Journal of International Economics 54 (2001) 75–96 www.elsevier.nl / locate / econbase The nature and growth of vertical specialization in world trade David Hummels a , Jun Ishii b , Kei-Mu Yi c , * a Department of Economics, Krannert School of Management, Purdue Univesity, West Lafayette, USA b Department of Economics, Stanford University, Stanford, CA 94305, USA c Federal Reserve Bank of New York, International Research, 33 Liberty St., New York, NY 10045, USA Received 24 August 1999; received in revised form 30 December 1999; accepted 24 April 2000 Abstract Dramatic changes are occurring in the nature of international trade. Production processes increasingly involve a sequential, vertical trading chain stretching across many countries, with each country specializing in particular stages of a good’s production sequence. We document a key aspect of these vertical linkages — the use of imported inputs in producing goods that are exported — which we call vertical specialization. Using input–output tables from 10 OECD and four emerging market countries we calculate that vertical specialization accounts for 21% of these countries’ exports, and grew almost 30% between 1970 and 1990. We also find that growth in vertical specialization accounts for 30% of the growth in these countries’ exports. 2001 Elsevier Science B.V. All rights reserved. Keywords: Growth of world trade; Globalization; Vertical specialization JEL classification: F1 1. Introduction It is widely acknowledged that in the last several decades the nature of international trade has changed dramatically. One of the most important changes *Corresponding author. Tel.: 11-212-720-6386; fax: 11-212-720-6831. E-mail address: kei-mu.yi@ny.frb.org (K.-M. Yi). 0022-1996 / 01 / $ – see front matter 2001 Elsevier Science B.V. All rights reserved. PII: S0022-1996( 00 )00093-3 76 D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 involves the increasing interconnectedness of production processes in a vertical trading chain that stretches across many countries, with each country specializing in particular stages of a good’s production sequence. This phenomenon — we follow Balassa and Findlay in calling it vertical specialization — has been studied quite extensively by trade economists. It has also been labeled quite extensively: ‘‘slicing up the value chain’’, ‘‘outsourcing’’, ‘‘disintegration of production’’, ‘‘fragmentation’’, ‘‘multi-stage production’’, and ‘‘intra-product specialization’’.1 Many models have been developed to study the impact of increased vertical specialization on factor prices, production and trade patterns, and welfare. Given the proliferation of models, one might presume that there is extensive empirical documentation of the increase in vertical specialization. Surprisingly, this is not so. Other than case studies and some anecdotes there is little systematic evidence quantifying the nature and growth of vertical specialization.2 Data on intermediate goods trade could be examined, but this would require relying on rather arbitrary classifications of goods into intermediate and final. Moreover, some existing classifications indicate that the intermediates share of total trade has been steadily declining since 1970.3 Instead, this paper provides evidence on a narrower concept of vertical specialization, involving those imported goods that are used as inputs to produce a country’s export goods. This narrow focus emphasizes the twin ideas that the production sequence of a good involves at least two countries, and that, during this sequence, the good-in-process crosses at least two international borders. The latter idea highlights the sequential production and back-and-forth aspect that case study evidence suggests characterizes an increasing amount of international trade.4 Japan now exports raw steel to Mexico, where the steel is stamped and pressed and then exported to the U.S., where it is manufactured into farm equipment, much of which is then exported again. This concept allows us to develop measures of vertical specialization that can be easily taken to the data. Our primary measure (VS) measures the value of 1 See, for example, Krugman (1995), Feenstra and Hanson (1996, 1997), Feenstra (1998), Deardorff (1998), Jones and Kierzkowski (1997), Dixit and Grossman (1982), and Arndt (1997). Balassa (1967) and Findlay (1978) were the first papers, to the best of our knowledge, to note this phenomenon. Sanyal (1983) also used ‘‘vertical specialization’’. 2 Most of the existing systematic evidence focuses on documenting several countries’ trends in outsourcing, usually defined to be the imported input shares of gross output or of material inputs. See, for example, Feenstra and Hanson (1996), Feenstra (1998), Campa and Goldberg (1997), Lawrence (1994), Slaughter (2000), and Berman et al. (1994). 3 Using annual trade data and the United Nations Broad Economic Categories classification scheme, we find that, for the OECD, both the intermediate goods share of imports and of exports declined steadily from about 1970 to 1992. (These calculations exclude oil.) Measuring the intermediate share of imports using the OECD Input–Output Database (OECD, 1995) also reveals a declining share during this period. However, Yeats (1998) shows that parts and components trade, a subset of intermediate goods trade, has grown as a share of total trade. 4 See Ishii and Yi (1997) and Hummels et al. (1998), for example. D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 77 imported inputs embodied in goods that are exported. Our main data source is input–output tables, which provide industry-level data on imported inputs, gross output, and exports, the data that are needed to calculate our measures. We use the OECD Input–Output Database, which covers 10 OECD countries, input–output tables from Ireland, Korea, and Taiwan, and data from Mexico’s maquiladoras. These countries account for more than three-fifths of world trade. Our calculations show that, as of 1990, the (export-weighted average) VS share of exports in these 14 countries was 0.21, up almost 30% from 0.165 in 1970. Also, the growth in VS exports accounts for 30% or more of the growth in overall exports between 1970 and 1990. In Section 2, we present our vertical specialization concepts and measures. In Section 3, we present our calculations of vertical specialization levels and trends. Section 4 presents the results of growth and cross-country accounting decompositions. Section 5 briefly discusses potential causes of the growth of vertical specialization. 2. Vertical specialization: concepts and measurement 2.1. Concepts To construct a good measure of vertical specialization, we first need to sharpen the concepts. The key idea behind fragmentation, outsourcing, slicing-the-valuechain, etc. is that countries increasingly link sequentially to produce goods. Our approach focuses on one feature of this sequential linkage: imported intermediate goods are used by a country to make goods or goods-in-process which are themselves exported to another country. This feature highlights the multipleborder-crossing, back-and-forth aspect of trade that much of the anecdotal and case study evidence suggests has risen dramatically. More formally, vertical specialization occurs when: A. a good is produced in two or more sequential stages, B. two or more countries provide value-added during the production of the good, C. at least one country must use imported inputs in its stage of the production process, and some of the resulting output must be exported. Note that vertical specialization involves both an import side and an export side. On the import side, vertical specialization is essentially a subset of intermediate goods trade. While all intermediate goods trade is consistent with (A) and (B), only the subset of intermediate goods imports that become embodied in exported goods is consistent with the third condition. On the export side, vertical specialization can involve either intermediate goods or final goods. Fig. 1 illustrates an example of a vertical specialization chain involving three 78 D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 Fig. 1. Vertical specialization. countries. Country 1 produces an intermediate good and exports it to Country 2. Country 2 combines the imported intermediates with capital and labor (valueadded), and domestically produced intermediate inputs to produce a final good (gross output). Finally, Country 2 exports some of the final good to Country 3. 2.2. Measurement We focus on one way in which a country can participate in a vertical specialization chain, when the country uses imported inputs to produce an exported good. For example, Japan uses imported oil to produce petrochemicals, some of which are exported. For country k and good or sector i, we define VS as follows: S D imported intermediates VS ki 5 ]]]]]]] ? exports gross output (1) D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 S D exports 5 ]]]] ? imported intermediates. gross output 79 (1a) For Country 2 in Fig. 1, then, VS 2i 5 (A /(D 1 E))*E 5 (E /(D 1 E))*A. As noted above, this is a subset of imported intermediates, A. Simply put, VS is the imported input content of exports, or equivalently, foreign value-added embodied in exports. The first term in Eq. (1) is the share of imported inputs into gross production. Multiplying this ratio by the amount that is exported provides a dollar value for the imported input content of exports. If Country 2 uses no imported inputs, or if it does not export its output, VS 2i 5 0.5 Note that the VS 2i /X2i , the VS share of exports, is A /(D 1 E). For good or sector i, then, the VS share of exports is equivalent to the imported input share of gross output. VS for country k is simply the sum of VS across all i, VS k 5 o i VS ki . We find it useful to calculate O O VS k i VS ki VS share of total exports ; ]] 5 ]]], Xk i Xki (1b) where X denotes exports. Of course, we can also calculate the VS share of total imports. These two shares will differ when trade is not balanced. The VS share of total exports can also be expressed as O O O VS k i VS ki i (VS ki /Xki )*Xki VS share of total exports ; ]] 5 ]]] 5 ]]]]] Xk X X i ki X VS ]] . OFS] X DS X DG ki 5 i k ki O i ki (1c) ki In other words, the VS share for a country k is an export-weighted average of the sector VS export shares. We noted above that the sector VS export shares are equivalent to the sector imported input shares of gross output. Eq. (1c) shows that the aggregate VS share, VS k /Xk , and the aggregate imported input share of gross output (IIGO) are not equivalent, because the latter is expressed as a gross output-weighed average of the sector imported input shares of gross output. The aggregate VS share of exports will be larger than the aggregate imported input / gross output ratio whenever the high VS share sectors also tend to be the high export sectors, i.e. when there is a positive correlation between sector VS shares and high export / output ratios. To empirically implement our VS measure, we would ideally use data on the production process and direction of trade flow for every stage of each good that is 5 We do not count re-exports, i.e. border-crossings that are merely in transit shipments. Much of Hong Kong’s exports are, in fact, re-exports of Chinese goods going through Hong Kong’s ports on their way to the U.S. and other countries. D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 80 traded. These data are impossible to obtain except on a case-by-case basis. Instead, we rely primarily on input–output tables. Our input–output (I–O) tables include sector-level data on inputs (distinguishing foreign and domestic sources), valueadded, gross output, and exports. A key advantage of using I–O tables is we avoid the arbitrariness of classification schemes that divide goods into ‘‘intermediate’’ and other categories. How should tires and engines, flour, and motherboards be classified? In some cases they are intermediate goods (when firms buy them to make cars, bread, and computers), and in some cases they are final goods (when households buy them).6 Input–output tables do not have this problem, because they classify the use (as an input into another sector’s production or as final demand) of each sector’s output. Another advantage is that the tables allow us to explore sector variation in VS; for example, we can examine how the auto sector differs from the textile sector on both the export and the imported input side. This facilitates a straightforward calculation of (1), (1a) or (1b) for each industry and for the country as a whole. In matrix notation, the formula for VS as a share of total exports for country k (the equivalent of (1b)) is VS share of total exports ;VS k /Xk 5 uAM X /Xk , (2) M where u is a 1 3 n vector of 1’s, A is the n 3 n imported coefficient matrix, X is an n 3 1 vector of exports, n is the number of sectors, and Xk is the sum of exports across the n sectors. Element a ij of AM denotes the imported inputs from sector i used to produce one unit (expressed in any common currency) of sector j’s output. Another attractive feature of I–O tables is they allow us to calculate the value of imported inputs used indirectly in production of an exported good. That is, imported inputs may be used in one sector, whose outputs are employed in a second, then a third, and eventually embodied in an export good. Imported inputs are allowed to circulate through several stages of the domestic economy before ‘‘exiting’’ as an export. In terms of Eqs. (1) and (1a) above, the imported intermediates term would include all direct and indirect (embodied in domestic inputs) imported inputs. The more general way to compute VS as a share of total exports for country k with these tables is VS share of total exports ;VS k /Xk 5 uAM [I 2 AD ] 21 X /Xk , (3) where u is a 1 3 n vector of 1’s, AM is the n 3 n imported coefficient matrix, I is the identity matrix, AD is the n 3 n domestic coefficient matrix, X is an n 3 1 6 Also, Sanyal and Jones (1982) make the important point that all goods trade can be thought of as intermediate goods trade, because even so-called final goods have marketing and distribution services added to them before they are sold. In our framework, we abstract from these services, and focus on the physical production process. D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 81 vector of exports, Xk is total country exports, and n is the number of sectors.7 [I 2 AD ] 21 is the term that captures allowing the imported input to be embodied in a domestic output at the 2nd, 3rd, 4th, . . . stage before it becomes embodied in the good that it is exported. It is the matrix analogue of an infinite geometric sum. We use (3) as our main measure of VS. Note that this formulation allows the goods to circulate through all sectors of the economy, including the service sectors.8 The relatively aggregate sector data from the input–output tables can lead to biases in calculating the true level of VS. If, within a sector, there is a positive (negative) correlation between exports and the imported inputs / gross output ratio, VS calculations involving the sector-level data will be downward (upward) biased. Suppose, for example, that a sector produces just two goods. One good uses imported intermediate inputs but is not exported. The other good uses no imported inputs but is exported. In this case, actual VS would be zero, yet at the sector level we would calculate a positive value. On the other hand, suppose that the first good relies heavily on imported intermediate inputs and is heavily exported, and the second good uses no imported inputs and is not exported. Then, at the sector level, we would underestimate VS. Indeed, Korean input–output table data for 1995 reveals a slight downward bias using aggregate data.9 However, in general, we cannot ascertain which case is more likely.10 The reader may have noticed that there is another way that a country can participate in a vertical specialization chain. This occurs when the country exports goods that are used as inputs into another country’s production of export goods. For example, Japan produces electronic components, most of which are exported to South East Asian countries, where they are used as inputs to produce TVs and VCRs, most of which are then exported to countries like the U.S. We call the value 7 Development economists have used (3), which they also call the import content of exports. See, for example, Chenery et al. (1987). However, their interest was traditionally with balance of payment issues, not the extent of vertical specialization, or its implications for trade growth. 8 While our measures of vertical specialization imply that a good-in-process crosses at least two borders, we cannot determine the average number of border-crossings. The number of border-crossings matter because it gives the ‘‘multiplier’’ trade effect that results from a given change in trade barriers. Also, we are only able to compute VS with merchandise trade; hence, we ignore trade in R&D or multinational headquarters services. This may be an important channel through which vertical specialization operates. 9 The 28-sector, 77-sector, and 168-sector input–output tables yielded VS shares of exports of 0.326, 0.338, and 0.342, respectively. 10 Our empirical measure of VS is based on imported intermediate goods. For many countries, imported capital goods constitute a large fraction of imports; logically, one can think of these goods as a type of intermediate good in the sense that rental services from the capital become embodied in the goods that are produced from it. An extended measure of VS would include the imported capital service content of exports. Due to space constraints, we do not include these calculations here. Our results from the OECD Input–Output tables show that the VSK share of exports is about 0.03–0.04. Details are available from the authors. D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 82 of exports that are embodied in a second country’s export goods VS1. In terms of Fig. 1, Country 1 has VS1; it is equal to A*E /(D 1 E). We do not calculate this measure, however, because of data constraints. VS1 is more difficult to measure than VS, as it requires matching bilateral trade flow data to the input–output relations.11 3. Growth of vertical specialization 3.1. Data sources The primary source of our input–output tables is the OECD Input–Output Database, which contains tables for 10 countries — the G-7 nations, plus Australia, Denmark, and the Netherlands — for several years between 1968 and 1990.12 The major advantage of this data set is that it provides a consistent set of tables to facilitate comparisons across countries and over time. The 10 countries account for about two-thirds of world GDP and more than 55% of world trade. The tables divide output into 35 sectors, including 24 goods-producing sectors, of which 22 are manufacturing. The concentration on manufacturing sectors is important because they increasingly dominate world trade.13 For each country, we focus on the goods sectors. We also employ input–output tables for Ireland (1964, 1975, 1990), Korea (1963, 1970, 1990, 1993, 1995), and Taiwan (1981, 1994). These were obtained primarily through these countries’ national statistical agencies or Central Banks. For Mexico, we use data on maquiladoras, including imported inputs, gross output and exports. Mexico’s maquiladoras are foreign-owned production plants that 11 In an earlier version of this paper (see http: / / www.ny.frb.org / rmaghome / staff rp / sr72.html), we ] perform a broad, relatively crude VS1 calculation for the OECD database countries, and a more narrow calculation involving U.S. trade with Canada and Mexico. We find that the VS1 share of exports is roughly 0.04–0.05 for the OECD countries; the more narrow calculation for the U.S. shows that, as of 1997, the VS1 share is 0.082. Details are available from the authors. 12 For most of the countries in this database, the imported inputs tables are imputed following the standard import proportionality assumption. The imputation occurs at a much more detailed level of disaggregation than is in the tables. For example, in the U.S. and Japan the imputations are performed over a range of 500 goods. For Germany and Denmark, the range is 2000 goods. In general, the OECD finds that higher levels of disaggregation are associated with smaller (downward) biases in classifying imports as intermediate goods. Hummels et al. (1998), which is an outgrowth of IY and this paper, also looks at this database. 13 In 1970, manufacturing accounted for about 60% of world merchandise trade; in 1994, it accounted for about 75% (UNCTAD, 1997). D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 83 complete processing or secondary assembly of imported components explicitly for export.14 3.2. VS levels and growth: OECD Input–Output Database Hereafter, we will use ‘‘VS export share of total exports’’, ‘‘VS share of exports’’, and ‘‘VS share’’ interchangeably. Fig. 2 presents our calculations.15 The overall VS shares vary widely across countries. The U.S., Japan and Australia have VS shares of about 0.05–0.10, while Canada, Denmark, and the Netherlands have VS shares around 0.30–0.35. We compute the sample correlation between the VS share of exports and GDP; it is 20.65 in the final year, indicating that smaller countries have higher VS shares.16 For every country but Japan, the VS share grew between the initial and final year of the sample. In Australia, Canada, France, the U.K., and the U.S., VS grew 25% or more.17 Aggregating across all 10 countries — by calculating the export-weighted average of each country’s VS share — we calculate the VS share in the initial and final years to be 0.162 and 0.198, respectively.18 This is an increase of about 22% over a 20-year period. If we use the initial year export weights for both the initial and final year, the VS share in the final year would be 0.202. This implies that more than 100% of the increase in the aggregate OECD 14 Maquiladora plants benefit from Mexican laws that exempt from Mexican tariffs parts and materials imported by Mexico for use in maquiladoras. Also, U.S. components of maquiladora-made goods exported back to the U.S. are exempt from U.S. tariffs. See Hummels et al. (1998) for more background information on the maquiladoras. 15 More detailed sector-level VS share numbers are presented in Tables 1 and 2 of an on-line version of the paper (http: / / www.ny.frb.org / rmaghome / staff rp / sr72.html). As with the aggregate VS shares, ] the sector VS shares vary widely across countries. There is also wide variation across sectors within a country. Trends in the sector VS shares mirror trends in the aggregate shares; between the first and last year, about three-quarters of the sector VS shares increase. 16 The GDP data are from IMF International Financial Statistics. The correlation between population and VS share is 20.69. 17 We also calculated the VS share of imports, which tended to be lower than the VS share of exports for countries whose imports exceed exports, such as the U.S. As with the VS share of exports, the VS share of imports grew over time in all countries but Japan. Countries that had very high (positive or negative) growth in VS shares of exports tended to have less high growth in their VS shares of imports, and vice versa. 18 In computing the OECD aggregate VS shares, we take 1970 to be the initial year and 1990 as the final year. For countries without input–output tables for that year we use the VS shares for the closest year if it was within 2 years of 1970 (or 1990). If the closest year to 1970 (or 1990) was more than 2 years away, as is the case for Germany, we used a linear extrapolation of the available data to get an estimate for 1970 (or 1990). For Italy we have only 1 year of data, 1985. We assume that Italy’s VS share is 0.2 in 1970, which is similar to the VS shares in France, Germany, and the U.K. around that time. 84 D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 Fig. 2. VS exports as a share of total merchandise exports: OECD countries. D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 85 VS share is accounted for by increases in every country’s VS share; countries with low VS shares — such as the U.S. — grew in export importance during this period.19 Fig. 2 also shows that many of the countries, including Japan, Denmark, France, the Netherlands, and the United Kingdom, experience sharp increases in their VS shares in the 1970s, followed by partially offsetting declines in the shares at some point after 1980. One possible explanation is oil prices, whose movements mirrored the VS shares during this period. The value of imported oil also rose and then fell. Changes in the value of imported oil could, ceteris paribus, change VS shares. To investigate this possibility, we recalculated our VS shares assuming that all oil (and other mining and quarrying) inputs are domestic; that is imported oil inputs are zero. The results are shown in the gray lines with circles in Fig. 2. Once oil is excluded, all five above-mentioned countries experience smoother and more monotonic increases in their VS shares over time.20 3.3. VS levels and growth: Ireland, Korea, Taiwan, and Mexico The small countries in our OECD sample have the largest VS shares. Because the rest of the world consists primarily of small countries, we would expect to find relatively large VS shares for countries outside our OECD database. To verify this, we calculate VS for Ireland, Korea, Taiwan, and Mexico. The calculations for these countries are presented in Fig. 3. For Ireland, Korea and Taiwan, the VS shares are quite high, (around one-third, similar to Netherlands in the OECD Fig. 3. Vertical specialization: Ireland, Korea, Taiwan, and Mexico. 19 20 To do this calculation, we use the ‘‘within’’ and ‘‘between’’ decomposition presented later in (5). We thank one of the referees for suggesting this exercise. 86 D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 database) but have not exhibited the large increases over time that Canada, for example, has experienced.21 Nevertheless, these three countries have contributed to growth in the world VS share. This is because the world VS share is an export-weighted average of the country VS shares, and these three countries’ share of world exports increased by a factor of 9 between 1963 and 1993. For Mexico we employ annual data from 1979 to 1997 on maquiladora imports, gross production, and exports, instead of input–output tables. To facilitate crosscountry aggregation, we calculate VS as a share of total, ‘‘maquiladoras1nonmaquiladoras’’, merchandise exports. Our calculations indicate that the VS share of exports has increased significantly, reflecting the growing importance of maquiladoras in Mexico’s trade.22 Fig. 3 shows that, between 1979 and 1984, VS as a share of total Mexican merchandise exports was about 0.1. After 1984, this share rose steadily and rapidly; as of 1997, it was 0.32 or $35 billion.23 3.4. VS levels and growth: calculations for the ‘‘ World’’ Above, we calculated VS shares for each country in our OECD database, as well as for several other countries. We now calculate the VS share for our entire 14-country sample, which accounted for 63% of world exports in 1990. The results are listed in column 1 of Table 1. We find that the VS share was 0.165 in 1970 and 21 This might seem surprising for Ireland given the surge in multinational activity there during the last two decades. According to Barry and Bradley (1997, p. 1798) ‘‘almost 60% of gross output and 45% of employment in manufacturing is in foreign-owned export-oriented firms.’’ However, there are three reasons why Ireland’s VS share did not increase much between 1964 and 1990. First, the presence of multinationals appears to have reduced the amount of ‘‘circulation’’ of imported inputs through the domestic economy. Calculations of (2), rather than (3), show that the VS share increases considerably between 1964 and 1990, from 0.17 to 0.27. Second, while manufacturing has undergone a structural shift from light activities such as clothing and textiles to the heavier activities dominated by multinationals, such as chemicals, machinery, and equipment, both light and heavy activities have relied greatly on imported inputs. Finally, much of the growth in manufacturing multinational activity occurred in the 1990s, after the final year of data. 22 Throughout this period, Mexican value-added has accounted for about 20–30% of gross production, with imported inputs accounting for the remainder. Hence, VS as a share of maquiladora exports has fluctuated between 0.7 and 0.8. 23 This is a lower bound for total VS because there is also VS originating from non-maquiladora channels. For example, the U.S. International Trade Commission estimates that non-maquiladora activity in 1996 led to an additional $8 billion of Mexican exports to the U.S. Assuming that the imported input content for these exports is about 80%, then VS exports for this activity is roughly $6.4 billion. Our 80% number is an informal estimate by Ralph Watkins, chief of the Minerals, Metals, Machinery, and Miscellaneous Manufacturing division of the USITC. Added to our maquiladora VS, the total VS share for 1996 would be 0.365. Assuming that an equal amount of this type of non-maquiladora activity occurred the following year, then Mexico’s VS share for 1997 would be about 0.4. D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 87 Table 1 World VS exports a Set of countries 1 Countries’ share of world exp 1970 1990 Growth in VS share Contribution of VS exports to growth in exports / GDP 0.60 0.63 2 VS share of exports 0.165 0.211 Countries’ share of world exp 0.74 0.82 VS share of exports 0.180 0.236 28.4% 31.3% 30.1% 32.5% a Source: (merchandise) export data for 1970 and 1990 is obtained from UNCTAD, Handbook of International Trade and Development Statistics. Set 1: includes 10 countries in OECD database1 Ireland, Korea, Mexico, and Taiwan. Set 2: includes column 1 countries1rest of Europe1other emerging East Asia. Other emerging East Asia includes China, Hong Kong, Indonesia, Malaysia, Singapore, and Thailand. 0.211 in 1990, a 28% increase over two decades.24 When we calculate the VS share in 1990 using the initial-year export weights, we obtain 0.204. This implies that 86% of the increase in the overall 14-country VS share is due to increases in the country VS shares, and 14% is due to increased export shares of the high VS share countries, namely Korea, Taiwan and Ireland.25 24 We assume the VS share for Mexico and Taiwan in 1970 is 0.1 and 0.3, respectively. The Mexico number is based on its 1979 VS share (0.1) and on Grunwald and Flamm (1985) p. 148. The Taiwan estimate is based on Korea’s 1970 VS share. We also find that, for 1980, the VS share was 0.212. Excluding oil, the VS shares are 0.136 in 1970, 0.160 in 1980, and 0.189 in 1990, suggesting a more steady growth path. 25 In a second calculation, we add the rest of Europe and other East Asian emerging market countries (China, Hong Kong, Indonesia, Malaysia, Singapore, and Thailand) to our 14 countries. The set of countries in this expanded sample accounted for 82% of world exports in 1990. We assume that the VS share for the rest of Europe, which includes the smaller countries such as Austria, Belgium, Ireland, Luxembourg, Portugal, Spain, and Switzerland, have VS shares slightly lower than Ireland’s: 0.25 in 1970 and 0.3 in 1990. For the East Asian emerging market countries, inclusive of Korea and Taiwan, we assume that the VS share is 0.25 in 1970 and 0.35 in 1990. We think it is likely that the other East Asian emerging market countries have experienced even greater growth in their VS shares than Korea and Taiwan did. For example, the two largest exporters in this group (not including re-exports from Hong Kong and Singapore) are China and Malaysia. After China liberalized its tariffs on imported inputs intended for production of export goods in the early 1980s, China experienced very rapid growth in its VS. Data from Naughton (1996) indicate that China’s VS share of imports was about 0.25 in 1988; this rose to 0.41 by 1994. We do not have data on Malaysia’s VS shares, but Malaysia’s export share of value-added in manufacturing rose from 0.52 to 1.91 between 1970 and 1990 (source: Bank Negara Malaysia, Quarterly Economic Bulletin, various issues). It is likely that the near quadrupling of the export share reflects a considerable increase in VS. Based on these assumptions, then, column 2 of Table 1 shows that the world VS share of exports rises from 0.18 in 1970 to 0.236 in 1990, a 30% increase. The above calculations focused only on VS. In a previous version of this paper, we also provide calculations for an extended measure of VS that included imported capital inputs embodied in exports and for VS1 for the 10 countries in the OECD database. With these additional numbers, we get an overall estimate of the world vertical specialization share of overall exports in 1990 of about 0.3. 88 D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 3.5. Linking VS growth to trade growth We use growth accounting to address the sources of growth in overall exports (as a share of gross output). We decompose exports into VS exports and other exports (this is equivalent to decomposing exports into foreign value-added embodied in exports and domestic value-added embodied in exports): 26 Xk,t VS k,t (Xk,t 2VS k,t ) D]] 5 D]] 1 D]]]], GO k,t GO k,t GO k,t (4) where DZt 5 Zt 2 Zt 21 . Table 2 presents the results for our OECD database (top panel) and for Ireland, Korea, Mexico, and Taiwan (bottom panel).27 For most of the OECD database countries and all four of the other countries, growth in VS accounts for more than 30% of export growth. In Canada and the Netherlands, roughly 50% of the growth of exports is accounted for by growth in VS. For Mexico between 1979 and 1994, and for Taiwan between 1961 and 1994, VS growth accounts for more than 50% of export growth. We also calculate the contribution of VS exports to world export / GDP growth for our two world aggregates described above. Table 1 shows that, in both cases, VS growth accounts for more than 30% of export / GDP growth between 1970 and 1990. 4. Nature of vertical specialization Having examined the growth of vertical specialization, in this section we provide decompositions that address two issues. First, how much of the variation in vertical specialization over time and across countries is due to variation in sectoral VS shares or variation in the sector composition of overall exports? Second, what can we say about the geographic orientation of vertical specialization? 26 Strictly speaking, the imported input content of exports, or VS, is equal to the foreign valued-added embodied in exports whenever there are only two stages of production. If there are more than two stages of production, and if a country re-imports some of its own value-added to use as inputs into other goods that are then exported, then foreign value-added is less than the imported input content of exports. 27 In some cases we use GDP instead of gross output, due to lack of data. We draw our Taiwan 1961 VS number from Chenery et al. (1987, p. 217). D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 89 Table 2 Contribution of VS exports to growth in (merchandise) export share of (merchandise) gross output from first to last year of sample a Country First, last year Increase in export share of gross output Contribution of VS share (%) Australia Canada Denmark France Germany Japan Netherlands United Kingdom United States 1968, 1989 1971, 1990 1972, 1990 1972, 1990 1978, 1990 1970, 1990 1972, 1986 1968, 1990 1972, 1990 0.06 0.08 0.17 0.11 0.09 0.03 0.10 0.15 0.07 16.2 50.9 30.8 32.4 22.2 6.1 48.2 31.7 14.1 Ireland Korea Korea Mexico b Mexico b Taiwan Taiwan b 1964, 1990 1963, 1993 1963, 1990 1979, 1997 1979, 1994 1981, 1994 1961, 1994 0.27 0.17 0.16 0.19 0.08 0.09 0.27 33.5 30.7 34.6 40.0 53.4 49.0 51.8 a Source: authors’ calculations based on OECD Input–Output Database; Ireland Central Statistical Office; Bank of Korea; INEGI, Mexico; Quarterly National Income Statistics in Taiwan Area, The Republic of China; Directorate-General of Budget, Accounting and Statistics, Executive Yuan, Taiwan; UNCTAD, Handbook of International Trade and Development Statistics, 1995; Chenery et al. (1987). b GDP used in place of gross output. 4.1. Decomposing variation in vertical specialization over time and across countries Following conventional ‘‘within’’ and ‘‘between’’ accounting, we decompose changes in country-level VS shares into changes in sector VS intensity (the sector VS share of sector exports) and changes in the sector composition of overall exports. Recall that the country VS share is an export-weighted average of the sector VS shares. The decomposition is given by VS k,t D]] 5 Xk,t VS OSD]] *0.5*(v X k,i,t k,i,t i 1 vk,i,t 21 ) k,i,t S VS k,i,t VS k,i,t 21 1 (Dvk,i,t )*0.5* ]] 1 ]]] Xk,i,t Xk,i,t 21 DD , (5) where VS k,t and Xk,t are total VS and exports for country k in period t, VS k,i,t and Xk,i,t are country k, sector i VS and exports in period t, and vk,i,t is country k, D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 90 sector i’s share of total exports in period t. Each sector’s contribution is divided into a contribution due to changes in the sector-level VS share (within), and a contribution due to the changes in the sector-level export share (between). For example, the decomposition tells us the contribution of the auto sector to Canada’s overall VS share growth, as well as a breakout of the auto sector’s contribution into growth in its vertical intensity and growth in its share of total Canadian exports. For expositional convenience, we add up all the sector contributions into a total ‘‘within’’ and a total ‘‘between’’ contribution. These results are given in Table 3. We see that changes in vertical intensity across all sectors account for most of the growth in overall VS share. Changes in the sector composition of overall exports play a small role. With the exception of Japan, increases in vertical intensity within sectors explain 73% or more of the increase in overall vertical specialization. We also group the sector contributions into the broader chemicals, machinery, and other industry categories; we find that, in every country except Australia and Japan, machinery and chemicals accounted for three-quarters or more of the growth in the VS share.28 These sectors contribute primarily through increases in their vertical intensity. We employ similar decompositions to assess whether cross-country variation in VS shares is due to sector vertical intensity or sector export composition. In this Table 3 Sources of growth in VS share of total exports a Increase in VS share of exports first year to last year Australia Canada Denmark France Germany Japan Netherlands United Kingdom United States a 0.022 0.070 0.007 0.060 0.012 20.024 0.032 0.057 0.049 Contribution of: (%) Change in sector VS intensity Change in sector share of overall exports 78.4 73.5 74.1 90.3 90.3 18.2 136.4 110.4 90.3 21.7 26.4 25.8 9.6 9.6 81.9 236.3 210.5 9.7 Source: authors’ calculations based on OECD Input–Output Database. 28 In Japan, the overall VS share declined. Chemicals and machinery VS shares grew, but it was not enough to offset the decline in the ‘‘other’’ VS share. Results for Japan are then consistent with other countries — vertical specialization in chemicals and machinery is growing, and accounting for a major portion of overall VS share growth. Most of the VS share growth in Australia came from mining and quarrying, and non-ferrous metals. D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 91 exercise, we examine differences relative to a ‘‘representative’’ country, constructed by taking a simple average of sector vertical intensity and sector composition over all countries.29 Table 4 presents the results. We again see that differences in vertical intensity account for most of the cross-country variation in VS shares. Differences in sector composition play a fairly minor role. However, unlike in the growth decompositions, we do not find that the chemicals and machinery sectors account for most of the cross-country variation in VS shares. A direct comparison of Canada relative to the U.S. is instructive. In 1990, Canada’s VS share, 0.27, was much higher than that of the U.S., 0.11. Our calculations show that 99% of this difference is due to greater sector vertical intensity in Canada. In particular, greater vertical intensity in the machinery sectors accounts for three-quarters of the overall VS share difference, and greater vertical intensity in the motor vehicles sector alone accounts for half. Finally, we note that it is straightforward to derive from growth accounting decompositions the result that a country’s VS share can grow even if its intermediates share of trade is declining. We illustrate this for the United States; for convenience we examine VS and intermediates trade as a share of imports. Based on our input–output tables, between 1972 and 1990, U.S. intermediates goods imports as a share of total imports fell from 0.51 to 0.37. But VS as a share of imports rose from 0.046 to 0.080. The VS share rose because exports increased in sectors that intensively use imported intermediates and because the sectors with high exports increased their share of total imported intermediates. Performing a Table 4 Sources of cross-country differences in VS share of total exports (final year)a Difference between VS share and VS share of representative country Australia Canada Denmark France Germany Italy Japan Netherlands United Kingdom United States a 20.108 0.051 0.077 0.020 20.024 0.049 20.110 0.150 0.039 20.112 Contribution of: (%) Difference in sector VS intensity Difference in sector share of total exports 52.4 86.1 85.3 70.9 134.2 68.1 92.1 88.3 83.7 103.8 47.5 13.8 14.7 29.1 234.2 31.9 7.9 11.7 16.3 23.8 Source: authors’ calculations based on OECD Input–Output Database. 29 If we had used an export-weighted or GDP-weighted average, as opposed to a simple average, the ‘‘representative’’ country would have essentially been the U.S. and Japan. 92 D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 decomposition similar to the growth accounting decomposition above, we find that growth of the sector export / output shares accounted for 77% of the growth in the VS share, with growth of the imported intermediate shares in the high export sectors accounting for the remaining 23%. The growth and cross-country decompositions deliver two main lessons. Variation in sector VS intensity accounts for almost all of the overall VS share variation over time and across countries. In most of the countries, the chemicals and machinery sectors account for most of the VS share growth over time. 4.2. Vertical specialization: geographic orientation Is vertical specialization, like most trade, primarily North–North, or is it primarily North–South, as in U.S. trade with the Mexican maquiladoras? To calculate the geographic orientation of vertical specialization we combine our OECD database input–output data with bilateral trade data taken from the Statistics Canada World Trade Database.30 We define the North as the OECD and the South as all other countries. Table 5 presents our calculations for each country using the initial year and final year of data. The vertical specialization orientation numbers are quite similar to the overall trade orientation for these countries.31 For both years, the most common geographic pattern of vertical specialization involves North’s imported inputs being transformed into export goods destined for other North countries. South–South vertical links are in general the smallest. Canada’s vertical specialization is almost exclusively North–North, while Japan’s is split evenly among the four origin / destination categories. Comparing the initial year to the final year, we see that the only country to have a substantial decline in its North–North vertical orientation is the U.S., which fell from 49 to 41%. The share of total U.S. vertical specialization consisting of imports originating from the South rose from 26 to 37%. 5. Conclusion Much attention has been focused on the extent of international integration as measured by the growing trade shares of output. In this paper, we identify and document a deeper dimension to international integration involving the inter- 30 Details on this calculation are available from the authors on request. Our calculation requires the assumption that all goods within a sector are homogeneous. 31 A rough comparison can be made by multiplying the northern share of trade in total imports and exports. For the U.S., this figure is 38%, slightly lower than the N–N vertical specialization share. Were there no differences in North versus South sector composition and their use in vertically specialized goods, the geographic orientation of vertical specialization will exactly match the geographic orientation of overall trade. D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 93 Table 5 North–South distribution of VS a Country Partner VS as a % of total VS Origin–destination N–N N–S S–N S–S Initial year Australia Canada Denmark France Germany Italy Japan Netherlands United Kingdom United States 47.4 87.5 64.6 50.5 59.1 50.1 29.2 67.0 48.5 48.8 26.3 6.6 15.3 21.6 18.2 19.0 28.5 14.3 24.5 25.1 17.3 5.3 17.3 20.0 17.2 22.8 20.9 15.6 18.2 17.2 9.0 0.6 2.8 7.9 5.5 8.1 21.3 3.1 8.9 8.8 Final year Australia Canada Denmark France Germany Italy Japan Netherlands United Kingdom United States 43.1 86.0 72.4 62.1 61.4 56.3 29.8 69.5 66.1 40.7 27.5 5.5 12.6 17.3 15.8 16.9 24.9 9.7 17.2 22.6 17.0 7.9 12.9 16.2 17.9 20.5 23.2 18.5 13.3 22.6 12.4 0.6 2.1 4.4 4.9 6.2 22.2 2.3 3.4 14.2 a Source: authors’ calculations. connection of production processes in a vertical trading chain that stretches across many countries. Our primary measure of vertical specialization, VS, is the imported input content of (or foreign value-added embodied in) exports. Our main data source is input–output tables from 10 OECD and four emerging markets countries. Our evidence indicates that, as of 1990, the VS share of merchandise exports for the OECD database countries is 0.20. Smaller countries have VS shares as high as 0.4. For our entire 14-country sample, we calculate that the VS share grew by about 30% between 1970 and 1990, and growth in VS exports accounted for 30% of the growth in the overall export / GDP ratio. Growth in VS intensities within sectors accounts for most of the increase in the national VS shares. The chemicals and machinery sectors, in particular, contribute the most to the aggregate increase. Finally, vertical specialization in the OECD primarily involves other OECD countries — inputs from developed nations are transformed and exported to other developed nations. A notable exception to this D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 94 pattern is the U.S., which has become more oriented toward developing countries at the same time that vertical specialization has grown rapidly. The next step in this research is to identify the forces that explain the growth in vertical specialization and the changing nature of trade. One possibility is that technological shocks have led to the fragmentation of production across different locations. These shocks may include changes in production technique such as an increase in the number of production stages (see Deardorff, 1998; Jones and Kierzkowski, 1997), or shocks that make it easier to separate existing stages of production. For example, the sequential nature of vertical specialization suggests that oversight and coordination of production are important. Improvements in communications technologies (faxes, e-mail, videoconferencing) and sharp declines in the costs of information transmission have made it easier for firms to coordinate and monitor production in diverse locations. These technological developments may also be tied to the increasing importance of foreign direct investment, which has grown faster than international trade in recent decades. Multinationals themselves may play an integral role in the vertical specialization process, as they allow production to fragment across countries while it remains internal to the firm. A second possibility is that reductions in the cost of moving goods have driven the increase in vertical specialization. Tariffs and transportation costs penalize a good produced sequentially in multiple countries each time the good-in-process is shipped to another country. Hence, reductions in these barriers yield a multiplied reduction in the cost of producing a vertically specialized good. This intuition is clearly related, but not identical, to the effective rate of protection concept.32 For concreteness, consider the following extreme example. A vertically specialized good is produced (under perfect competition) in N sequential stages with each stage produced in a different country. The first stage is produced with value-added, but each succeeding stage has infinitesimally small value-added. In this case, the cost of the final good will be P 5 (1 1 t )N P1 , where P1 is the price of the first stage. A one percentage point reduction in tariffs leads to an N percentage point decline in the price of the vertically-specialized final good. Exports of vertically specialized goods thus increase relative to exports of goods which cross only one border.33 This effect can be seen in the anecdotal and case study evidence linking the formation of regional trading blocs to increased vertical specialization. For example, Mexico’s maquiladoras resulted from changes in trade laws specifically designed to enhance vertical specialization. Imported inputs from the U.S. are not subject to Mexican tariffs as long as they are used in the maquiladoras, and only 32 The modern, theoretically consistent and general approach to effective rate of protection owes much to Jim Anderson. See, for example, Anderson (1998). 33 In a previous version of this paper, we developed an extension of the Dornbusch et al. (1977) Ricardian model that generates the two effects discussed above. D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 95 the Mexican value-added is subject to U.S. tariffs when maquiladora goods are exported back to the U.S.34 Investigating the relative importance of these two hypotheses, technological shocks and trade barrier reduction, in driving the increase in vertical specialization is an important task for future research. Acknowledgements We thank the editor and two anonymous referees for very helpful suggestions. This paper is a revision and extension of ‘‘The Growth of World Trade’’ by Ishii and Yi (1997). Since then, the authors have benefited from comments by Jim Anderson, Marianne Baxter, Don Davis, Paul Evans, John Fernald, Raquel Fernandez, Caroline Freund, Jess Gaspar, Jim Harrigan, Jane Ihrig, Boyan Jovanovic, Chinhui Juhn, Bob King, Narayana Kocherlakota, Ayhan Kose, Michael Kouparitsas, Robert E. Lipsey, Dana Rapoport, and Lucinda Vargas as well as seminar participants at Brandeis University, New York University, University of Kentucky, Ohio State University, University of Virginia, Duke University, University of North Carolina, Federal Reserve Bank of Philadelphia, University of Toronto, Queen’s University, University of Chicago GSB Brown Bag, 1997 NBER Summer Institute, the 1998 Winter Econometric Society Meetings, and the 1999 SED meetings. The authors thank Josh Greenfield, Rema Hanna, Stefan Papaioannou, and Dana Rapoport for outstanding research assistance. The views expressed in this paper are those of the authors and are not necessarily reflective of views at the Federal Reserve Bank of New York or the Federal Reserve System. References Anderson, J.E., 1998. Effective protection redux. Journal of International Economics 44, 21–44. Arndt, S.W., 1997. Globalization and the open economy. North American Journal of Economics and Finance 8, 71–79. Balassa, B., 1967. Trade Liberalization Among Industrial Countries. McGraw-Hill, New York. Barry, F., Bradley, J., 1997. FDI and trade: the Irish host-country experience. The Economic Journal 107, 1798–1811. Berman, E., Bound, J., Griliches, Z., 1994. Changes in the demand for skilled labor within U.S. manufacturing: evidence from the Annual Survey of Manufactures. Quarterly Journal of Economics 109, 367–398. 34 Two of the case studies in Ishii and Yi (1997) and Hummels et al. (1998), U.S.–Mexico maquiladora trade and the 1965 U.S.–Canada Auto Pact, are results of preferential tariff reductions. As noted earlier, these reductions led to rapid growth in vertical specialization. Recently, there has been much research on regional trade agreements versus multilateral trade agreements. See, for example, Krugman (1992), Grossman and Helpman (1995), and Freund (1998). 96 D. Hummels et al. / Journal of International Economics 54 (2001) 75 – 96 Campa, J., Goldberg, L.S., 1997. The evolving external orientation of manufacturing: a profile of four countries. Federal Reserve Bank of New York Economic Policy Review July, 53–81. Chenery, H., Robinson, S., Syrquin, M., 1987. Industrialization and Growth: A Comparative Study. Oxford University Press, New York. Deardorff, A., 1998. Fragmentation in simple trade models. Manuscript, University of Michigan. Dixit, A.K., Grossman, G.M., 1982. Trade and protection with multistage production. Review of Economic Studies 49, 583–594. Dornbusch, R., Fischer, S., Samuelson, P., 1977. Comparative advantage, trade and payments in a Ricardian model with a continuum of goods. American Economic Review 67, 823–839. Feenstra, R.C., 1998. Integration of trade and disintegration of production in the global economy. The Journal of Economic Perspectives. 12, 31–50. Feenstra, R.C., Hanson, G.H., 1997. Foreign direct investment and relative wages: evidence from Mexico’s maquiladoras. Journal of International Economics 42, 371–394. Feenstra, R.C., Hanson, G.H., 1996. Foreign investment, outsourcing and relative wages. In: Feenstra, R.C., Grossman, G.M., Irwin, D.A. (Eds.), The Political Economy of Trade Policy. Papers in Honor of Jagdish Bhagwati. MIT Press, Cambridge, MA, pp. 89–127. Findlay, R., 1978. An Austrian model of international trade and interest rate equalization. Journal of Political Economy 86, 989–1008. Freund, C., 1998. Regionalism and permanent diversion. International Finance Discussion Paper [602, Board of Governors of the Federal Reserve System. Grossman, G., Helpman, E., 1995. The politics of free trade agreements. American Economic Review 85, 667–690. Grunwald, J., Flamm, K., 1985. The Global Factory: Foreign Assembly in International Trade. Brookings, Washington, DC. Hummels, D., Rapoport, D., Yi, K.-M., 1998. Vertical specialization and the changing nature of world trade. Federal Reserve Bank of New York Economic Policy Review June, 79–99. Ishii, J., Yi, K.-M., 1997. The growth of world trade. Federal Reserve Bank of New York Research Paper No. 9718, May. Jones, R.W., Kierzkowski, H., 1997. Globalization and the consequences of international fragmentation. Manuscript, University of Rochester and Graduate Institute of International Studies, Geneva. Krugman, P.R., 1992. Is bilateralism bad? In: Helpman, E., Razin, A. (Eds.), International Trade and Trade Policy. MIT Press, Cambridge, MA, pp. 9–24. Krugman, P.R., 1995. Growing world trade: causes and consequences. Brookings Papers on Economic Activity 1, 327–377. Lawrence, R., 1994. Trade, multinationals, and labor. NBER Working Paper No. 4836, August. Naughton, B., 1996. China’s Emergence and Prospects as a Trading Nation. Brookings Papers on Economic Activity, Vol. 2. Brookings, Washington, DC. Organization for Economic Cooperation and Development, 1995. The OECD Input–Output Database. Organization for Economic Cooperation and Development, Paris. Sanyal, K.K., 1983. Vertical specialization in a Ricardian model with a continuum of stages of production. Economica 50, 71–78. Sanyal, K.K., Jones, R.W., 1982. The theory of trade in middle products. American Economic Review 72 (1), 16–31. Slaughter, M.J., 2000. Production transfer within multinational enterprises and American wages. Journal of International Economics 50, 449–472. United Nations Conference on Trade and Development, 1997. Handbook of International Trade and Development Statistics 1995. United Nations, New York. Yeats, A.J., 1998. Just how big is global production sharing? Manuscript, World Bank.