 w c 2c yc Yw yc {n c ' c c' 1 c' p yc ' T (nc " pc1" ) } 0; (20-III) c " c ' (VM c / Yc ) 0. yc (21-III) In the two-country setting, from Equations (13’-III) and (14’-III); (VX c / Yc ) VQ 1 Tnc pc1 w 2 0; yc Yw yc {( N nc ) Tnc pc1 } (20’-III) (VM c / Yc ) 0. yc (21’-III) Thus, if the prices of products and the numbers of varieties in an industry are the same across countries, the country that has a larger GDP share in the world has a smaller value of exports per GDP in an industry, but the values of imports per GDP in the industry do not vary across countries. These results are summarized in the following table. nc ↑ pc ↑ yc ↑ VXc/Yc ↑ ↓ ↓ VMc/Yc ↓ ↑ -- According to this summary table, one can derive the following predictions for Case III: VXc/Yc and VMc/Yc could be both higher (lower) when: 15 (i) nc is lower (higher) as yc is lower (higher); (ii) pc is higher (lower) as yc is lower (higher); or (iii) pc is higher (lower) and nc is lower (higher) as yc is lower (higher). Can production share (xc) be a ‘predictor’ in Case III, too? As shown above, in Case III, VXc/Yc and VMc/Yc depend on the three variables: pc, nc, and yc, while they depend only on the two variables, xc and yc, in Case I. However, prices (pc’s) and the numbers of firms/varieties (nc’s) are in general much more difficult to observe in data24 than countries’ shares of production in an industry (xc = VQc/VQw). Then, isn’t it possible to give similar predictions of the ‘co-movement’ of VXc/Yc and VMc/Yc only from the observable variables xc and yc without having the information on prices and the numbers of varieties? My answer is YES: it is possible. The reason is as follows:25 Using the technique of total differentiation, the following conditions are derived for a change in the value of a country’s total production in an industry, VQc: dVQc 0 dnc BQ dpc dVQc 0 dnc BQ dpc ; where BQ ( VQc / pc ). VQc / nc In the same manner, for the values of exports and imports per GDP; VX (VX c / Yc ) / pc d ( c ) ()0 dnc () BX dpc where BX ( ); Yc (VX c / Yc ) / nc d( VM c (VM c / Yc ) / pc ) ()0 dnc () BM dpc where BM ( ). Yc (VM c / Yc ) / nc However, it is derived that: BQ BX BM (1 )nc / pc 0 (for the multi-country version) (1 )nc ( N nc ) / Npc 0 (for the two-country version) In addition, sign(dVQc ) sign(dxc ) 26. 24 The price information may be observable if data are detailed enough. However, the number of firms (nc) is generally unobservable because this variable comes from the assumptions of the model, which does not mean the real number of firms. 25 For the technical details of the following part, see Appendix C-4. 26 See Appendix C-4. 16 Therefore, one can conclude that, if countries’ shares of GDP (yc’s) are fixed; dxc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 dxc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 That is, the country that has a larger production share in an industry has a larger value of exports per GDP, and a smaller value of imports per GDP. One can thus give a prediction of the ‘co-movement’ of VXc/Yc and VMc/Yc from the two observables xc and yc also in this Case III such as in the table below:27 xc ↑ yc ↑ VXc/Yc ↑ ↓ VMc/Yc ↓ -- This prediction of the model is thus the same as that in Case I: if a country’s share in the production value in an industry and its GDP share are both smaller (greater)28, then that country’s exports and imports in values per GDP in the same industry could both be higher (lower). Otherwise, the export value (per GDP) in an industry is higher for the country whose import value (per GDP) in that industry is lower. 5. Data Analysis of Correlation (Case I only) Previous sections have shown that, according to the model, the exports and imports in an industry per GDP, VXci/Yci and VMci/Yci, and their ‘co-movement’ can be predicted by ‘predictors,’ which are each country’s share of production in that industry (xc) and its share of GDP (yc) for Case I; the (common) price of each country’s product varieties in an industry (pc), the number of varieties in each country (nc), and yc for Case III. In this section, I use data to examine how close the predicted measure of the correlation coefficient between VXci/Yci and VMci/Yci (ρi(VXci/Yci, VMci/Yci)) based on these ‘predictors’ is to the direct measure of the correlation coefficient. This is a test of the extent to which the standard model (and its variation) can explain the observed positive correlation between exports and imports within the same industry. Cross-country data on production and 27 Note that, however, for an accurate analysis of the correlation coefficients with data, one still needs the variables pc and nc rather than xc. I discuss this in the next section and propose a procedure for a data analysis in Appendix B. 28 See Footnote 22. 17 international trade in manufacturing industries and GDP are employed. Here I examine only the Case I model, for which all the necessary ‘predictors’ are available in the data, but not the Case III model that requires the prices and the numbers of varieties as ‘predictors’, both of which are not directly observable in the data. Instead, in Appendix B, I propose a possible procedure for a data analysis of Case III without having these unobserved variables. In the following parts of this section, I first re-present the expressions for the correlation coefficient for each of Cases I and III,29 then describe the data, and finally show the results for Case I. (1) Predicted Measure of Correlation Coefficient Let us recall the following expression for the coefficient of correlation across countries, ρi(VXci/Yci, VMci/Yci), which has been derived in Section 3 for each of Case I and Case III. Case I: (15-I): i ( VX ci VM ci 1 yc i 1 yc i i , ) i( xc ,1 xci ) i ( xc , xc ) . Yc Yc yc yc Case III: For the multi-country version, (15-III): VX i VM ci ni p i1 i ( c , ) i ( c c Yc Yc yc { n c ' c i c' yc ' (n i c' pci1' ) c ' c }, ) pci1' T i (nci " pci1" ) nci pci1 T i (nci ' pci1' ) c " c ' c ' c For the two-country version, (15’-III): i i nci pci1 N i nci i VX c VM c i (1 yc ) ( , ) ( , ) Yc Yc yc {N i (1 T i pci1 )nci } T i N i ( pci1 T i )nci i ( (1 yc ) T i nci pci1 nci pci1 , ) yc {N i (1 T i pci1 )nci } T i N i ( pci1 T i )nci The right-hand side of each expression above is the predicted measure of the correlation coefficient. That is, if the model correctly predicts countries’ export and import values per GDP, the cross-country correlation coefficients calculated by the formula above showed 29 Case II has been omitted for further examination, as mentioned in Section 3. 18 equal the actual values of ρi(VXci/Yci, VMci/Yci). Note, however, that the predicted measure of the correlation coefficient for Case III requires the price information for pci, which is not available in the data used for this paper, as well as the number of product varieties, nci, which is unobservable in reality.30 On the other hand, the Case I model requires only the data on countries’ GDPs and production, both of which are available in the dataset described below. (2) Data The data I employ for this paper contain the variables of production, exports, imports, and GDP of various countries. The data on production are from UNIDO (2003). The values of the (gross) output in manufacturing industries in U.S. dollars (USD) were selected. The data on exports and imports are from Feenstra, Lipsey, and Bowen (1997) and Feenstra (2000). The original datasets provide the values of exports and imports for each bilateral pair, and I summed up these bilateral trade values for each origin country (for exports) and for each destination country (for imports). The trade data are in thousands of USD. The GDP data in current USD were taken from the World Bank (2002). There is an issue concerning the industry classifications employed by the data sources. UNIDO’s production dataset categorizes manufacturing industries according to the International Standard Industrial Classification (ISIC: Revision 2) at the 3-digit level, while Feenstra et al.’s trade data are categorized by the Standard International Trade Classification (SITC: Revision 2) at the 4-digit level. I transformed the original trade data into ISIC utilizing the concordance information provided by OECD, which is made available on the web page of Jon Haveman’s Industry Concordances.31 I selected the data for the years of 1970, 1975, 1980, 1985, 1990, and 1995.32 Since availability of the data is different across industries, countries and years, I used for each industry only the countries for which all the variables of production, exports, imports, and GDP are available in order to calculate the across-country correlation coefficient for each industry in each year.33 The sample size thus varies from 22 countries to 77 countries 30 See Footnote 24. URL: http://www.macalester.edu/research/economics/PAGE/HAVEMAN/Trade.Resources/ tradeconcordances.html 32 UNIDO’s dataset on production covers the period of 1963-2001, Feenstra et al.’s (1997) trade data covers 1970-1992, which is supplemented by Feenstra (2000) through the year of 1997. The GDP data from the World Bank (2002) covers the period of 1960-2000. 33 To obtain x i = VQ i/VQ i and y = Y /Y , I needed to calculate the world total value of production in each c c w c c w 31 i industry ( VQW c VQci ) and world GDP ( YW c Yc ). For each summation I include the countries for which the figure to be summed (VQci or Yc) is available even if the countries are not included in the sample for 19 by industry and year. Table 2 shows the number of countries included in the sample for each industry in each year. (3) Result (for Case I) The actual ρi(VXci/Yci, VMci/Yci) and the correlation coefficient predicted by the Case I model are compared for each industry and each period in Table 3. Although the actual correlations are significantly positive in many industries and years, the correlations predicted by the model are mostly small or negative (see Figure 2). Although there are several industries in which the predicted correlations are fairly close to, or even greater than, the actual, the model predicts much lower values for the correlation coefficients than the actual coefficients in most industries (see Figure 3). That is, at least in its Case I version, the model cannot explain the strong positive correlations found in the data. In Section 4, I used comparative statics to show that the necessary condition for VXci/Yci and VMci/Yci to ‘move together’ (i.e., both be higher or lower) is that a country with a larger (smaller) production share (xci) has a larger (smaller) GDP share (yc). Indeed, as shown in Table 4, in the data xci and yc are almost perfectly correlated: that is, in any industry, a country with a larger GDP share almost always has a larger production share.34 However, the more direct term that is included in the expression of VXci/Yci derived from the model is (1-yc)/yc: that is, GDP of ‘the rest of the world’ relative to a country’s own GDP, rather than the country’s own GDP itself. In other words, a country’s exports and imports (in values per GDP) become both smaller (greater) when the country has a larger (smaller) production share but the GDPs of the other countries are smaller (greater) relative to that country’s. This mechanism also applies to Case III. Therefore, I also list the correlations between xci and (1-yc)/yc in Table 4. The signs of the correlations are indeed all negative, as the model requires for positive correlations between exports and imports, but the values are small: they are all between -0.1 and -0.3. It might thus be necessary to have more strongly negative correlations between xci and (1-yc)/yc to predict the strongly positive correlations between exports and imports such as found in the data. This may be so also in Case III.35 each industry/year. Therefore, the sum of xci’s and that of yc’s are not necessarily equal to one. (However, I should note that, when I include only the countries in the sample, in order for the sum of xci’s and that of yc’s to be one, the results presented below do not change.) 34 Moreover, the values of x i and y are almost equal to each other for many countries in most industries and c c years: the two variables are almost on a ’45-degree line’. 35 Appendix B also presents the result of the data analysis for Case III by the proposed ‘indirect’ method. 20 6. Concluding Discussion In this paper I employed the monopolistic competition model of intra-industry trade that is widely used in the trade literature, first with the assumptions of free trade and identical production technologies in an industry; and then with the more relaxed assumptions allowing existence of transport costs and price asymmetry within an industry across countries, in order to see whether such a model can explain the positive correlation between countries’ exports and imports in values per GDP within one industry that is observed in the data for most manufacturing industries. The theoretical model tells, in both its restricted and relaxed versions, that a country’s exports and imports in value per GDP in an industry could both be higher (lower) if the country’s share of the production value in that industry in the world and its GDP share in the world are both smaller (greater); otherwise the export value per GDP in an industry is higher for the country whose import value per GDP in that industry is lower. However, the cross-country correlations between exports and imports predicted by the model from the data on manufacturing production and GDP of various countries showed that the model cannot explain, at least in its restricted form, such positive correlations between exports and imports as observed in the data, while the correlations between the countries’ production shares and GDPs are almost equal to one in most of the industries. For the relaxed case allowing iceberg transport cost and price asymmetry across countries, a data analysis is not straightforward because of data limitation, but inferring from the result of the comparative statics in Section 4, this relaxing of the assumptions may not be enough to make the conventional model capable of explaining the fact.36 What can explain the positive correlation between exports and imports within the same industry? What can be an alternative model to the conventional monopolistic competition model of trade? Harrigan (1994) pointed out that the “love of varieties” assumption for the consumer’s preference, which is represented by a CES utility function over all available varieties, is the driving force of intra-industry trade in this type of model, and similar models with the preference assumption such as Deardorff and Stern (1986) also generate intra-industry trade. In fact, the analysis in this paper relies on the demand side of the conventional model, so that the result here implies that the assumption on the consumers’ preferences may need to be reconsidered. Or, being a little less ambitious, the model with further-relaxed assumptions on transport costs could explain the fact. Hummels 36 Indeed, as shown in Appendix B, the data analysis performed according to the proposed ‘indirect’ method concludes that the model does not explain the positive correlation well even under the relaxed assumptions. 21 and Levinsohn (1995) presented evidence of a significant effect of distance between two countries on the volume of bilateral intra-industry trade. Haveman and Hummels (2004) pointed out that countries may be choosing their sources of import (or trading partners) taking account of bilateral transport costs.37 Therefore, the monopolistic competition model might be able to explain the positive correlation between exports and imports if it incorporates transport cost that differs for each pair of countries. These could be directions for future research. A considerable amount of research has been conducted on the volume of intra-industry trade, including a series of studies on the gravity equation. All of them, however, focused on the total volume of trade without paying much attention to the direction of the trade: how much is exported or imported. By pointing out the ‘fact’ of the positive correlation between exports and imports, which is inconsistent with conventional models based only on comparative advantage, and which may be puzzling according to the conventional monopolistic competition model, this study presents a new issue to be considered when examining what is a plausible theory of international trade. 37 It should be noted that Haveman and Hummels discussed this in the context of the model with imperfect specialization of countries’ production, while the model analyzed in this paper features countries’ perfect specialization in certain varieties of a good. 22 Appendix A Section 4 has performed comparative-static analysis to provide some intuition on the ‘co-movement’ of countries’ export and import values per GDP within the same industry (VXci/Yc and VMci/Yc), but it does not necessarily directly tell what the correlation coefficient of these two variables will be in various circumstances. In this appendix, I present a direct examination of the correlation coefficient. Specifically, limiting the analysis to the Case I model and focusing on five potential circumstances on the patterns of countries production sizes and GDPs, I show whether the model predicts the correlation to be positive or negative in each circumstance.38 Preliminaries Recall Equation (15-I): VX i VM ci 1 yc i 1 yc i i i ( c , ) i( xc ,1 xci ) i ( xc , xc ) . Yc Yc yc yc Then; i ( VX ci VM ci 1 yc i i , ) ()0 i ( xc , xc ) ()0 Yc Yc yc 1 yc i i Cov( xc , xc ) ()0 yc Var ( xci ) ()Cov ( xci i , xc ) . yc …… (*) Therefore, in the following parts of this appendix, with which inequality (*) holds is examined in each case. Five Potential Cases and Expected Correlations Case (i): yc is the same for all countries The first case to be examined is when all countries are in an equal size: i.e., the countries all have the same GDP regardless of their industrial production compositions. In this case, each country’s share of GDP in the world, yc, is equal to 1/N where N(≥2) is the number of countries in the world. Therefore; 38 The detailed derivation or the proof of the expressions or (in)equalities that appear in this appendix are suppressed to avoid lengthiness, but these can be provided by the author upon request. 23 Cov( xci i , xc ) Cov( Nxci , xci ) N Var ( xci ) Var ( xci ) ; yc and thus the model predicts the correlation between export and import per GDP to be negative. More specifically, in this case, the correlation coefficient is: i ( VX ci VM ci 1 yc i i , ) i ( xc , xc ) i (( N 1) xci , xci ) 1 : Yc Yc yc that is, a perfect negative correlation is predicted. Case (ii): xci is constant for all countries The next case is when, in a certain industry, all the countries produce the same value: i.e., the countries’ production sizes are all equal, regardless of the sizes of their economies or GDPs. In this case, each country’s share of production in that industry in the world, xci, is equal to 1/N. Therefore; xi 1 1 Var ( xci ) Var (1/ N ) Cov ( c , xci ) Cov ( , ) 0. yc N yc N That is, in this case, the correlation coefficient is undefined, while the covariance between exports and imports per GDP becomes zero. It is natural to think that the larger a country’s GDP is, the larger the country’s production size in an industry is.39 The following three cases are all in such a circumstance. Case (iii): xci = yc This case is when a country’s share of production in an industry equals its GDP share: i.e., the production size of each country in an industry is proportional to the country’s GDP. In this case; xi Cov( c , xci ) Cov(1, xci ) 0 Var ( xci ) ; yc and thus the correlation between export and import per GDP is expected to be positive. More specifically, the correlation coefficient in this case is: VX i VM ci 1 yc i i i ( c , ) i ( xc , xc ) i (1 yc , yc ) 1 : Yc Yc yc that is, exports and imports are predicted to be perfectly correlated. 39 Indeed, as shown in Table 4, countries’ production shares and GDP shares are almost perfectly correlated in almost all manufacturing industries. 24 Case (iv): xci = (1-a)/N + a∙yc; 0<a<1 This case implies that countries’ GDPs are more diversified than their sizes of production in an industry, or in other words, the degree of countries’ specialization in that industry is not very large.40 In this case; Cov( xci i (1 a )a 1 , xc ) Cov( , yc ) 0 Var ( xci ) ; yc N yc and thus the correlation between export and import per GDP is expected to be positive. Case (v): yc = (1-a)/N + a∙xci; 0<a<1 This case contrasts with Case (iv) above: that is, this case implies that countries’ sizes of production in an industry are more diversified than their GDPs. In other words, the degree of countries’ specialization in that industry is very large.41 In this case; 1 Var ( yc ) a xi (1 a) 1 Cov( c , xci ) Cov( , yc ) 2 yc N a yc Var ( xci ) In fact, the former value becomes greater than the latter value ( Var ( xci ) Cov ( xci / yc , xci ) ) only when the value of a is very close to one. That is, in this case, the correlation between export and import per GDP can be positive when the production of each country in an industry is almost proportional to its GDP, but the correlation is predicted to be negative otherwise.42 Summary: When Will Correlation Be Positive? By summarizing the five cases examined above, the following conclusion is derived: The Case I model predicts that the correlation between countries’ export and import values per GDP within the same industry will be positive when the following 40 Being more accurate, this is a special linear case of such a circumstance. Note that, in this case, it holds that 41 N c xci c yci 1 . N This is a special linear case of such a circumstance. Note that, also in this case, it holds that N c xci c yci 1 . N 42 In fact, in the limit as N rises to infinity, any a such that 0<a<1 cannot generate a positive correlation, which means that the correlation is expected to be always negative. 25 conditions are all satisfied: (i) countries vary both in their shares of production in the industry and in their shares of GDP; (ii) the larger a country’s GDP, the larger the country’s production share in the industry; and (iii) a country’s production share is almost equal to its GDP share, or the dispersion of countries’ GDP shares is greater than that of their production shares (i.e., countries are not very specialized in production). 26 Appendix B As mentioned in Section 5, direct calculation of the predicted correlation coefficient between exports and imports per GDP, ρi(VXci/Yci, VMci/Yci), is not possible for Case III due to lack of data availability on the ‘predictors’: the price of varieties produced in each country (pci), which is not available in the data currently used; and the numbers of varieties in each country (nci), which is unobservable in data. In this appendix, I propose an ‘indirect’ way of computing the predicted correlation coefficient for Case III using only variables that are available in the data but without relying on the unavailable ‘predictors.’ It has turned out, however, that the proposed method needs to rely not only on each country’s production share in an industry (xci) and GDP share (yc), which are the ‘predictors’ for Case I and observable in the data, but also on the actual value of each country’s export in an industry per GDP, VXci/Yc, rather than having the prediction of it. Therefore, the correlation coefficient obtained by this method is between the actual export per GDP (VXci/Yc) and the predicted import per GDP (VMci/Yc). Moreover, it should be noted that this method also requires knowledge of the values of the ‘iceberg’ transport cost, τi, and the elasticity of substitution among varieties, σi, for each industry. I use plausible figures for these values, which are estimated by Hummels (2001)43. In the rest of this appendix, I present the procedure for the computation of the predicted correlation coefficient and the result of the comparison of the predicted correlation to the actual correlation between the observed values of exports and imports per GDP. Preliminaries (The script i for an industry is suppressed below.) Let us recall, for Case III: (10-III): (11-III): VQc nc pc1 b[ 1 c nc p VX c nc pc1 bT { c ' c Yc Yc ' T { }] ; 1 1 1 T (nc ' pc ' ) c ' c nc ' pc ' T (nc " pc " ) c ' c 1 c' nc ' p c " c ' Yc ' }; T (nc" pc1" ) c " c ' 43 Another approach is estimating these values from the data employed for this analysis, rather than borrowing figures from other sources. For example, the values of τi and σi could be determined to minimize the difference between the actual and the predicted correlation coefficient. Such a method might be an interesting exercise, but it does not guarantee that the estimated figures are plausible. 27 (12-III): VM c bYcT (nc ' pc1' ) 1 c nc p c ' c T (nc ' pc1' ) . c ' c Define: zc nc pc1 ; wc 1/( zc T zc ' ) . c ' c Then, Equations (10-III) through (12-III) above can be expressed as follows: VQc zcb[ wcYc T wc 'Yc ' ] ; (i) VX c zcbT wc 'Yc ' VQc zcbwcYc ; (ii) VM c wcYcbT zc ' . (iii) c ' c c ' c c ' c Equation (ii) implies: zcbwcYc VQc VX c wc VQc VX c . bzcYc (ii)’ Then, by the definition of wc, the equation below follows: (1/ wc ) zc T zc ' bzcYc /(VQc VX c ) . (iv) c ' c Equation (iv) holds for any country c. Also note that T z c ' T ( z c ' z c ) . Thus, c ' c c' considering Equation (iv) for two countries c and c’ (c ≠ c’), and subtracting one from another on both sides, yields: zcYc zc 'Yc ' (1 T )( zc zc ' ) b( ). VQc VX c VQc ' VX c ' Define Rc Yc /(VQc VX c ) . Then, since 0 T 1 1, the equation above implies: {1 b b Rc }zc {1 Rc ' }zc ' . 1 T 1 T Here, define the “benchmark” country c = 0 such that z0 n0 p01 1 . Then, the equation above can be expressed as follows: 28 zc {1 b b R0 }/{1 Rc } where R0 Y0 /(VQ0 VX 0 ) . 1 T 1 T (v) Now zc is the (unobservable) variable measuring some combination of a country’s product price and its number of varieties relative to that of the benchmark country44. Proposed Procedure for a Predicted Measure of ρi(VXci/Yci, VMci/Yci) Now I propose steps to calculate the measure of ρi(VXci/Yci, VMci/Yci) predicted from the Case III model. Step 1: Choose a country in the sample as the “benchmark” country c =0 (e.g., USA). Consider z0 =1 for the benchmark country. Step 2: Using the data on production, export, and GDP, and according to Equation (v), calculate zci for every country in the sample other than the benchmark country with the parameter Ti: i.e.; bi bi i R }/{1 Rci } 0 i i 1 T 1 T i where R0 = Y0/(VQ0i - VX0i) Rci = Yc/(VQci – VXci) bi = VQwi/Yw (by market clearing in industry i). Note here that all the variables other than zci are observable. The value for the parameter Ti ≡ (τi)1-σ for each industry, where τi is the “iceberg” transport cost and σi is the CES preference parameter, needs to be estimated separately, or obtained from other sources. I use the estimates by Hummels (2001).45 Step 3: Obtain the import per GDP predicted from the model: i.e., from Equation (14-III) and using the calculated zci ;46 zci {1 44 zci can be considered as an indirect measure of a country’s production size in an industry relative to that of the benchmark country. Indeed, xc / zc 0 , so that a country with larger zci has a larger production share i i in that industry (xci = VQci/VQwi). 45 Hummels (2001) has measured transport costs by the freight rate relative to value of imports for seven countries, and also estimated the CES preference parameter for each commodity using the data on bilateral trade flows among those countries. For the transport cost, since the model of this paper assumes that the transport cost in an industry is identical for every pair of countries in the world, I have taken the trade-value-weighted average of the values that Hummels has estimated for the seven countries. In addition, since the model needs the parameter value for each 3-digit ISIC industry but Hummels’ estimation is for each 2-digit SITC (Rev.2) commodity, I have converted the figures of the transport cost and the CES preference parameter estimated by Hummels into those for each ISIC industry based on concordance information. 46 The predicted export could be calculated from Equation (13-III) and using z i, but it is tautological since c the actual value of export has been used to estimate the value of zci. 29 T i (nci ' pc1' ) i i c i w VM VQ c ' c [ i i ]. Yc Yw nci pc1 T i (nci ' pc1' ) c ' c Step 4: Calculate the correlation coefficient, ρi(VXci/Yci, VMci/Yci), using the actual value of VXci/Yci and the estimated value of VMci/Yci obtained in Step 3 above. Result Table A1 compares the actual and predicted correlation coefficients for each 3-digit ISIC industry in each year for Case III, which is similar to Table 3 for Case I. Although the actual correlation coefficients are positive in most of the industries and years, the predicted correlation coefficients by this method are mostly negative, as shown in Figure A1 (similar histograms to Figure 2 for Case I). This result seems to imply that, even in Case III that allows transport cost and price asymmetry across countries, the standard monopolistic competition model of intra-industry trade does not explain the observed positive correlation between export and import per GDP within the same industry. Figure A2, which is similar to Figure 3 for Case I, shows how the correlation predicted by this method over- or under-estimates the actual correlation. What is striking is that the distribution of the estimation gaps is not continuous: there are two clusters and the group of the industry-year observations for which the correlations are over-estimated is clearly separated from the group of the under-estimated observations. About a quarter of all the industry-year cells (41 out of 168: 28 industries in 6 year periods) obtain the predicted correlation over-estimating the actual, but I do not find a systematic pattern between the over-estimation and certain industries or years. This estimation method seems to be very sensitive to data, and it may be caused by the discontinuity of the estimated zci 47. In fact, the discontinuity of the estimated zci seems to imply that the model’s assumption of identical and constant expenditure shares in each industry is inconsistent with reality. The method of estimating zci presented here relies on this assumption, while the expenditure share on a certain industry seems to vary across countries, which may result in inappropriate estimation of the figure of zci in some cases. 47 30 References Deardorff, Alan V. (1980), “The General Validity of the Law of Comparative Advantage”, Journal of Political Economy, 88(5), pp.941-957. Deardorff, Alan V., and Stern, Robert M. (1986), The Michigan Model of World Production and Trade, MIT Press, Cambridge, MA. Feenstra, Robert C.; Lipsey, Robert E., and Bowen, Harry P. (1997), “World Trade Flows, 1970-1992, with Production and Tariff Data”, NBER Working Paper No.5910. Feenstra, Robert C. (2000), “World Trade Flows, 1980-1997”, Center for International Data, Institute of Governmental Affairs, University of California, Davis. Finger, J. M. (1975), “Trade Overlap and Intra-Industry Trade”, Economic Inquiry, 13(4), pp.581-589. Harrigan, James (1994), “Scale Economies and The Volume of Trade”, Review of Economics and Statistics, 76(2), pp.321-328. Haveman, Jon, and Hummels, David (2004), “Alternative Hypotheses and The Volume of Trade: The Gravity Equation and The Extent of Specialization”, Canadian Journal of Economics, 37(1), pp.199-218. Helpman, Elhanan, and Krugman, Paul R. (1985), Market Structure and Foreign Trade, MIT Press, Cambridge, MA. Hummels, David (2001), “Toward a Geography of Trade Costs”, manuscript. Hummels, David, and Levinsohn, James (1995), “Monopolistic Competition and International Trade: Reconsidering The Evidence”, Quarterly Journal of Economics, 110, pp.799-836. Krugman, Paul R. (1979), “Increasing Returns, Monopolistic Competition, and International Trade”, Journal of International Economics, 9, pp.469-479. 31 Krugman, Paul R. (1980), “Scale Economies, Product Differentiation, and the Patterns of Trade”, American Economic Review, 70(5), pp.950-959. Romalis, John (2004), “Factor Proportions and the Structure of Commodity Trade”, American Economic Review, 94(1), pp.67-97. UNIDO: United Nations Industrial Development Organization (2003), Industrial Statistics Database at the 3-digit level of ISIC Code (Rev.2) (INDSTAT3 2003 ISIC Rev.2). World Bank (2002), World Development Indicators 2002. 32 Table 1: Correlations between Volumes of Exports and Imports per GDP across Countries Industry* 311 (Food products) 313 (Beverages) 314 (Tobacco) 321 (Textiles) 322 (Wearing apparel) 323 (Leather products) 324 (Footwear) 331 (Wood products) 332 (Furniture) 341 (Paper and products) 342 (Printing and publishing) 351 (Industrial chemicals) 352 (Other chemicals) 353 (Petroleum refineries) 354 (Misc. petroleum & coal products) 355 (Rubber products) 356 (Plastic products) 361 (Pottery, china, earthenware) 362 (Glass and products) 369 (Other non-metallic mineral products) 371 (Iron and steel) 372 (Non-ferrous metals) 381 (Fabricated metal products) 382 (Non-electric machinery) 383 (Electric machinery) 384 (Transport equipment) 385 (Professional & scientific equipment) 390 (Other manufactured products) 1970 .183 .646 .431 .084 .432 .378 .162 -.006 .418 -.166 .555 .397 .320 .127 .508 .982 .652 .052 .100 .053 .247 -.046 .167 .161 .322 .077 .340 .507 1975 .304 .041 .498 .494 .208 .136 .124 -.241 .598 -.136 .312 .234 .381 .346 .235 -.008 .584 .107 -.020 -.022 -.013 .371 .191 .036 .785 .005 .242 .555 Year 1980 1985 .085 .202 .058 .384 .513 .670 .424 .123 .375 .470 .336 .232 .215 .202 -.189 -.122 .329 .483 -.091 -.041 .459 .570 .381 .567 .492 .494 -.032 -.042 .382 .121 .084 .013 .533 .762 -.030 -.094 .163 .238 .143 .084 -.189 -.056 -.092 .001 .242 .278 .141 .297 .543 .951 -.083 -.041 .056 .330 .697 .187 1990 .314 .381 .248 .131 .077 .184 -.019 -.136 .099 .009 .443 .359 .231 .086 -.029 .047 .749 -.013 .118 .049 -.107 -.060 .433 .137 .939 -.007 .090 .281 1995 .184 .269 .185 .412 .165 .235 -.092 -.207 .329 -.056 .495 .286 .494 -.023 .051 -.016 .618 -.050 .105 -.005 .243 -.075 .343 .057 .963 .185 .665 .694 * Industries are classified by the International Standard Industry Classification (ISIC: Revision 2) at the 3-digit level. 33 Table 2: Sample Sizes: Number of Countries included in Each Sample Year Industry 311 (Food products) 313 (Beverages) 314 (Tobacco) 321 (Textiles) 322 (Wearing apparel) 323 (Leather products) 324 (Footwear) 331 (Wood products) 332 (Furniture) 341 (Paper and products) 342 (Printing and publishing) 351 (Industrial chemicals) 352 (Other chemicals) 353 (Petroleum refineries) 354 (Misc. petroleum & coal products) 355 (Rubber products) 356 (Plastic products) 361 (Pottery, china, earthenware) 362 (Glass and products) 369 (Other non-metallic mineral products) 371 (Iron and steel) 372 (Non-ferrous metals) 381 (Fabricated metal products) 382 (Non-electric machinery) 383 (Electric machinery) 384 (Transport equipment) 385 (Professional & scientific equipment) 390 (Other manufactured products) 1970 60 59 56 59 57 59 54 59 62 55 61 57 63 43 30 51 51 42 50 55 43 43 61 47 57 54 39 55 1975 71 77 69 68 66 66 61 70 75 70 74 62 72 51 40 63 67 51 60 72 50 50 72 58 68 65 40 62 1980 68 75 68 68 63 63 58 71 72 68 74 69 73 49 34 61 64 51 55 73 55 47 77 63 67 67 37 61 1985 71 69 62 64 54 54 54 71 66 68 68 64 69 47 33 61 68 49 53 63 47 43 75 56 63 61 36 54 1990 76 68 63 70 60 53 53 66 69 72 74 60 71 45 34 58 65 51 56 65 50 42 71 50 65 67 33 61 Note: The names of the countries included in each sample can be provided upon request. 34 1995 63 67 56 61 50 47 50 56 63 66 69 51 64 44 22 57 61 49 55 68 50 40 66 48 59 52 33 46 Table 3: Correlations between VXci/Yc and VMci/Yc: Actual and Predicted (Case I Model) Industry 311 313 314 321 322 323 324 331 332 341 342 351 352 353 354 355 356 361 362 369 371 372 381 382 383 384 385 390 Note: 1970 Act. Prd. .183 .108 .646 .118 .431 .153 .084 -.009 .432 -.024 .378 -.009 .162 .043 -.006 -.136 .418 -.024 -.166 -.122 .555 -.199 .397 -.221 .320 -.071 .127 -.013 .508 -.082 .982 .004 .652 -.188 .052 -.276 .100 -.114 .053 .008 .247 -.228 -.046 -.069 .167 -.185 .161 -.266 .322 -.278 .077 -.249 .340 -.164 .507 -.072 1975 Act. Prd. .304 .092 .041 .116 .498 .092 .494 .027 .208 -.034 .136 -.022 .124 -.038 -.241 -.172 .598 -.139 -.136 -.188 .312 -.271 .234 -.205 .381 -.069 .346 .042 .235 -.182 -.008 -.030 .584 -.175 .107 -.366 -.020 -.143 -.022 -.043 -.013 -.247 .371 -.096 .191 -.212 .036 -.335 .785 -.213 .005 -.310 .242 -.165 .555 -.104 Year 1980 1985 Act. Prd. Act. Prd. .085 .099 .202 .142 .058 .134 .384 .125 .513 -.021 .670 .012 .424 -.190 .123 -.109 .375 -.020 .470 .071 .336 -.179 .232 -.038 .215 .079 .202 .064 -.189 -.160 -.122 -.044 .329 -.122 .483 -.019 -.091 -.190 -.041 -.098 .459 -.293 .570 -.192 .381 -.285 .567 -.168 .492 -.101 .494 .030 -.032 .022 -.042 .127 .382 -.062 .121 -.094 .084 -.081 .013 -.040 .533 -.207 .762 -.129 -.030 -.516 -.094 -.389 .163 -.150 .238 -.056 .143 -.078 .084 .027 -.189 -.309 -.056 -.197 -.092 -.038 .001 .036 .242 -.230 .278 -.106 .141 -.444 .297 -.398 .543 -.314 .951 -.193 -.083 -.452 -.041 -.339 .056 -.296 .330 -.279 .697 -.106 .187 -.070 i Act. = Actual correlation (direct measure from the data) of ( VX ci VM ci , ); Yc Yc Prd. = Predicted correlation from the Case I version of the model: i ( VX ci VM ci 1 yc i i , ) i ( xc , xc ) . Yc Yc yc 35 1990 Act. Prd. .314 .203 .381 .140 .248 .060 .131 -.195 .077 .081 .184 -.149 -.019 -.028 -.136 -.020 .099 -.068 .009 -.130 .443 -.260 .359 -.113 .231 .014 .086 .185 -.029 -.034 .047 -.068 .749 -.182 -.013 -.186 .118 -.072 .049 .034 -.107 -.100 -.060 .023 .433 -.195 .137 -.479 .939 -.212 -.007 -.465 .090 -.512 .281 -.052 1995 Act. Prd. .184 .219 .269 .184 .185 .095 .412 -.084 .165 .053 .235 -.370 -.092 -.124 -.207 -.078 .329 -.142 -.056 -.121 .495 -.258 .286 -.024 .494 .019 -.023 .144 .051 .009 -.016 -.024 .618 -.199 -.050 -.053 .105 -.082 -.005 -.071 .243 -.139 -.075 .012 .343 -.159 .057 -.523 .963 -.182 .185 -.373 .665 -.520 .694 -.034 Table 4: Correlations of Country’s Production Share (xci) with Own GDP Share (yc) or with Rest of the World’s GDP Share Relative to Own GDP ((1-yc)/yc). Year Industry 311 313 314 321 322 323 324 331 332 341 342 351 352 353 354 355 356 361 362 369 371 372 381 382 383 384 385 390 Note: 1970 1975 1980 1985 1990 1995 O R O R O R O R O R O R .996 .981 .988 .962 .995 .965 .983 .936 .994 .990 .996 .982 .999 .988 .948 .997 .947 .770 .993 .974 .934 .984 .994 .987 .970 .995 .994 .991 -.149 -.156 -.168 -.177 -.128 -.197 -.157 -.164 -.150 -.143 -.128 -.145 -.137 -.224 -.300 -.141 -.156 -.213 -.132 -.171 -.258 -.219 -.137 -.130 -.133 -.133 -.104 -.131 .989 .970 .945 .958 .992 .938 .918 .931 .970 .981 .984 .986 .995 .986 .884 .994 .952 .686 .994 .962 .942 .986 .992 .990 .972 .990 .982 .967 -.125 -.149 -.206 -.227 -.183 -.248 -.237 -.197 -.212 -.174 -.113 -.191 -.195 -.218 -.158 -.114 -.190 -.220 -.171 -.211 -.281 -.243 -.182 -.167 -.182 -.167 -.113 -.160 .990 .969 .871 .900 .982 .850 .899 .938 .973 .977 .981 .985 .991 .990 .930 .966 .945 .617 .988 .931 .906 .976 .992 .987 .960 .987 .974 .960 -.159 -.228 -.232 -.164 -.196 -.266 -.256 -.214 -.151 -.183 -.119 -.200 -.213 -.232 -.242 -.234 -.205 -.214 -.193 -.160 -.220 -.251 -.102 -.178 -.179 -.178 -.124 -.116 .992 .982 .928 .907 .986 .773 .809 .978 .989 .991 .990 .975 .994 .995 .974 .970 .953 .629 .979 .946 .829 .980 .988 .973 .926 .986 .988 .951 -.189 -.190 -.212 -.223 -.168 -.276 -.253 -.169 -.170 -.157 -.136 -.193 -.176 -.217 -.118 -.196 -.167 -.194 -.191 -.207 -.275 -.272 -.091 -.156 -.150 -.144 -.105 -.161 .994 .972 .943 .892 .969 .747 .683 .975 .975 .987 .982 .987 .989 .979 .962 .947 .961 .634 .969 .919 .849 .981 .957 .940 .856 .973 .946 .945 -.108 -.143 -.151 -.266 -.235 -.276 -.249 -.124 -.093 -.110 -.076 -.198 -.091 -.235 -.195 -.231 -.186 -.243 -.241 -.240 -.246 -.254 -.145 -.214 -.109 -.103 -.060 -.113 .994 .954 .936 .922 .963 .577 .755 .984 .960 .980 .993 .986 .997 .964 .957 .975 .984 .800 .981 .905 .911 .988 .970 .963 .928 .988 .970 .952 -.192 -.225 -.249 -.197 -.195 -.190 -.272 -.168 -.249 -.214 -.145 -.176 -.174 -.244 -.186 -.217 -.195 -.258 -.243 -.194 -.234 -.232 -.164 -.201 -.151 -.209 -.167 -.144 O: Correlation of a country’s share in production value (xci) with that country’s own GDP share (yc); R: Correlation of xci with the share in GDP of the rest of the world (all countries other than country c) relative to country c’s own GDP share ((1-yc)/yc); where: xci VQci / VQci ; VQci = the value of production of country c in industry i; c yc Yc / Yc ; Yci = GDP of country c. c 36 Table A1: Correlations between VXci/Yc and VMci/Yc: Actual and Predicted (Case III Model) Industry 311 313 314 321 322 323 324 331 332 341 342 351 352 353 354 355 356 361 362 369 371 372 381 382 383 384 385 390 Note: 1970 Act. Prd. .183 -.047 .646 -.211 .431 -.064 .084 .070 .432 -.192 .378 .059 .162 .057 -.006 -.130 .418 -.394 -.166 -.685 .555 -.395 .397 -.187 .320 -.099 .127 -.105 .508 .025 .982 -.842 .652 -.135 .052 -.272 .100 -.242 .053 -.140 .247 -.600 -.046 -.039 .167 -.464 .161 -.576 .322 -.603 .077 -.618 .340 -.888 .507 .107 1975 Act. Prd. .304 -.031 .041 -.021 .498 -.003 .494 -.396 .208 -.069 .136 .184 .124 .179 -.241 -.281 .598 -.499 -.136 -.776 .312 -.342 .234 -.104 .381 -.210 .346 .204 .235 -.165 -.008 -.009 .584 -.609 .107 -.339 -.020 -.323 -.022 -.116 -.013 -.519 .371 -.318 .191 -.462 .036 -.680 .785 -.370 .005 -.684 .242 -.753 .555 .163 Year 1980 1985 Act. Prd. Act. Prd. .085 -.002 .202 -.098 .058 -.065 .384 .042 .513 .070 .670 .108 .424 -.064 .123 -.089 .375 -.097 .470 .122 .336 -.559 .232 .064 .215 .316 .202 -.496 -.189 -.350 -.122 -.149 .329 -.182 .483 .018 -.091 -.784 -.041 -.517 .459 -.150 .570 -.126 .381 -.196 .567 -.126 .492 .110 .494 .105 -.032 .117 -.042 .028 .382 -.050 .121 .073 .084 .074 .013 .174 .533 -.339 .762 -.321 -.030 -.366 -.094 -.304 .163 -.202 .238 -.088 .143 -.103 .084 -.021 -.189 -.381 -.056 -.283 -.092 -.240 .001 -.381 .242 -.379 .278 -.210 .141 -.204 .297 -.219 .543 -.331 .951 -.156 -.083 -.586 -.041 -.199 .056 -.507 .330 -.600 .697 -.037 .187 -.202 i Act. = Actual correlation (direct measure from the data) of ( 1990 Act. Prd. .314 .070 .381 .067 .248 .042 .131 -.041 .077 .211 .184 -.097 -.019 .161 -.136 .090 .099 -.414 .009 -.254 .443 -.194 .359 -.024 .231 .091 .086 .093 -.029 -.385 .047 -.673 .749 .138 -.013 -.087 .118 -.138 .049 -.001 -.107 -.253 -.060 -.691 .433 -.211 .137 -.328 .939 -.047 -.007 -.336 .090 -.470 .281 -.209 1995 Act. Prd. .184 .049 .269 .016 .185 .056 .412 -.403 .165 .114 .235 -.411 -.092 .749 -.207 -.359 .329 -.083 -.056 -.643 .495 -.253 .286 -.094 .494 .124 -.023 .019 .051 .075 -.016 -.749 .618 -.142 -.050 .048 .105 -.383 -.005 -.373 .243 -.104 -.075 -.026 .343 -.021 .057 -.420 .963 -.416 .185 -.221 .665 -.305 .694 -.231 VX ci VM ci , ); Yc Yc Prd. = Correlation between the actual VXci/Yc and the predicted VMci/Yc (based on Case III version of the model). 37 Figure 1: Cross-Country Correlations between Exports/GDP and Imports/GDP: Histograms (Number of Sectors) at Each Level of Aggregation 4-digit SITC (Rev.2) #sectors = 786; #countries = 95(max); Mean=0.219; median=0.132 100 Frequency 0 0 50 20 Frequency 40 150 60 200 3-digit SITC (Rev.2) #sectors = 239; #countries = 97(max); Mean=0.232; median=0.183 -1 -.5 0 corr(VX/Y, VM/Y) .5 1 -1 .5 1 6-digit HS (1992) #sectors = 5,035; #countries = 31(max); Mean=0.209; median=0.116 800 400 400 Frequency 600 300 200 0 200 100 0 Frequency 0 corr(VX/Y, VM/Y) 1000 5-digit SITC (Rev.2) #sectors = 1,464; #countries = 94(max); Mean=0.223; median=0.132 -.5 -1 -.5 0 corr(VX/Y, VM/Y) .5 1 -1 38 -.5 0 corr(VX/Y, VM/Y) .5 1 Figure 2: Histogram of ρ(VXci/Yc, VMci/Yc): Actual and Predicted (Case I Model) (Frequency = numbers of 3-digit ISIC industries) Predicted: mean = -0.109, median = -0.097 30 20 0 0 10 10 Frequency Frequency 20 40 50 30 Actual: mean = 0.233, median = 0.189 -1 -.5 0 corr(VX/Y, VM/Y): Actual .5 1 -1 -.5 0 corr(VX/Y, VM/Y): Case I Pred .5 Figure 3: Histogram of Gap (Predicted Correlation - Actual Correlation) (Case I Model) (Frequency = numbers of 3-digit ISIC industries) 15 10 5 0 Frequency 20 25 mean = -0.342, median = -0.326 -1.5 -1 -.5 rho_Gap: Case I Pred - Actual 39 0 .5 Figure A1: Histogram of ρ(VXci/Yc, VMci/Yc): Actual and Predicted (Case III Model) (Frequency = numbers of 3-digit ISIC industries) 20 0 10 Frequency 0 10 Frequency 20 30 Predicted: mean = -0.194, median = -0.153 30 Actual: mean = 0.233, median = 0.189 -1 -.5 0 corr(VX/Y, VM/Y): Actual .5 1 -1 -.5 0 .5 corr(VX/Y, VM/Y): Case III Pred 1 Figure A2: Histogram of Gap (Predicted Correlation - Actual Correlation) (Case III Model) (Frequency = numbers of 3-digit ISIC industries) 10 5 0 Frequency 15 20 mean = -0.745, median = -1.080 -2 -1 0 rho_Gap: Case III Pred - Actual 40 1 Appendix C-1 This appendix shows how the sets of equations in Section 3 are derived for each of the three cases (Case I, II and III). Derivation of the common equations (Equations (1) through (9)) to all the cases is as explained in Section 3. Case I Derivation of Equations (8-I) through (12-I) is shown in Section 3. For Equation (13-I), from Equation (11-I); VX c 1 n b c ( Yc ' ) . Yc Yc N c ' c Define yc ≡ Yc/Yw. Also note that Yc ' Yw Yc . Hence; c ' c VX c n (1 y c ) nc 1 b c (1 y c )Yw b . Yc y cYw N yc N Define xc ≡ VQc/VQw. In this Case I, since all firms (in an industry) in the world charge the same price and produce the same amount, VQc = ncpq and VQw = Npq, so that xc = nc/N. Therefore; VX c (1 y c ) bxc . Yc yc Finally, by market clearing in an industry, the value of the total world production must equal the world total expenditure. That is; VQw bYw b VQw / Yw . Hence; VX c VQw (1 y c ) xc . Yc Yw yc (13-I) For Equation (14-I), from Equation (12-I); VM c ( N nc ) n 1 bYc b(1 c ) . Yc Yc N N Hence, since xc = nc/N and b = VQw/Yw as shown above; VX c VQw (1 xc ) . Yc Yw (14-I) Case II Equations (8-II) through (14-II) for Case II are derived as follows: In this case, since the prices of all the varieties charged by firms (excluding trade cost) are the same, p(ω) = p for all ω. Thus, from Equation (8); i Gc [ n c p 1 {nc ' (p) 1 }] 1 1 c ' c p[nc 1 n c ' c c' ] 1 1 1 p[nc 1 ( N nc )]1 Thus, by defining T ≡ τ1-σ; 1 1 Gc p[nc T ( N nc )] . (8-II) Equation (9-II) follows by substituting p(ω) = p and Equation (8-II) into Equation (9). Equations (10-II) and (11-II) are then self-explaining. The value of imports of a country is equal to the expenditure of consumers in that country to all the foreign varieties in an industry. Since p(ω) = p for all ω in this model; p VM c {nc ' bYc ( )1 } . Gc c ' c Then, by substituting T = τ1-σ and Equation (8-II) for Gc into the expression above; nc ' ( N nc )bTYc VM c bYc T { } . (12-II) nc T ( N nc ) c ' c n c T ( N n c ) The last term of the equation is because nc ' N nc . c ' c Now for the values of exports and imports per GDP: from Equations (11-II) and (12-II); VX c nc b Yc ' T { }; Yc Yc c ' c nc ' T ( N nc ' ) VM c ( N nc )bT . Yc nc T ( N nc ) Note that Yc = ycYw for all c. Also, since xc = nc/N in this case, nc = xcN for all c. Therefore; VX c x N y c 'Yw x yc' c bT { } c bT { }; Yc y cYw yc c ' c x c ' N TN (1 x c ' ) c ' c x c ' T (1 x c ' ) VM c N (1 xc )bT (1 xc )bT . Yc xc N TN (1 xc ) xc T (1 xc ) Thus, Equations (13-II) and (14-II) follow by substituting b = VQw/Yw into the two equations above. Case III Equations (8-III) through (14-III) for Case III are derived as follows: In this case, since there is a single price for all the varieties produced in each country, p(ω) = pc for all ω produced in c. Thus, Equation (8) becomes: ii Gc [nc p 1 c {nc ' (p c ' ) 1 }] 1 1 . c ' c Then, Equation (8-III) follows by defining T ≡ τ1-σ. Thus, each firm in country c earns the same revenue (or produces the same value of output), which is, by substituting Equation (8-III) into Equation (9): pc qc bYc ( bYc pc 1 p ) {bYc ' ( c )1 } Gc Gc ' c ' c {nc pc1 . pc1 1 pc1 b [ { Y } ] c' T (nc ' pc1' )} {nc ' pc1' T (nc" pc1" )} c ' c c ' c c" c ' 1-σ Equation (9-III) then follows by substituting T = τ into the expression above. Equation (10-III) also follows by substituting (9-III) into the expression of the value of the total output in country c: VQc = ncpcqc. The expression for the value of total exports from country c, Equation (11-III), is obvious since the second term of RHS of Equation (10-III) is the value exported abroad, while the first term is the value consumed domestically. On the other hand, since the value of imports is equal to the expenditure of consumers in country c to all the foreign varieties in an industry; pc ' VM c {nc ' bYc ( c ' c Gc bYc 1 {nc ' ( pc ' )1 } )1 } c ' c Gc1 . Then, Equation (12-III) is obtained by substituting T = τ1-σ and Equation (8-III) for Gc into the expression above. Now the expressions for the values of exports and imports per GDP are derived as follows. From Equations (11-III) and (12-III); VX c nc p c1 Yc ' bT { } 1 Yc Yc T (nc" p c1" ) c ' c n c ' p c ' c " c ' 1 c nc p y c 'Yw nc p c1 yc' bT { } bT { } 1 1 1 y c Yw yc T ( nc" p c" ) T (nc" p c1" ) c ' c nc ' p c ' c ' c nc ' p c ' c " c ' c " c ' bT (nc ' p ) VM c c ' c . Yc nc pc1 T (nc ' pc1' ) 1 c' c ' c The second equality in the former expression (VXc/Yc) is because Yc = ycYw for all c. Then, Equations (13-III) and (14-III) are derived by substituting b = VQw/Yw into the two expressions above. Finally, consider the two-country version of the model for this Case III. In the two-country setting, there is only one foreign country for country c (home), which is ‘the rest of iii the world.’ Having the home country and ‘the rest of the world’ denoted by c and –c, respectively, Equations (13-III) and (14-III) are modified as follows: VX c VQw nc p c1 y c [ T{ }] ; 1 Yc Yw yc n c p c Tn c p c1 VM c VQw Tn c p 1c [ ]. Yc Yw nc p c1 Tn c p 1c By normalizing the price of varieties produced in ‘the rest of the world’ to one (i.e., p-c = 1) and substituting n-c = N - nc and y-c = 1 - yc, the expressions above become Equations (13’-III) and (14’-III), respectively. iv Appendix C-2 This appendix shows derivation of the expressions of the correlation between VXci/Yci and VMci/Yci (ρi(VXci/Yci, VMci/Yci)) for the three cases of the model that are presented in Section 3. This appendix also examines whether the possibility that the correlation becomes positive or negative can be rejected theoretically or not. Since the correlation is across countries within an industry, the script i for an industry is suppressed below. Case I: From Equations (13-I) and (14-I) in Section 3; VX VM c VQ (1 yc ) VQw ( c , ) ( w xc , (1 xc )) . Yc Yc Yw yc Yw Since VQw/Yw is constant (both VQw and Yw do not vary across countries); VX VM c (1 yc ) (1 yc ) ( c , ) ( xc , (1 xc )) ( xc , xc ) a; Yc Yc yc yc and Equation (15-I) follows. Let us check whether this correlation could be either positive or negative: first, note that: (1 yc ) (1 yc ) ( xc , xc ) 0 Cov( xc , xc ) . yc yc Thus one can see the sign of the correlation of the two variables from the sign of their covariance. By definition; (1 yc ) (1 yc ) 2 (1 yc ) Cov( xc , xc ) E[ xc ] E[ xc ] E[ xc ] . yc yc yc Since 0<xc ≡VQc/VQw <1, 0< xc2 < xc <1 for any c. Thus; (1 yc ) 2 (1 yc ) E[ xc ] E[ xc ] . yc yc However, since 0< E[xc] <1; (1 yc ) (1 yc ) E[ xc ] E[ xc ] E[ xc ] . yc yc Then, it is ambiguous which of the first term or the second term of RHS of the expression of the covariance above is greater, and thus the covariance could be positive or negative, depending on a Note that, for random variables x and y and constants a and b; ρ(ax, by) = ρ(x, y). Also; ρ(x, 1-y) = -ρ(x, y). v the data on xc and yc. Therefore, one cannot reject either possibility that the correlation between VXc/Yc and VMc/Yc is positive or negative. Case II: From Equations (13-II-2) and (14-II) in Section 3; VX VM c VQ xc VQ 1 xc ( c , ) ( w T ( M 1) , w T ). Yc Yc Yw xc T (1 xc ) Yw xc T (1 xc ) Since VQw/Yw, T, and M-1 are all constant; VX VM c xc T (1 xc ) ( c , ) ( , ) Yc Yc xc T (1 xc ) xc T (1 xc ) ( xc xc , )b xc T (1 xc ) xc T (1 xc ) = -1. The second equality is because (15-II) T (1 xc ) xc . 1 xc T (1 xc ) xc T (1 xc ) Case III: Here I first show the two-country version, and then the multi-country version. (Two-country Version) From Equations (13’-III) and (14’-III) in Section 3; VX VM c VQ (1 yc ) Tnc pc1 VQ T ( N nc ) ( c , ) ( w , w ) 1 1 Yc Yc Yw yc {( N nc ) Tnc pc } Yw {nc pc T ( N nc )} VQ (1 yc ) Tnc pc1 VQ T ( N nc ) ( w , w ) 1 Yw yc {N (1 Tpc )nc } Yw {TN ( pc1 T )nc } Since VQw/Yw is constant; VX c VM c (1 yc ) Tnc pc1 T ( N nc ) , ) ( , ). (i): ( 1 Yc Yc yc {N (1 Tpc )nc } {TN ( pc1 T )nc } Consider the second argument of the equation (i) above. Note that: T ( N nc ) nc pc1 1 . TN ( pc1 T )nc TN ( pc1 T )nc Thus; b See the previous footnote a. vi . (1 yc ) Tnc pc1 T ( N nc ) ( , ) 1 yc {N (1 Tpc )nc } {TN ( pc1 T )nc } (ii): ( (1 yc ) Tnc pc1 nc pc1 , ) yc {N (1 Tpc1 )nc } TN ( pc1 T )nc On the other hand; (1 yc ) Tnc pc1 T ( N nc ) ( , ) 1 yc {N (1 Tpc )nc } {TN ( pc1 T )nc } (iii): (1 yc ) nc pc1 N nc ( , ) 1 yc {N (1 Tpc )nc } TN ( pc1 T )nc since T ≡ τ1-σ is constant. Therefore, by combining (i), (ii), and (iii) above, Equation (15’-III) is derived as follows: VX VM c (1 yc ) nc pc1 N nc ( c , ) ( , ) 1 Yc Yc yc {N (1 Tpc )nc } TN ( pc1 T )nc (15’-III) (1 yc ) Tnc pc1 nc pc1 ( , ) yc {N (1 Tpc1 )nc } TN ( pc1 T )nc Now check whether this correlation could be either positive or negative. Since: (1 yc ) Tnc pc1 nc pc1 ( , )0 yc {N (1 Tpc1 )nc } TN ( pc1 T )nc (1 y c ) Tn c p c1 nc p c1 Cov( , ) 0; y c {N (1 Tp c1 )nc } TN ( p c1 T )nc the sign of the correlation of the two variables is the same as that of their covariance. By definition; Cov( E[ E[ (1 yc ) Tnc pc1 nc pc1 , ) yc {N (1 Tpc1 )nc } TN ( pc1 T )nc (1 yc ) Tnc pc1 nc pc1 ] yc {N (1 Tpc1 )nc } {TN ( pc1 T )nc } (1 yc ) Tnc pc1 nc pc1 ] E [ ] yc {N (1 Tpc1 )nc } TN ( pc1 T )nc Note that: nc pc1 nc pc1 1 for any nc, pc TN ( pc1 T )nc nc pc1 ( N nc )T since the term (N – nc)T in the denominator above is positive. Hence; (1 yc ) Tnc pc1 nc pc1 (1 yc ) Tnc pc1 c yc {N (1 Tpc1 )nc } {TN ( pc1 T )nc } yc {N (1 Tpc1 )nc } vii (1 yc ) Tnc pc1 nc pc1 (1 yc ) Tnc pc1 E[ ] E[ ]. yc {N (1 Tpc1 )nc } {TN ( pc1 T )nc } yc {N (1 Tpc1 )nc } nc pc1 ] 1; However, since E[ TN ( pc1 T )nc (1 yc ) Tnc pc1 nc pc1 (1 yc ) Tnc pc1 E[ ] E[ ] E[ ]. yc {N (1 Tpc1 )nc } TN ( pc1 T )nc yc {N (1 Tpc1 )nc } It is thus ambiguous which of the first term or the second term of RHS of the expression of the covariance above is greater, so that one cannot reject either possibility that the correlation between VXc/Yc and VMc/Yc is positive or negative. (Multi-country Version) From Equations (13-III) and (14-III); T (nc ' pc1' ) VX c VM c VQw nc pc1 yc ' VQw c ' c ( , ) ( T { }, ) Si 1 i 1 1 Yc Yc Yw yc n p T ( n p ) Y { n p T i (nc ' pc1' )} c ' c c" c" c' c' w c c c " c ' c ' c i nce VQw/Yw and T are both constant; VX VM c n p1 ( c , ) ( c c Yc Yc yc { n c ' c (n c' yc ' pc1' ) c ' c }, ) 1 i 1 1 p T ( n p ) n p T i (nc ' pc1' ) c" c" c' c' c c c " c ' c ' c and thus Equation (15-III) follows. This correlation could be either positive or negative, depending on the data for xc and yc, by the same token as the two-country version. viii Appendix C-3 In the Case II model, all firms (or varieties) in an industry charge the same price and produce the same quantity across countries in equilibrium, by assumption. When such an equilibrium holds, the following conditions that are presented in Section 3 must be satisfied among xc ≡ VQc/VQw, yc ≡ Yc/Yw, and T ≡ τ1-σ where τ >1 is the ‘iceberg’ transport cost. yc x T (1 xc ) c ' c (II-A) c yc ' xc ' T (1 xc ' ) In this appendix, I derive the condition above for each version. From Equation (9-II), the value of output from a single firm in country c is: Yc Yc ' pqc b[ T { }] . nc T ( N nc ) c ' c nc ' T ( N nc ' ) In the same manner, a typical firm in country c’≠ c produces: Yc ' Yc" pqc ' b[ T { }] . nc ' T ( N nc ' ) c " c ' nc " T ( N nc " ) Define zc (D1) (D2) bYc bYc . Then, (D1) and (D2) above are expressed as nc T ( N nc ) (1 T )nc TN follows: pqc z c T ( z c ' z c ) ; (D1’) c' pqc ' zc ' T ( zc zc ' ) . (D2’) c In equilibrium, pqc pqc ' ( pq) for any pairs of c and c’; c ≠ c’. Thus, subtracting each side of (D1’) from that of (D2’) yields: 0 ( zc ' zc ) T ( zc zc ' ) (1 T )( zc ' zc ) . In order for this equality to hold, it must be that zc = zc’ (since T ≡ τ1-σ ≠ 1 by assumptionc). Hence; bYc bYc ' . (D3) zc z c ' (1 T )nc TN (1 T )nc ' TN b in both sides cancel each other. Moreover, by substituting nc = xcN and Yc = ycYw; ycYw yc 'Yw (D3) . (D3’) (1 T ) xc N TN (1 T ) xc ' N TN Yw/N in both sides cancel, so that: c On the other hand, if T = 1 (i.e., free international trade: Case I), zc does not have to equal zc’, so that the presented condition does not have to hold for equal-price equal-quantity equilibrium in Case I. ix (D3’) yc yc ' . (1 T ) xc T (1 T ) xc ' T Or, this is equivalent to: yc x T (1 xc ) . c yc ' xc ' T (1 xc ' ) This is the condition presented above, and this must hold for any pairs of countries. What does this condition Imply? Note that the codition (II-A) implies: y yc yc ' ( c 1) xc xc ' yc ' That is, if prices and quantities of all varieties in an industry are equal in equilibrium, then the country that has larger income (GDP) must have a larger world production share in that industry. x Appendix C-4 This appendix is to provide technical details of the discussion in Section 4 about why the observable xc = VQc/VQw can be the ‘predictor’ of the ‘co-movement’ of VXc/Yc and VMc/Yc without having pc and nc. I first show for the multi-country version, and then show for the two-country version. Multi-country Setting (Value of production: VQc) By totally differentiating (with respect to the variables for the home country c) Equation (10-III); VQc VQc VQc dVQc ( ) dnc ( ) dpc ( ) dYc . nc pc Yc Fixing the country’s GDP: i.e., setting dYc = 0; VQc VQc dVQc ( ) dnc ( ) dpc . nc pc Thus; dVQc 0 ( VQc VQc ) dnc ( ) dpc 0 dnc BQ dpc nc pc VQc VQc dVQc 0 ( ) dnc ( ) dpc 0 dnc BQ dpc nc pc by defining BQ ( ; (E-1) VQc / pc ). VQc / nc From Equation (10-III); (n p )} 0 T (n p )} Yc (nc ' pc1' ) Yc '{nc ' pc1' T VQc c ' c (1 )bTnc pc [ [ 1 pc {nc pc T (nc ' pc1' )}2 c ' c {nc ' pc1' c ' c c" c " c , c ' c " c ' c" 1 c" 1 c" ( 1 1 0) ; Yc (nc ' p 1 c' ) (n p )} 0. T (n p )} 1 c' Yc '{nc ' p VQc c ' c bTpc1 [ [ 1 nc {nc pc T (nc ' pc1' )}2 c ' c {nc ' pc1' c ' c T c" c " c , c ' c " c ' c" 1 c" 1 c" 2 Therefore; BQ ( VQc / pc (1 )bTnc pc (1 )Tnc ) 0. VQc / nc bTpc1 pc (Value of exports per GDP: VXc/Yc) By the same token as for VQc; xi (E-2) 2 d( VX c (VX c / Yc ) (VX c / Yc ) )0( ) dnc ( ) dpc 0 dnc BX dpc Yc nc pc VX (VX c / Yc ) (VX c / Yc ) d( c ) 0 ( ) dnc ( ) dpc 0 dnc BX dpc Yc nc pc by defining BX ( (E-3) (VX c / Yc ) / pc ). (VX c / Yc ) / nc Note that: (17-III): (VX c / Yc ) VQW (1 )Tnc p pc YW yc c (VX c / Yc ) VQW Tpc1 (15-III): nc YW yc [ c ' c [ c ' c 1 c' {nc ' p {nc ' pc1' T c " c , c ' c " c ' c " c , c ' c " c ' (n T (n p 1 c' yc '{nc ' p (n T (n p yc '{nc ' pc1' T c" c" c" c" pc1" )} 1 c" )}2 0; pc1" )} 1 c" )}2 0. Therefore; VQw (VX c / Yc ) (1 )Tnc pc pc Yw yc (1 )Tnc BX 0 . (E-4) VQw 1 p (VX c / Yc ) c Tpc nc Yw yc (Value of imports per GDP: VMc/Yc) By the same token; VM c (VM c / Yc ) (VM c / Yc ) d( )0( ) dnc ( ) dpc 0 dnc BM dpc Yc nc pc VM c (VM c / Yc ) (VM c / Yc ) d( )0( ) dnc ( ) dpc 0 dnc BM dpc Yc nc pc by defining BM ( (VM c / Yc ) / pc ) . Notice the sign of the last inequality in each of the two (VM c / Yc ) / nc lines above: this is because (VM c / Yc ) / nc 0 as shown below. Note that: T (1 )nc pc { (nc ' pc1' )} (19-III): (VM c / Yc ) VQ c ' c w 0; 1 pc Yw {nc pc T (nc ' pc1' )}2 c ' c Tpc1 (nc ' pc1' ) (17-III): (E-5) (VM c / Yc ) VQ c ' c w 0. 1 nc Yw {nc pc T (nc ' pc1' )}2 c ' c xii Therefore; VQ (VM c / Yc ) w (1 )Tnc pc pc Yw (1 )Tnc BM 0 . (E-6) VQw 1 p (VM c / Yc ) c Tpc nc Yw From (E-2), (E-4), and (E-6) above; (1 )Tnc BQ BX BM 0. pc Thus, from (E-1), (E-3), and (E-5) above; (1 )Tnc dnc dpc dVQc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 ; pc dnc (1 )Tnc dpc dVQc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 . pc Finally, since: xc VQc / VQW VQc /(VQc ' VQc ) ; c ' c then: dxc / dVQc (VQW VQc ) / VQW2 0 ; which implies sign(dxc ) sign(dVQc ) . Hence, in conclusion; dxc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 dxc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 . Two-country Setting (Value of production: VQc) By the same token as for the multi-country setting shown above; VQc VQc dVQc 0 ( ) dnc ( ) dpc 0 dnc BQ dpc nc pc VQc VQc dVQc 0 ( ) dnc ( ) dpc 0 dnc BQ dpc nc pc by defining BQ ( VQc / pc ). VQc / nc xiii ; By modifying Equation (10-III) for the two-country version; Yc (YW Yc ) VQc nc pc1 b[ T ]. 1 nc pc T ( N nc ) ( N nc ) Tnc pc1 Thus; VQc Yc YW Yc (1 )bTnc ( N nc ) pc [ ]0 1 2 pc {nc pc T ( N nc )} {( N nc ) Tnc pc1 }2 VQc Yc YW Yc bTNpc1 [ ]0 1 2 nc {nc pc T ( N nc )} {( N nc ) Tnc pc1 }2 Therefore; BQ ( VQc / pc (1 )bTnc ( N nc ) pc (1 )nc ( N nc ) ) 0. VQc / nc bTNpc1 Npc For the value of exports per GDP (VXc/Yc) and that of imports (VMc/Yc), all the steps are the same as those for the multi-country setting shown above, except using Equations (16’-III) through (19’-III) instead of (16-III) through (19-III), respectively. Then, one obtains the following: (1 )nc ( N nc ) BX BM BQ 0 ; Npc which implies: dnc (1 )nc ( N nc ) dpc dVQc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 ; Npc dnc (1 )nc ( N nc ) dpc dVQc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 . Npc Furthermore, since sign(dxc ) sign(dVQc ) as shown above, one reaches the same conclusion as that for the multi-country setting: i.e.; dxc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 dxc 0 d (VX c / Yc ) 0 d (VM c / Yc ) 0 . xiv