Journal of International Economics 48 (1999) 71–100 Business groups and product variety in trade: evidence from South Korea, Taiwan and Japan a, b c * Robert C. Feenstra , Tzu-Han Yang , Gary G. Hamilton a Department of Economics, University of California, Davis, CA 95616, USA b Council for Economic Planning and Development, Executive Yuan, Taiwan, R.O.C. c Department of Sociology, University of Washington, Seattle, WA 98195, USA Received 7 October 1997; accepted 29 October 1997 Abstract We analyze the impact of market structure on the trade performance of South Korea, Taiwan and Japan. South Korea has many large, vertically-integrated business groups known as ‘chaebol’, whereas business groups in Taiwan are smaller and more specialized in the production of intermediate inputs. We test the hypotheses that a greater presence of business groups leads to less product variety but higher product quality, using data on exports from these countries to the United States. We find that Taiwan tends to export a greater variety of products to the US than Korea, and this holds particularly for final goods. In addition, Taiwan exports higher-quality intermediate inputs, whereas Korea exports higher-quality final goods. A comparison with Japan is also presented, which has greater product variety in its sales to the US than either Taiwan or Korea, due to its larger size. 1999 Elsevier Science B.V. All rights reserved. Keywords: Product variety; Business group; South Korea; Taiwan; Japan JEL classification: F14; O53 1. Introduction Since the early 1980s, it has been argued that market structure matters for trade patterns (Helpman and Krugman, 1985). Models developed to capture this link * Corresponding author. Tel.: 11-530-7527022; fax: 11-530-7529382. E-mail address: [email protected] (R.C. Feenstra) 0022-1996 / 99 / $ – see front matter 1999 Elsevier Science B.V. All rights reserved. PII: S0022-1996(98)00024-5
Transcript
Journal of International Economics 48 (1999) 71–100
Business groups and product variety in trade: evidencefrom South Korea, Taiwan and Japan
a , b c*Robert C. Feenstra , Tzu-Han Yang , Gary G. HamiltonaDepartment of Economics, University of California, Davis, CA 95616, USA
bCouncil for Economic Planning and Development, Executive Yuan, Taiwan, R.O.C.cDepartment of Sociology, University of Washington, Seattle, WA 98195, USA
Received 7 October 1997; accepted 29 October 1997
Abstract
We analyze the impact of market structure on the trade performance of South Korea,Taiwan and Japan. South Korea has many large, vertically-integrated business groupsknown as ‘chaebol’, whereas business groups in Taiwan are smaller and more specialized inthe production of intermediate inputs. We test the hypotheses that a greater presence ofbusiness groups leads to less product variety but higher product quality, using data onexports from these countries to the United States. We find that Taiwan tends to export agreater variety of products to the US than Korea, and this holds particularly for final goods.In addition, Taiwan exports higher-quality intermediate inputs, whereas Korea exportshigher-quality final goods. A comparison with Japan is also presented, which has greaterproduct variety in its sales to the US than either Taiwan or Korea, due to its larger size. 1999 Elsevier Science B.V. All rights reserved.
Keywords: Product variety; Business group; South Korea; Taiwan; Japan
JEL classification: F14; O53
1. Introduction
Since the early 1980s, it has been argued that market structure matters for tradepatterns (Helpman and Krugman, 1985). Models developed to capture this link
72 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
were motivated by empirical observations such as the large amount of ‘intra-industry’ or two-way trade in similar products between countries. It is nowcommon to incorporate intra-industry trade, due to monopolistic competition, intothe specification of equations explaining trade flows. It can be questioned,however, whether these estimating equations really depend on the marketstructure. Helpman (1987) notes that his specification of trade volume equationsconsistent with monopolistic competition also applies when countries are special-
1ized in different products for any other reason. Hummels and Levinsohn (1993),(1995) have shown that these equations fit well not only for industrial countries,but also for trade from developing countries, where we do not expect that
2monopolistic competition is the prevailing market structure.Despite the well-developed theoretical literature on monopolistic competition
and trade, we are still lacking actual measures of product variety and marketstructure that can be applied to the theory. In this paper, we apply the productvariety measure developed in Feenstra (1994), which is exact for a constantelasticity of substitution (CES) aggregator function. As a measure of marketstructure, we focus on the ‘business groups’ found in many Asian countries, but
3also in Europe, Latin America, and elsewhere. Known as ‘keiretsu’ in Japan and‘chaebol’ in Korea, these groups involve some degree of common-ownershipacross firms. The keiretsu in Japan have attracted a considerable amount ofacademic attention, including some analysis of their impact on Japanese imports
4(Fung, 1991; Lawrence, 1991). Less attention has been paid to the groups inSouth Korea and Taiwan, and their impact on trade patterns (exceptions areMartins, 1992; Rodrik, 1993). The chaebol in Korea share certain hierarchicalfeatures with the keiretsu, and are even more strongly vertically-integrated fromthe production of raw materials to the export of finished goods. In contrast, thebusiness groups found in Taiwan are smaller and less vertically-integrated thanthose in Korea or Japan, and more focused in the upstream sector. Theintermediate inputs produced by the groups are sold to small and medium-sizedfirms downstream, for further processing and export.
1For example, complete specialization may occur due to technological differences across countries,as modeled by Davis (1995).
2Hummels and Levinsohn (1993), (1995) test a key hypothesis from Helpman’s model over a set ofOECD countries, and over a set of non-OECD countries for which intra-industry trade should not beimportant. They find substantial support for the hypothesis in both sets of countries, suggesting thatsomething other than monopolistic competition explains the results. Jensen (1995) has re-examined theresults of Hummels and Levinsohn, and argues that there are greater differences between the OECDand non-OECD countries than first appears.
3See Caves (1989), Granovetter (1994), (1995) for a general review of the literature on businessgroups. For research on business networks in Asia, see Gerlach (1992), Futatsugi (1986), Hamilton andBiggart (1988), Orru et al. (1990), (1991) and the papers in Hamilton (1991).
4Fung (1991) and Lawrence (1991) have examined the effects of the keiretsu on Japanese trade, andargued that they constitute a (limited) barrier to entry. This conclusion was questioned by Bhagwati(1992) and Saxonhouse (1993), however, who argue that the changing membership of firms belongingto each keiretsu invalidates any conclusions about the groups as a whole.
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 73
In Section 2, the business groups in South Korea, Taiwan and Japan aredescribed in greater detail, using available data from published sources. Then inSection 3, several hypotheses about how the groups will affect trade patterns aredeveloped. We work from the assumption that the groups maximize the jointprofits of the firms involved, and should therefore be analyzed like multi-product
5companies. We hypothesize that a multi-product group will have less incentive todevelop new product varieties as compared to single-product firms, since any newvariety will reduce existing sales. On the other hand, a multi-product group willhave greater incentive to develop a reputation in high-quality, that would lead toan increase in demand for all its products. We therefore expect that a greaterpresence of business groups leads to less product variety but higher productquality. The reasoning behind these hypotheses is reinforced when we alsoconsider vertical-integration within the groups.
These hypotheses are tested using data on US imports of highly disaggregateproducts from South Korea, Taiwan and Japan. Indexes of product variety areconstructed at the 5-digit industry level to reflect the range of products sold fromeach country. In addition, a ‘product mix’ index is constructed to reflect whethereach country tends to export high-priced or low-priced products within each5-digit category. This index can be interpreted as a measure of product quality, andwe show that both indexes are theoretically related to the consumption servicesprovided per unit of imports.
Our results are presented in Section 4. We find that Taiwan tends to export agreater variety of products to the US than Korea, and this holds particularly forfinal goods. In addition, Taiwan exports relatively more high-priced intermediateinputs, whereas Korea exports relatively more high-priced final goods. Theseresults are discussed in relation to the business group structure of the twoeconomies. A comparison with Japan is also presented, and we find that Japan hasgreater product variety in its sales to the US than either Taiwan or Korea, whichcan be explained by the very large size of that economy. The striking differencebetween the product variety and mix of Taiwan and Korea, despite the similarityof factor endowments across these economies, appears to confirm the importanceof market structure as a determinant of trade patterns. The significance of theseresults and directions for further research are discussed in Section 5.
2. Descriptions of the groups
A central feature of Korea’s industrial organization is the business groups, or6chaebol. These groups, consisting of legally-independent firms, are affiliated
5Note that even with maximization of joint profits, the equilibrium profits of the groups need not beabnormally high if there is free entry. In fact, the evidence suggests that profits of the keiretsu are nohigher than average (Weinstein and Yafeh, 1995).
6The chaebol are described in Amsden (1989), Biggart (1990), Hamilton and Biggart (1988),Hamilton et al. (1990), Kim (1991), (1993), (1997), Orru et al. (1991), Steers et al. (1989) and Zeile(1991).
74 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
under a common group name and are centrally controlled through direct familyownership and mutual shareholding among member firms. As shown in Table 1,the 50 largest business groups accounted for 45% of total sales in the manufactur-ing sector in 1983, and even more in other sectors. These sales figures give aninflated estimate of the importance of the chaebol, however, because transactionsof semi-finished goods between firms within a group are included. In Table 1, thefigures in parentheses give the value-added shares accounted for by the businessgroups within each sector, and these figures are not affected by the frequency of
7intra-group transactions. Overall, the value-added of the top 50 business groupsaccounted for one-fifth of GDP in 1983 (Zeile, 1991).
In the manufacturing sector, the chaebol spread across many industries. Asshown in Table 2, there were five manufacturing industries in which the chaebolaccounted for more than 50% of the total sales, and eight others in which theyaccounted for between 25 and 50%. Many of these are chemical or heavyindustries. This pattern of concentration is a direct result of the government’scredit policy. During the 1970s, the Korean government applied discriminatoryinterest rates and controlled both domestic and foreign loans in order to influenceindustrial development. It supplied a large amount of credit with low interest ratesto business groups for investment in the heavy and chemical industries, with theresult that these industries and groups grew rapidly.
To measure the degree of vertical integration for the Korean business groups,we have used self-reported accounting data that were compiled for the KoreanInvestors Service (Feenstra, 1997). These data contain, among other things, theinter-firm transactions for all the firms within each chaebol in 1989. Using this, wecan measure the total sales between firms within the same group, and express thisas a ratio of purchases of all sales by the firms in that group. This internalization
Table 1Business group shares by major sector, 1983 (percent)
Sector Korea Taiwan JapanSales share of 50 largest Sales share of 96 Sales share of 16
a bchaebol (value-added) largest business groups largest keiretsu
Source: Hamilton (1988), Table 3; Hamilton et al. (1990), Table 4.aFigures in parentheses give value-added of all firms in business groups selling in that sector, relative tototal value-added of the sector, for the year 1986.bFigures for Japan are for fiscal year 1982.
7Value-added figures were not available for the other countries in Table 1.
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 75
Table 2Business group shares by industry, 1983 (percent)
Industry (SIC code) Korea Taiwan Japan50 largest 96 largest groups 16 largest
Source: Hamilton et al. (1990), Table 5, p. 116.aThe figures in parentheses include the sales of state-owned businesses, computed from the Yearbook ofFinancial Statistics of the Republic of China, and the Report on 1983 Industrial and CommercialCensus of the Republic of China.bThe Taiwanese business group share is principally in pulp and paper.cIncludes chemical products.dIncludes beverage and tobacco.eThe Korean chaebol share is principally in wood products and furniture.
ratio is a measure of the degree of vertical integration for the group, but can bebiased upward by the presence of trading companies that simply transfer productswithin a group. We therefore calculated this measure both with and without the
8purchases of trading companies. For the largest 44 chaebol, the weighted averageinternalization ratio is 22.1% when trading companies are included, and 13.8%
8In the data on intra-group transactions, the chaebol trading companies are usually among the mainbuyers of goods sold by other firms in the group, and they are main seller of products in a final form. Inorder to not double count a purchase and a sale of the same goods within the chaebol network, weignored the internal purchases but not the sales of chaebol trading companies.
76 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
9when trading companies are excluded (Feenstra, 1997). There is a considerablerange of internalization across different groups. The largest five groups —Samsung, Hyundai, Lucky-Goldstar, Daewoo and Sunkyong — have 1989 salesbetween $8.9 and $26 billion, and together account for 60% of the total sales ofthe 44 groups. These five groups have a weighted average internalization ratio of28.7% including their trading companies, and 17.1% without them. In comparison,the remaining 39 have a weighted average internalization ratio of 11.5% whentheir trading companies are included, and 8.6% otherwise.
10In Taiwan, business groups are much less dominant. The total sales of the 96largest Taiwanese business groups accounts for only 19% of sales in themanufacturing sector (see Table 1). As shown in Table 2, they are influential in asmaller number of manufacturing industries, principally those producing inter-
11mediate goods. There is only one industry — textiles — in which the group shareof total sales is over 50%, and another three industries — chemical materials,non-metallic mineral products, and food products — in which they have a sharebetween 25 and 50%.
The shares shown in parentheses in Table 2 include the sales of both businessgroups and enterprises owned by the state government. The state-owned en-terprises are also concentrated in intermediate goods. Two out of the five industrieswith significant government shares — basic metal and petroleum — are obviousupstream industries. In the food industry, state-controlled enterprises mainlyproduce sugar, salt, and animal feeds — all raw materials. In transportationindustries, the state is involved in shipbuilding and highway construction. Addingup the shares of Taiwanese business groups and state-owned enterprises in theoverall economy, their dominance in intermediate goods industries is quiteapparent, with the exception of fabricated metal, a category that includes bothintermediate and final goods.
The smaller size of Taiwanese business groups and their focus on intermediateinputs are two of the major distinctions between the Taiwanese and Koreanbusiness groups. Although both economies are heavily export oriented, the largestbusiness groups in each economy occupy very different structural locations: theKorean chaebol dominate in the export sector, and the biggest business groups inTaiwan produce intermediate goods that are sold domestically. These domesticsales are primarily to small and medium size firms that have only an arm’s lengthrelationship with the producers of intermediate inputs. While the Korean govern-
9These means are weighted by the sales of each business group.10The literature on business groups in Taiwan is relatively small when compared with the literature
on the Korean chaebol. However, see Chou (1985), Greenhalgh (1988), Hamilton and Biggart (1988),Hamilton and Kao (1990), and Numazaki (1986), (1991).
11Some of the industries in Table 2 include both intermediate and final goods, such as pulp andpaper, printing and publishing (SIC 26127) and lumber and wood products (SIC 24125). In thesecases, we have classified the industry according to the principal output in the country with the largestbusiness group share.
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 77
ment has provided extensive subsidies to the chaebol for exporting, there is much12less direct support for exporting in Taiwan.
Information on the internal transactions of the business groups in Taiwan iscollected and published (in Chinese) in annual volumes of the China CreditInformation Service (CCIS). In 1983, for example, over 40% of Taiwan’s businessgroups reported that none of their member firms were linked by ongoing businesstransactions. An additional 33% reported five or fewer routine transaction linkages
13among member firms.A limitation of this dataset is that it does not provide the magnitude of
intra-group purchases and sales, but just whether or not any transaction occurredbetween the firms in each group. However, for the publicly-traded corporations,the magnitude of their intra-group transactions is contained in annual reports madeto the securities exchange commission.
In ongoing research, we have contracted with the CCIS to collect thisinformation for all publicly-traded firms in the largest 80 business groups, andobtain the same information for private firms in those groups by using a survey. Asfor Korea, we measure internalization as the ratio of all sales between firms in thesame business group to the total sales by that group. This dataset shows that theweighted average internalization ratio for Taiwan is 9.5% when trading companiesare included, and 8.5% when they are excluded (Feenstra, 1997). This compareswith the 17.1–28.7% range obtained for the largest five business groups in Korea,and the 8.6–11.5% range obtained for the remaining 39 groups. Thus, it is evidentthat the top five groups in Korea are the outliers in this comparison, being bothextremely large in terms of sales and much more highly integrated than othergroups in that country or in Taiwan.
In Japan, the independent firms in intermarket business groups mutually owneach other’s shares. For any one firm, the controlling interest is held only by thegroup as a whole (Orru et al., 1990; Gerlach, 1992). Typically, individualownership, whether through stock or through private holdings, accounts for verylittle of the total ownership of Japanese business groups. Most firms are publiclylisted on one of Japan’s large stock exchanges, but only a small percentage of thetotal shares are actually available for purchase. Most equity in business groupfirms is held by other firms in the same business group.
Structurally speaking, there are two types of business groups in Japan: one ahorizontally and the second a vertically arranged network among firms (Orru et al.,
12Wade (1990 p. 110) notes that small and medium sized firms accounted for 65% of exports fromTaiwan in 1985, but argues that government support for upstream firms such as chemicals, metals andelectronics indirectly supports the downstream firms. Also, firms exporting from Taiwan do obtainsome assistance from the sogo shosha, the Japanese trading companies, as well as a state exportpromotion agency.
13Hamilton et al. (1990 p. 122). This data is based upon the information found in China CreditInformation Service (1983).
78 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
1990; Gerlach, 1992). The first type, known as ‘intermarket groups’, or ‘inter-market keiretsu’ have ownership and loan relationships that extend acrossunrelated industries. There are six major intermarket groups, each of which has abank and a trading company (or sogo shosha) at its center, and its member firmsbelong to a ‘presidents club’. Because a bank is at the center of the group, thesegroups are also called ‘main bank groups’ (Aoki and Patrick, 1994). The majorfirms in these groups, along with a set of relatively autonomous, very large firms(e.g. Toyota), organize a second type of ownership network, called ‘keiretsu’ or‘vertical keiretsu’. They consist of interlinkages among many small, medium, andlarge independent firms so that inter-firm networks overlap directly with pro-duction sequences. The activities of these production networks are coordinated bythe large firms in the network. The data in Table 1 and Table 2 includes the sales
14of the six major ‘intermarket’ groups along with 10 other vertical keiretsu. Incomparison with Korea, the Japanese business groups are more specialized inparticular industries, and account for substantial shares of sales in intermediateindustries, chemicals, machinery, electronics and transportation equipment.
For the intermarket groups in Japan, Gerlach (1992 pp. 143–149) reports thatthe rate of internal transactions has been variously calculated to be around 10%.Numbers of roughly this magnitude were also calculated by the Japanese FairTrade Commission in a widely-publicized study, finding that intermarket groups
15rely on other group members for 13% of purchases. These groups are thereforeless integrated than the largest five chaebol in Korea, with the internalization ratioreported above 17.1–28.7%, depending on whether the trading companies areincluded or not. The situation is not so clear, however, when the vertical keiretsuin Japan are compared with the Korea chaebol. The same study by the JapaneseFair Trade Commission asks groups what they buy from companies in which theyhave more than 10% equity, even when those companies do not attend the‘presidential council’ (i.e. are not part of the same intermarket group). Theseresponses add another 25% to the 13% rate of internal purchase, for a total
16internalization ratio of 38%. Estimates this high also occur for Japaneseautomobile manufacturing groups. Dyer (1998) cites MITI statistics showing that
14The six ‘intermarket’ groups are Mitsubishi, Mitsui, Sumitomo, Fuyo, Ikkan, and Sanwa, while the10 other keiretsu are Tokai Bank, IBJ, Nippon Steel, Hitachi, Nissan, Toyota, Matsushita, Toshiba-IHI,Yokyu and Seiba.
15Cited in ‘Japan’s Industrial Structure: Inside the Charmed Circle’, The Economist, January 5, 1991,p. 54.
16When the groups are also asked what they buy from overseas related companies in which they havemore than 10% equity, that adds a further 28% to the intragroup purchases. By this broadest measure,then, the total rate of internalization is as high as 66%. These numbers are cited (with accompanyingtext of the Japanese Fair Trade Commission report) in an internal memorandum of the Department ofTreasury, Embassy of the United States of America, prepared by Timothy Geithner, February 26, 1992,and also by Richard K. Nanto of the Congressional Research Service at a meeting of the Center forAsian Pacific Affairs, The Asia Foundation, on April 15, 1994.
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 79
31% of the costs of products sold are manufactured internally by firms within thegroup. In comparison, the Hyundai group in Korea had an internalization ratio of19.9–33.0% in 1989, depending on whether its trading companies are included or
17not.We can also compare the business groups to American corporations that have
accounting procedures based on multiple internal profit centers. In a survey of 237US corporations engaged in pricing internal transfers,Vancil (1978, p. 176; cited inEccles, 1985 pp. 106–113), found that over 77% of these corporations had 15% orless of their total purchase of intermediate inputs satisfied by divisions within theirown corporations. This degree of internalization appears to be roughly comparableto the intermarket groups in Japan, but less than the Korean chaebol. We do notview the large American conglomerate as fundamentally different from a Japanesekeiretsu, in that both are involved with the exchange of goods and informationbetween multiple divisions /firms. The particular structure taken by the keiretsuversus the American conglomerate reflects in part the legal provisions of eachcountry. For example, the Glass-Steagal Act in the United States separates banksand securities companies, so that a bank cannot act as both a lender and ashareholder, as occurs in Japan. Furthermore, the US Investment Company Act of1940 prevents one company from taking a managerial role in another, unless itactually owns it (Gilson and Kraakman, 1993). It is precisely this managerial orfinancial control that some feel is played by the banks at the center of groups inJapan, despite that fact that they are limited in their holdings of individual
18companies.Based on these comparisons, it would appear that the Korean chaebol would
rank among the world’s most vertically-integrated networks. Their rate ofinternalization exceeds the average for the intermarket groups in Japan, though notnecessarily for the vertical keiretsu. The top five Korean chaebol, in particular, areextremely large and much more integrated than their Taiwanese counterparts, andthan other groups in Korea. Some of these groups have become household namesin the United States, like Samsung and Hyundai, and also have a dominantpresence in other international markets. The aggressive exporting of the Korean
17The internalization ratios for 44 groups in Korea are reported in Feenstra (1997). The internaliza-tion figure for automobile manufacturing groups in Japan includes the value of products manufacturedinternally within each firm, in addition to those traded between group firms. Thus, it is biased upwardsas compared to the value of intra-group transactions for Korea.
18In Japan, the Anti-Monopoly Law Reform of 1977 specifies that banks cannot own more than 5%of the shares of any company. This does not necessarily prevent them from exercising some financialcontrol, particularly in bad states. Aoki (1990) and Lincoln et al. (1994) argue that the banks providean insurance role for groups firms that are in difficulty, and that these firms recover faster due to theirgroup affiliation. Fung (1997) theoretically examines the impacts of bank control on welfare of thegroup, and on global welfare. Note that in South Korea, commercial banks are not allowed to be part ofthe chaebol, and these groups are closer to the vertical keiretsu in structure than to the intermarketkeiretsu or main bank groups.
80 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
groups contrast with those in Taiwan, which seldom market under their ownbrandnames (with the exception of Acer computers), and prefer to act as ‘originalequipment manufacturers’ for retail companies in the United States and elsewhere.The trading companies found in many Korean groups are actively involved inoverseas markets, while this is less true for the Taiwanese groups. Theseobservations suggest that the group structure may affect their trade patterns, and inthe next section we develop some hypotheses along these lines.
3. Product variety in trade
A business group is one example of a ‘network’ structure between firms, and ingeneral, networks can affect production and trade in a number of ways. Theadvantages to being within a network potentially include: information flowsbetween firms and customers (Egan and Mody, 1992; Rauch, 1997); informationflows on production techniques between firms and suppliers (Aoki, 1990);financial insurance provided by a bank within a group (Aoki and Patrick, 1994;Hoshi et al., 1990, 1991; Lincoln et al., 1994); improved access to intermediateinputs sold by member firms (Feenstra et al., 1997a), etc. A two-country modelwith an endogenous number of business groups is developed by Fung andFriedman (1996), and that analysis suggests that there are multiple equilibria, sothat the group structure is not uniquely determined by economic considerations.
We shall focus on the interactions of the group’s firms in selling multipleproducts, and in purchasing intermediate inputs from within the group. The type ofgroup we have in mind has its sales concentrated within a small set of industries,and may also be vertically-integrated. This description applies to the chaebol inKorea, to the groups found in upstream sectors in Taiwan, and to the verticalkeiretsu in Japan, but less so to the intermarket (or main bank) groups in Japan andother groups in Taiwan that are highly diversified across products. We shallconceptualize a business group as a set of firms that jointly maximize profits, andcompare the equilibrium achieved in a market dominated by business groups withthat achieved in a market of the same size, but with unaffiliated firms.
Our first hypothesis is that a market dominated by business groups will have alower level of product variety than the same market dominated by single-productfirms. The reason is that a new product will partially take away demand fromexisting products sold by the group, so the marginal revenue is lower than it wouldbe for a single-product firm. If the fixed costs associated with product developmentare the same, then we expect that a multi-product group will have less incentive todevelop new product varieties. This hypothesis in confirmed theoretically by Pakesand McGuire (1994), for example, who compare the Nash equilibria obtained withsingle-product and multi-product firm with product differentiation.
On the other hand, a group will have greater incentive to develop a reputation inhigh-quality, that would lead to an increase in demand for all its products in that
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 81
industry. That is, any action that shifts out the demand curves for all its products,such as building reputation, will be more valuable to a multi-product group than toa single-product firm. Our second hypothesis is then that a market dominated bybusiness groups will have a higher level of product quality than the same marketdominated by single-product firms. This result is obtained theoretically by Rodrik(1993), who supposes that the level of product quality perceived by buyers equalsthe average product quality within an industry. In that case, groups that have ahigh share of sales within the industry have a greater incentive to improve productquality. Rodrik measures quality by comparing the unit-values of productsexported from South Korea with those from Taiwan. The measure of productquality we shall use generalizes the unit-value comparison made by Rodrik, andwill indicate whether a country primarily sells high-priced or low-priced productswithin each industry.
We have derived these two hypotheses using the horizontal concentration of agroup within an industry, but they are reinforced when we also consider vertical-integration. If the firms in a group are jointly profit-maximizing, then intermediateinputs will be sold internally at their marginal cost, even if this does not cover the
19fixed costs of product development. This means that the marginal cost ofproducing final goods will be lower for firms in a group than for unaffiliated firms,who purchase inputs at a price including a mark-up. As a group sells more, it canalso develop more intermediate inputs, thereby further lowering costs: the vertical-integration leads to economies of scale within the group. The profit-maximizingstrategy we would expect in this situation is for the group to produce longerproduction runs than unaffiliated firms, but focus on a narrower range of productvariety. This is confirmed theoretically by Feenstra et al. (1993), who find a higherlevel of product variety with business groups than with unaffiliated firms,consistent with our first hypothesis. The implications of vertical-integration for thesecond hypothesis are less clear, but it is possible that the longer production runsof a business group would make it more willing to invest in product quality, if thisincreases demand by a proportionate amount.
These two hypotheses apply to a comparison of like-sized economies, such asSouth Korea and Taiwan. We anticipate that the dominance of the chaebol inKorea, especially in the downstream markets, will lead to less product variety buthigher product quality as compared to Taiwan. In a larger economy, such as Japan,we expect there to be greater product variety due to its size, and do not make anyspecific prediction concerning product quality. While we are primarily interested inthe contrast between Korea and Taiwan, we include Japan as a comparison case tocheck that our product variety measure indeed gives higher values there. We shalltest our hypotheses using disaggregate data on US imports from these countries.
19Fixed costs of developing intermediate inputs would have to be covered by internal transfers withinthe group, or by sales external to the group. The internal transfers suggests a reason for financialinstitutions to be a part of groups.
82 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
While it would be preferable to have data on their domestic production, orworldwide sales, the advantage of using the US import data is that it is much moredisaggregate than any other data source: over 10 000 commodities are dis-tinguished annually. We use this data to construct an index of product variety, anda ‘product mix’ index, which is interpreted as a measure of product quality. Theseare derived in the next section.
3.1. Product variety and mix indexes
For each industry, treat the US imports from each country j 5 1, . . . , J asdifferentiated across i 5 1, . . . , N varieties, where each country may supply only asubset I # h1, . . . , Nj of these varieties. Let x 5 (x , x , . . . ,x ) denote thej j 1 j 2 j Nj
vector of import quantities from country j, and suppose that the total servicesobtained from imports of country j for the industry in question are given by theCES function f(x , I ):j j
s / (s 21)(s 21) /sf(x , I ) 5 Oa x , a . 0, (1)j j i ij iS D
i´Ij
where the elasticity of substitution is s.1. If the product in question is aconsumer good, then f(?) represents the utility function for the varieties fromcountry j, and otherwise it is a production function for importing firms. We assumethat total utility or output obtained from imports from all source countries is givenby the function:
U 5 G[ f(x , I ), . . . , f(x , I )] (2)1 1 J J
which aggregates the services obtained from all countries. Eq. (2) assumes that theimport varieties from each country are weakly separable from each other withinthe function G(?), which is convenient in developing our indexes.
Let X 5 o x denote the total quantity of country j’s imports, measured inj i´I ijj
physical units. Then the services obtained per unit of import is obtained bydividing total services by the physical quantity X :j
A ; f(x , I ) /X . (3)j j j j
Then (2) can be rewritten as:
U 5 G(A X , . . . , A X ). (29)1 1 J J
The services per unit of import A cannot be measured directly, since it dependsj
on the unknown level of service f(x , I ), but an empirical measure can be obtainedj j
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 83
by considering the ratio of relative services A /A . Letting p .0 denote the pricej k j20vector from country j, this ratio can be measured by:
A E /X c( p , I )j j j j j] ]] ]]]5 Y , (4)S D S DA E /X c( p , I )k k k k k
where E denotes total expenditure on imports from country j, and c( p , I ) is thej j j
unit-cost function dual to (1):
1 / (12s )12s sc( p , I ) 5 Ob p , b 5 a . (5)j j i ij i iS D
i´Ij
Expression (4) is the ratio of unit-values of imports from country j and k,divided by the ratio of unit-costs from the two countries. While the unit-values aredirectly obtained from import data, the unit-costs are not observed. However, theirratio can be measured by an exact price index. In particular, suppose that x and xj k
are the cost-minimizing quantities with prices p and p , respectively, and that thej k
set of common goods I;(I >I ) imported from both countries is not empty.t t 21
Then from Feenstra (1994), the ratio of unit-costs can be measured as:
1 / (s 21)c( p ,I ) /c( p ,I ) 5 P( p , p ,x ,x ,I)(l /l ) , (6)j j k k j k j k j k
where the components of this expression are as follows: (a) P( p , p ,x ,x ,I) ;j k j kw (I )iP( p /p ) is the price index of Sato (1976) and Vartia (1976), constructedij ik
i´Iover the common goods I. The weights w (I) are computed as logarithmic meansi
21of the expenditure shares of the two countries, normalized to sum to unity; (b)
l ;o p x / o p x , with the same formula applying for l .j ij ij ij ij ki´I i´Ij
The result in (6) states that the ratio of unit-costs equals the price index of thecommon or ‘overlapping’ goods (in the set I) times the additional term (l /j
1 / (s21)l ) . To interpret this term, note that l equals the proportion of thek j
expenditure on the common goods i´I relative to the entire set of goods i´I .jAlternatively, l measures one minus the expenditure share of the goods outsidej
the set I. If country j has a larger share of revenue from selling products outsidethe set of common goods, so that l ,l , that tends to lower the unit-cost ratio byj k
an amount depending on the power 1/(s21). Thus, the greater the value of unique
20Expression (4) follows directly from (3), because total expenditure equals unit-costs multiplied byoutput, or E 5c( p , I )f(x , I ).j j j j j
21Making use of the expenditure shares s (I);p x /o p x and s (I);p x /o p x theij ij ij i´I ij ij ik ik ik i´I ik ik
formula for the weights is
s (I) 2 s (I) s (I) 2 s (I)ij ik ij ik]]]] ]]]]w (I) ; YO .S D S Di ln s (I) 2 ln s (I) ln s (I) 2 ln s (I)i´Iij ik ij ik
The numerator in this expression is a logarithmic mean of s and s , and lies between these cost shares.ij ik
Then the weights w (I) are a normalized version of the logarithmic means, and add up to unity.i
84 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
products supplied by country j, the lower will be the relative cost of obtainingimport services from that country.
We rewrite the service ratio in (4) using (6):
1 / (s 21)A (E /X ) /(E /X ) lj j j k k k] ]]]]] ]5F GS DA lP( p , p ,x ,x ,I)k jj k j k (49)
1 / (s 21)5 (Product Mix) 3 (Product Variety) .
Thus, the relative services per unit import of the two countries is decomposedinto two sources. The first term on the right of (49) is the ratio of the unit-values tothe price index, or the ‘product mix’. A higher value for this term indicates thatcountry j sells relatively more of the higher-priced varieties than does country k.The second term represents the relative effect of product variety. Note that theexpenditure share ratio l /l in (49) has inverse subscripts to the quality ratiok j
A /A . Therefore, the greater is the expenditure share on varieties from country jj k
(or the smaller from country k) that are outside the set of common goods I, thehigher will be the variety index. We will interpret the product mix index as ameasure of product quality (like Aw and Roberts, 1986, 1988). Equation (49)shows that both the product mix and variety are components of A /A , the servicesj k
per unit of country j imports relative to those of country k, so both indexes arewell-motivated in terms of the preferences of importing consumers or firms.
3.2. Data and hypothesis tests
To contrast the product structures of South Korea, Taiwan and Japan in productvariety and mix, we will use disaggregate US import statistics for 1978–1988.Since the US is the largest destination market for all these countries (more than30% of Korean exports and 40% of Taiwanese exports came into the US in the lastdecade) their performances in this market should reflect the features of theirproduction quite well. We shall take each 7-digit Tariff Schedule of the UnitedStates (TSUSA) number as a variety, and then construct the product mix and
22variety indexes within each 5-digit Standard Industrial Classification (SIC). Inother words, the 5-digit SIC level is taken as the ‘industry’ for which product
23variety and mix are measured.
22The value and quantity of each 7-digit TSUSA commodity are reported in US Bureau of theCensus (1978–1988), which was obtained on magnetic tape. The price of each variety is a unit-value,computed by dividing total import value by total quantity at the 7-digit TSUSA level. A concordancefile matching TSUSA categories with import-based SIC code numbers was used to construct theproduct groups. These data are provided in a readily accessible CD-ROM format in Feenstra (1995).
23An example of a 5-digit SIC category is ‘men’s and boy’s suits, coats and overcoats’. We alsocalculated all indexes using the 8-digit SIC as the ‘industry’ level, an example of which is ‘men’s andboy’s suits’. The 5-digit and 8-digit SIC levels gave very similar results for product mix and variety;see Yang (1993).
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 85
The total sample of 1978–1988 was broken into the two periods 1978–1982 and1983–1988, to check for changes in product variety and mix that may haveoccurred. The 5-digit industries used are those with more than three varietiesexported by both countries in the full first or second period. For each of theseindustries, the product variety and mix indexes are calculated in each year. Todetermine which country dominates in product variety or mix, we compute themean of each index (measured in logs) over the years within each period, and testwhether the log index is greater or less than zero at the 10% level, using aone-sided t-test. More formally, letting z denote the log of the product variety orn t
mix index for some industry n in year t, and m denote its mean value, we assumen
that:
2z 5 m 1 ´ , where ´ is distributed N(0, s ), (7)nt n nt nt n
and t lies in the ranges 1978–1982 or 1983–1988. Then we test the hypotheses:
H : m # 0 versus H : m . 0,0 n 1 n
and also,
9 9H : m $ 0: versus H : m , 0. (8)0 n 1 n
¯ ¯9H or H are rejected if z /S.t (t21) or z /S,2t (t21), respectively,0 0 n 0.9 n 0.9
¯where z is the sample mean, S is its standard deviation and t is the number ofyears in each period. We have described this familiar test in detail because we shallgeneralize it below.
In Table 3, measuring the index as Taiwan (T) relative to Korea (K), if the nullhypothesis that m 5ln(l /l )#0 is rejected, indicating that Taiwan has greatern nK nT
expenditure on varieties not in the set of common goods, then we conclude thatTaiwan has greater product variety (denoted T .K); if on the contrary, the nullhypothesis that m 5ln(l /l )$0 is rejected, then we conclude that Korea hasn nK nT
greater product variety (denoted K.T ); and if neither of these hypotheses arerejected, then the conclusion is uncertain (denoted U ). The same is done for theproduct mix index. In Table 3, we have summarized the results of these hypothesistests by 2-digit categories, each of which contain multiple 5-digit industries.Entries in the columns marked T .K (K.T ) show the number of 5-digit industriesfor which the hypothesis m #0 (m $0) was rejected, while entries in the columnsn n
marked U are the number of 5-digit industries for which neither hypothesis wasrejected.
In addition, we shall test the joint hypothesis that all 5-digit industries within a2-digit category have a log index that is positive, or negative. Letting n index the5-digit industries, and n´N denote the 2-digit category, these joint hypotheses arestated as:
H : m # 0 for all n´N, versus H : m . 0 for some n´N,0 n 1 n
and also,
86R
.C.
Feenstra
etal.
/Journal
ofInternational
Econom
ics48
(1999)71
–100Table 3Hypothesis tests for 5-digit SIC: Taiwan versus Korea
T .K (K.T ) means the hypothesis that the Taiwan index is less (greater) than the Korean index at the 5-digit level was rejected at the 10% level; U means that boththese hypotheses could not be rejected.
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 87
9 9H : m $ 0 for all n´N, versus H : m , 0 for some n´N, (9)0 n 1 n
For example, if there are three 5-digit industries within the 2-digit category, thenthese are hypotheses on the vector m5(m , m , m ). The null hypotheses H1 2 3 0
3specifies that m must lie in the negative quadrant of R , while the alternative H13allows m to lie anywhere else in R .
The test statistics for either hypothesis in (9) is constructed as a likelihood ratiousing the model in (7). In particular, the likelihood ratio for H is constructed as:0
Table 4Hypothesis tests for 2-digit SIC: Taiwan versus Korea
Intermediate products Number of common Variety index Product mixgoods index
1980 1985 78–82 83–88 78–82 83–88
Textile mill products (22) 44 157 U(T ) U U(K) ULumber and wood products (24) 14 18 T T T TPulp and paper products (26) 7 16 U U T U(T )Chemical products (28) 9 39 U U(T ) U(T ) TStone, clay and glass products (32) 51 72 T T U UPrimary metal (33) 35 74 U K T KFabricated metal (34) 151 222 T T T TSubtotal 311 598 T–3 T–3 T–4 T–3
K–0 K–1 K–0 K–1U –4 U –3 U –3 U –3
Final productsFood products (20) 58 67 T T U(T ) UApparel and textile products (23) 376 1170 T T U UFurniture (25) — 15 — U — KPrinting and publishing (27) 19 25 T U(T ) K KRubber and plastic products (30) 29 76 U(T ) U(T ) U ULeather products (31) 93 159 T T K UIndustrial machinery (35) 17 62 U(T ) T K U(K)Electrical equipment (36) 191 236 U(T ) T U UTransportation equipment (37) 10 22 T T K KPrecision instruments (38) 71 68 U(T ) T U KMisc. manufacturing (39) 94 132 T T K USubtotal 958 2032 T–6 T–8 T–0 T–0
K–0 K–0 K–5 K–4U –4 U –3 U –5 U –7
Total 1269 2630 T–10 T–12 T–4 T–3K–0 K–0 K–5 K–5U –7 U –6 U –10 U –10
T (K) means the hypothesis that the Taiwan index is less (greater) than the Korean index for all 5-digitindustries within each 2-digit group was rejected at the 10% level; U means that these two hypotheseswere both accepted or both rejected; U(T ) and U(K) are borderline cases.
88 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
t / 22¯O (z 2 z )t nt n]]]]]]L 5P , (10)
2n´N 3 4min O (z 2 m )t nt nm #0n
¯where t denotes the number of years in each sample, and z is the sample mean ofn
z . The expression in the numerator of (10) is simply the sum of squared residualsn
¯(SSR) from (7), with z as the optimal choice for m , while the expression in then n
denominator is the SSR when the choice of m is constrained to be non-positive.n
¯The likelihood ratio L is less than unity, and will be smaller if z is positive andn
large for some n, so that forcing m #0 in the denominator substantially increasesn2the SSR. For large t, the value 22logL is asymptotically distributed as x (q),
where q is the number of industries within the 2-digit class N. Then a low value¯for L will make it more likely that H is rejected, as should occur when z is large0 n
for some n. Like the hypotheses in (8), it is possible that neither of (9) arerejected; but in contrast to (8), it is also possible that both the hypotheses in (9)are rejected.
In Table 4, we report the results of the testing hypotheses (9) when the indexes9are measured as Taiwan relative to Korea. If H (H ) is rejected at the 10% level0 0
9and H (H ) is not rejected at the 25% level, then we conclude that Taiwan0 0
(Korea) has higher product variety or mix, which is denoted by T (K). Borderlinecases occur when the first hypothesis is not rejected at the 10% level, but isrejected at the 25% level; or when the second hypothesis is not rejected at the 25%level, but is rejected at the 10% level; and these are denoted by U (for uncertain)followed by the letter of the country that has the higher index at the weakersignificance level. Cases where the hypotheses in (9) are both rejected or bothaccepted are denoted by U, indicating that the conclusion is entirely uncertain.
4. Empirical results
4.1. Taiwan-Korea comparison
4.1.1. Product varietyThe sharpest results in Table 3 are obtained for the product variety index, where
Taiwan had greater variety in more industries within each 2-digit category thanKorea. In the 5-digit industries, it had higher product variety in 42–44% of theindustries in each period (bottom of Table 3), while Korea showed greaterdiversity in only 5–8%. For the other half of the industries, the hypothesis test wasinconclusive. In the 2-digit results in Table 4, Taiwan was found to have greaterproduct variety in 10 industries for the first period and 12 industries for the second,while Korea did not show greater diversity in any of the industries during bothperiods.
A closer inspection of these results shows that the Taiwanese advantage in
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 89
product variety holds more strongly in final goods than in intermediate inputs. Thisis consistent with our first hypothesis, since the business groups in Taiwan aremainly focused in the upstream sector. Thus, in textile mill products and pulp andpaper, Taiwan has a share of business groups that exceeds that for Korea in Table2, and in both these sectors the product variety ranking in Table 4 is uncertain. Forchemical products and primary metal, the share of groups (including the state) inTaiwan and Korea are roughly comparable, and in these cases the product varietyranking in Table 4 is again uncertain (though in favor of Korea for 1983–1988 inprimary metals). In contrast, for all other intermediate sectors and final goods wefind higher product variety in Taiwan, at least by the weak hypothesis test.
Evidently, in the upstream sectors where the groups in Taiwan are strong, theirpresence offsets the tendency to find higher product variety as compared to Korea.The only exception to this is stone, clay and glass, where Taiwan has higherproduct variety despite having about the same share of business groups as inKorea. But in those sectors where the presence of business groups are markedlyless in Taiwan than in Korea, we also find higher product variety from Taiwan,providing quite strong support for our first hypothesis.
4.1.2. Product mixFrom the product mix indexes reported in Table 3, we find that Korea
specializes in higher-value final products (both consumption and capital goods),while Taiwan specializes in higher-value intermediate goods. The evidence can befound in the textile, wood, paper, and metal products industries. In textile millproducts, Korea and Taiwan had their own specialization’s in different (but aboutthe same number of) 5-digit industries, which made the 2-digit category uncertain;but Korea had a clear lead in apparel, which uses the former as the intermediateinput and creates the final products. In the lumber and wood industry, Taiwan hadhigher product mix in lumber and wood products for both periods, while Koreawas leading in furniture during the second period. The third example is paperproducts. Korea and Taiwan had their own strength in particular materials ofpaper, paperboard and paper boxes, but Korea obviously had higher product mixrelative to Taiwan in the printing and publishing industry, which is the last step tomake paper products ready to be consumed. The last case is the metal productssector. Taiwan had higher product mix in fabricated metal for both periods and inprimary metal during the first, while Korea led in industrial machinery.
By dividing industries into intermediate and final products, the respectivespecialization’s of the two countries becomes more evident. In Table 3, for the firstperiod, there are 18 intermediate industries in which Taiwan has higher productmix, versus seven in Korea; but for final products, Korea had higher product mixin 37 industries versus 17 for Taiwan. Korea moderately increased its product mixfor intermediate goods over time, and in the second period Taiwan has higherproduct mix in 19 industries versus 14 for Korea; while for final products, Koreahad higher product mix in 57 industries and Taiwan in 28. If we check this finding
90 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
with the results in Table 4, all of the 2-digit categories in which Taiwan had higherproduct mix are intermediate products (with the exception of a weak result in foodproducts), for both the first and the second period. In contrast, Korea has higherproduct mix in nearly one-half of the 2-digit final goods, with the other final goodscategories being uncertain.
These results of the product mix index can be associated with the businessgroups shares in Table 2. After adding up Taiwanese business group and state-owned shares, there are six industries whose shares are greater than 30% of thetotal sales — food (40.7%), textile mill products (50.7%), chemical materials(42.4%), stone, clay and glass products (47.6%), primary metal (30.7%) and
24transportation equipment (39.0%). Except for transportation equipment, in allother cases Taiwan was either ahead or equivalent to Korea in product mix in thefirst period, even though Korea had similar or even greater business group shares.Taiwan’s lead in some cases was overtaken by Korea in the second period,particularly in food and in primary metals, where Korea had chaebol shares of33.7% and 28.0%, respectively. Both of these were higher than the Taiwanesebusiness group shares, but lower than the sum of the Taiwanese business groupand state-owned shares.
Summarizing, the sectors in which Taiwan maintains a lead in product mix arenearly all intermediate inputs, where it also has high business groups shares. Incontrast, Korea has higher product mix in many final products, where it also hashigh chaebol shares. Thus, the presence of business groups in either case appear tobe closely related to the production of high-value product varieties, consistent withour second hypothesis.
4.2. Comparison with Japan
The product variety measure we have used is a bilateral comparison, and it isquite sensitive to the ‘overlapping’ set of products that the two countries produce.Since Korea is less diversified than Taiwan, the index for these countries reflectsKorea’s performance in exports better than Taiwan’s. To present the overallperformance of Taiwan, we need a country with broader range of exports to theUS as the benchmark, and Japan is used for that purpose. We expect that Japanshould have high product variety in its exports, due to its large size, so that therewill be a large number of ‘overlapping’ products with both Korea and Taiwan.Accordingly, in this section we present results for the Taiwan-Japan and Korea-Japan comparisons, which turn out to be in close agreement to the directTaiwan-Korea comparison.
24The transportation industry is a special case in which Taiwanese business groups’ production isconcentrated in automobile manufacturing and state-owned in shipbuilding, most of which is fordomestic consumption rather than export.
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 91
4.2.1. Variety indexJapan dominated both Taiwan and Korea in product variety. In the Taiwan-
Japan comparison, among 18 2-digit industries (Table 6, bottom), Japan hadgreater product variety in 11 industries in the first period, but Taiwan caught upslightly and narrowed down Japan’s dominance to nine industries in the secondperiod. Taiwan itself had greater product variety in two cases, and the ranking wasinconclusive in seven other industries in this latter period. The industries in whichTaiwan led — lumber and wood products, leather products, and weakly in foodproducts — were all light industries; four of the other 2-digit industries that couldnot be ranked with Japan in variety were also light industries. Japan had greatervariety than Taiwan across nearly all of the heavy industries, however. In addition,Japan had greater product variety than Korea in all 2-digit industries except leather
25products (Table 8). Higher product variety is expected from an economy that ismuch larger in its resource base, and this is confirmed by the comparison of eitherTaiwan or Korea with Japan.
4.2.2. Product mix indexFor the product mix index, reported in Table 5 and Table 7, Taiwan and Korea
each had higher product-mix than Japan within some 5-digit industries within most2-digit categories, and Japan led in other 5-digit industries. As a result, for the2-digit hypothesis tests reported in Table 6 and Table 8, neither Taiwan nor Koreacould be ranked with Japan in the vast majority of cases.
Considering the results in Table 6, during the first period Taiwan had higherproduct mix than Japan in chemicals (weakly) and in rubber and plastic industries,but this result was reversed in the second period. In heavy industries, Taiwan hadhigher product mix only in fabricated metal (second period), and weakly highermix during only one period in chemical products and transportation equipment(consisting of mainly bicycles and parts and auto parts).
For the Korea-Japan comparison in Table 8, we find that Korea had higherproduct mix in some of the light industries — food and leather products — as wellas heavy industries — industrial machinery in the first period, and transportationequipment. For the latter, Korea had an advantage in selling more higher-valuebicycles and parts, while its auto parts could not be ranked with Japan;automobiles are not included because Korea did not continuously export them inthe full first and second periods. Generally, Korea leads Japan in product mix onlyin selected final goods industries, while Japan leads in a number of theintermediate products, especially in the second period.
25The manufacture of leather products carries a social stigma in Japan (because the inputs are animalcarcasses), which explains why it has less product variety in that industry.
92R
.C.
Feenstra
etal.
/Journal
ofInternational
Econom
ics48
(1999)71
–100Table 5Hypothesis tests for 5-digit SIC: Taiwan versus Japan
T .K (K.T ) means the hypothesis that the Taiwan index is less (greater) than the Korean index at the 5-digit level was rejected at the 10% level; U means that boththese hypotheses could not be rejected.
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 93
Table 6Hypothesis tests for 2-digit SIC: Taiwan versus Japan
Industry (SI) Number of common Variety index Product mixgoods index
1980 1985 78–82 83–88 78–82 83–88
Intermediate productsTextile mill products (22) 127 248 J J U ULumber and wood products (24) 25 22 T T U UPulp and paper products (26) 17 40 U(J) U(J) J JChemical products (28) 51 195 J J U(T ) UStone, clay and glass products (32) 93 115 J J U(J) UPrimary metal (33) 45 126 J J U UFabricated metal (34) 217 289 J J U TSubtotal 575 1035 T–1 T–1 T–0 T–1
J–5 J–5 J–1 J–1U –1 U –1 U –6 U –5
Final productsFood products (20) 95 98 U U(T ) U UApparel and textile products (23) 407 1256 U U U UFurniture (25) — 16 — U — JPrinting and publishing (27) 25 27 J U(J) U(J) U(J)Rubber and plastic products (30) 61 86 J U(J) T ULeather products (31) 71 120 T T U UIndustrial machinery (35) 89 193 J J U UElectrical equipment (36) 257 297 J J U(J) UTransportation equipment (37) 26 35 U U U U(T )Precision instruments (38) 106 126 J J J JMisc. manufacturing (39) 139 164 J J U USubtotal 1276 2418 T–1 T–1 T–1 T–0
J–6 J–4 J–1 J–2U –3 U –6 U –8 U –9
Total 1851 3453 T–2 T–2 T–1 T–1J–11 J–9 J–2 J–3U –4 U –7 U –14 U –14
T (K) means the hypothesis that the Taiwan index is less (greater) than the Korean index for all 5-digitindustries within each 2-digit group was rejected at the 10% level; U means that these two hypotheseswere both accepted or both rejected; U(T ) and U(K) are borderline cases.
5. Conclusions
In his recent review of the book Trust: The Social Virtues and the Creation ofProsperity, by Francis Fukuyama, Robert Solow makes the following remarkablycandid comments:
94R
.C.
Feenstra
etal.
/Journal
ofInternational
Econom
ics48
(1999)71
–100Table 7Hypothesis tests for 5-digit SIC: Korea versus Japan
T .K (K.T ) means the hypothesis that the Taiwan index is less (greater) than the Korean index at the 5-digit level was rejected at the 10% level; U means that boththese hypotheses could not be rejected.
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 95
Table 8Hypothesis tests for 2-digit SIC: Korea versus Japan
Industry (SIC) Number of common Variety index Product mixgoods index
1980 1985 78–82 83–88 78–82 83–88
Intermediate productsTextile mill products (22) 76 226 J J U(J) ULumber and wood products (24) 12 11 J U(J) J JPulp and paper products (26) 7 20 U J J JChemical products (28) 38 90 J J U(J) JStone, clay and glass products (32) 62 75 J J U(J) JPrimary metal (33) 82 167 J J U UFabricated metal (34) 160 229 J J J JSubtotal 437 818 K–0 K–0 K–0 K–0
J–6 J–6 J–3 J–5U –1 U –1 U –4 U –2
Final productsFood products (20) 77 95 J J K KApparel and textile products (23) 333 1114 J J U UFurniture (25) 2 15 2 J 2 KPrinting and publishing (27) 20 25 J J U URubber and plastic products (30) 32 73 J J U ULeather products (31) 60 116 U K U(K) KIndustrial machinery (35) 28 81 J J K UElectrical equipment (36) 208 247 J J U UTransportation equipment (37) 10 24 J J K KPrecision instruments (38) 82 80 J J U UMisc. manufacturing (39) 102 127 J J U USubtotal 952 1997 K–0 K–1 K–3 K–4
J–9 J–10 J–0 J–0U –1 U –0 U –7 U –7
Total 1389 2815 K–0 K–1 K–3 K–4J–15 J–16 J–3 J–5U –2 U –1 U –11 U –9
Note: T(K) means the hypothesis that the Taiwan index is less (greater) than the Korean index for all5-digit industries within each 2-digit group was rejected at the 10% level; U means that these twohypotheses were both accepted or both rejected; U(T ) and U(K) are borderline cases.
. . . I, for various reasons, would like him to be right. Academic economicslikes to pretend that economic behavior is pretty much the same, always andeverywhere, almost uninfluenced by socially conditioned perceptions andnorms. If Fukuyama’s thesis could be proved to be right, it might help toloosen up the profession’s view in other contexts as well. (The NewRepublic, September 11, 1995, p. 37)
Unfortunately, the book does not meet Solow’s criterion of proof, and heconcludes:
96 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
I believe that the sorts of things that Fukuyama wants to talk about are moreimportant than my colleagues in economics are willing to admit. I wouldrather they are talked about imprecisely than not discussed at all. Butimprecision is not a virtue, and ‘for example’ is not an argument. (p. 39)
To convince economists that organizational structure matters, it is necessary topoint to objective measures of the economic performance that are affected. Wehave identified testable hypotheses on trade resulting from a model of businessgroups: in an economy with highly integrated business groups, there should be lessproduct variety of final goods, but higher product quality. These hypotheses applyto a comparison of like-sized economies, such as South Korea and Taiwan, whereKorea has larger and more integrated groups. In a larger economy, such as Japan,we expect there to be greater product variety due to its size, and do not make anyspecific prediction concerning product quality.
We have tested these hypotheses using data on US imports of highly disaggre-gate products from South Korea, Taiwan and Japan. Applying index-numbertechniques from Feenstra (1994), an index of product variety was constructed atthe 5-digit SIC industry level to reflect the range of products sold from eachcountry. In addition, a ‘product mix’ index was constructed to reflect whether eachcountry tends to export high-priced or low-priced products within each 5-digitcategory. This index can be interpreted as a measure of product quality, and wehave shown that both indexes are theoretically related to the consumption servicesprovided per unit of imports.
In the results, we have found that Taiwan exports a greater variety of productsto the US than Korea, and this holds particularly for final goods. This fits thepattern of business groups in the two economies, which are located principally inthe upstream sectors of Taiwan. It also fits the anecdotal evidence that Taiwantends to focus on niche markets for its products, while changing product design tomeet the demands of overseas purchasers. In contrast, Korea tends to aggressivelymarket particular products (such as the Hyundai car, or microwave ovens), withhigh volume production but more limited product variety. Japan exceeds either ofthese countries in product variety, which can be explained by its very large size.The results also show that Taiwan exports relatively more high-priced intermediateinputs, whereas Korea exports relatively more high-priced final goods.
We feel that these results confirm the importance of market structure indetermining trade patterns, and also demonstrate the usefulness of using businessgroups as a measure of market structure. Business groups of various types arefound in many other Asian and Western countries, and lead to large differences inmarket structure, as described by Caves (1989). It can be hoped that theseinternational comparisons may be used to more fully determine the impact ofmarket structure on international trade, and on other aspects of economicperformance. One direction for further research would be to explore whetherdiffering level of product variety across Taiwan and Korea (or other countries)
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 97
have an impact on the productivity growth. The hypothesis would be thatproductivity growth should be higher in Taiwan, reflecting the greater range ofproduct varieties developed there. Some evidence in support of this hypothesis isobtained by Feenstra et al. (1997b). More generally, the business groups in Asiaand elsewhere offer a rich testing ground for hypotheses relating market structureand economic performance, and we hope that the results in this paper havedemonstrated their importance, while also suggesting directions for furtherresearch.
Acknowledgements
The research on this paper has been supported by the Pacific Rim Business andDevelopment Program (PACBAD) at the Institute of Governmental Affairs,University of California, Davis. Financial support from the Ford Foundation isgratefully acknowledged.
References
Aoki, M., 1990. Towards an economic model of the Japanese firm. Journal of Economic Literature 28(1), 1–27.
Aoki, M., Patrick, H., 1994. The Japanese Main Bank System. Oxford University Press, New York.Amsden, A.H., 1989. South Korea and Late Industrialization. Oxford University Press, New York.Aw, B.Y., Roberts, M.J., 1986. Estimating quality change in quota-constrained import markets: the case
of US footwear. Journal of International Economics 21, 45–60.Aw, B.Y., Roberts, M.J., 1988. Price and quality comparisons for US footwear imports: an application
of multilateral index numbers. In: Feenstra, R.C. (Ed.), Empirical Methods for International Trade.MIT Press, Cambridge, pp. 257–275.
Bhagwati, J.N., 1992. The fraudulent case against Japan. The Wall Street Journal, January 6.Biggart, N.W., 1990. Institutionalized patrimonialism in Korean business. Comparative Social Research
12, 113–133.Caves, R.E., 1989. International differences in market structure. In: Schmalensee, R., Willig, R.D.
(Eds.), Handbook of Industrial Organization, vol. II, chap. 21. North-Holland, Amsterdam, pp.1225–1250.
China Credit Information Service, 1983. Business groups in Taiwan 1983–84. China Credit In-formation Service, Ltd., Taipei.
Chou, T.-C., 1985. Industrial Organization in the Process of Economic Development: The Case ofTaiwan, 1950–1980. Ciaco, Louvain-la-Neuve.
Davis, D.R., 1995. Intra-industry trade: a Hecksher-Ohlin-Ricardo approach. Journal of InternationalEconomics 39 (3 /4), 201–226.
Dyer, J.H., 1998. Vertical keiretsu alliances and asset specialization: a new perspective on Japaneseeconomic success. In: Fruim, W.N. (Ed.), Networks and Markets: Pacific Rim Investigations. OxfordUniversity Press, New York, in press.
Eccles, R.G., 1985. The Transfer Pricing Problem: A Theory for Practice. Lexington Books, Lexington,MA.
98 R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100
Egan, M.L., Mody, A., 1992. Buyer-seller links in export development. World Development 20 (3),321–334.
Feenstra, R.C., 1994. New product varieties and the measurement of international prices. AmericanEconomic Review 84 (1), 157–177.
Feenstra, R.C., 1995. US imports, 1972–1994: data and concordances. NBER Working Paper no. 5515,March, with accompanying CD-ROM.
Feenstra, R.C., 1997. Business groups in South Korea and Taiwan: a database and comparison. PacificRim Business and Development Program, Working Paper no. 24, Institute of Governmental Affairs,University of California, Davis.
Feenstra, R.C., Yang, T.-H., Hamilton, G.G., 1993. Market structure and international trade: businessgroups in East Asia. NBER Working Paper no. 4536, November.
Feenstra, R.C., Huang, D.-S., Hamilton, G.G., 1997. Business groups and trade in East Asia: Part 1,networked equilibria. NBER Working Paper no. 5886, January.
Feenstra, R.C., Madani, D., Yang, T.-H., Liang, C.-Y., 1997. Testing endogenous growth in SouthKorea and Taiwan. NBER Working Paper no. 6028.
Fung, K.C., 1991. Characteristics of Japanese industrial groups and their potential impact on US-Japanese trade. In: Baldwin, R.E. (Ed.), Empirical Studies of Commercial Policy. Univ. of Chicagoand NBER, Chicago, pp. 137–164.
Fung, K.C., 1997. International trade and bank groups: welfare-enhancing or welfare-reducing.Manuscript, University of California, Santa Cruz, July.
Fung, K.C., Friedman, D., 1996. International trade and the internal organization of firms: anevolutionary approach. Journal of International Economics 41 (1 /2), 113–137.
Futatsugi, Y., 1986. Japanese enterprise groups. Kobe University, School of Business, Kobe.Gerlach, M., 1992. Alliance Capitalism: The Strategic Organization of Japanese Business. University of
California Press, Berkeley.Gilson, R.J., Kraakman, R., 1993. Investment companies as guardian shareholders: the place of the
MSIC in the corporate governance debate. Stanford Law Review 45, 985–1010.Granovetter, M., 1994. Business groups. In: Smelser, N., Swedberg, R. (Eds.), Handbook of Economic
Sociology. Princeton University Press, Princeton, pp. 453–475.Granovetter, M., 1995. Coase revisited: business groups in the modern economy. Industrial and
Corporate Change 4 (1), 93–130.Greenhalgh, S., 1988. Families and networks in Taiwan’s economic development. In: Winckler, E.,
Greenhalgh, S. (Eds.), Contending Approaches to the Political Economy of Taiwan. M.E. Sharpe,Armonk, NY, pp. 224–245.
Hamilton, G.G., 1988. Patterns of Asian capitalism: The cases of Taiwan and South Korea. Program inEast Asian Culture and Development Research. Working Paper no. 28, Institute of GovernmentalAffairs, University of California, Davis.
Hamilton, G.G. (Ed.), 1991. Business Networks and Economics Development in East and SoutheastAsia. Center of Asian Studies, Hong Kong.
Hamilton, G.G., Biggart, N.W., 1988. Market, culture, and authority: a comparative analysis ofmanagement and organization in the far east. American Journal of Sociology, Special Issue onEconomic Sociology 94, S52–S94.
Hamilton, G.G., Zeile, W., Kim, W.-J., 1990. The network structures of East Asian economies. In:Clegg, S.R., Redding, S.G. (Eds.), Capitalism in Contrasting Cultures. de Gruyter, Berlin, pp.105–129.
Hamilton, G.G., Kao, C.-S., 1990. The institutional foundations of Chinese business: the family firm inTaiwan. Comparative Social Research 12, 135–157.
Helpman, E., Krugman, P., 1985. Market Structure and Foreign Trade. MIT Press, Cambridge.Helpman, E., 1987. Imperfect competition and international trade: evidence from fourteen industrial
countries. Journal of the Japanese and International Economies 1 (1), 62–81.Hoshi, T., Kashyap, A., Sharfstein, D., 1990. The role of banks in reducing the costs of financial
distress in Japan. Journal of Financial Economics 27, 67–88.
R.C. Feenstra et al. / Journal of International Economics 48 (1999) 71 –100 99
Hoshi, T., Kashyap, A., Sharfstein, D., 1991. Corporate structure, liquidity, and investment: evidencefrom Japanese industrial groups. Quarterly Journal of Economics 106 (1), 33–60.
Hummels, D., Levinsohn, J., 1993. Product differentiation as a source of comparative advantage?.American Economic Review, Papers and Proceedings 83 (2), 445–449.
Hummels, D., Levinsohn, J., 1995. Monopolistic competition and international trade: reconsidering theevidence. Quarterly Journal of Economics 101 (3), 799–836.
Jensen, P.E., 1995. Monopolistic competition and bilateral patterns of international trade. Manuscript,Pennsylvania State University.
Kim, E.M., 1993. Contradictions and limits of a developmental state: with illustrations from the SouthKorean case. Social Problems, May.
Kim, E.M., 1991. The industrial organization and growth of the Korean chaebol: integratingdevelopment and organizational theories. In: Hamilton, G.G. (Ed.), Business Networks andEconomic Development in East and Southeast Asia. Center of Asian Studies, University of HongKong, pp. 272–299.
Kim, E.M., 1997. Big business, strong state: collusion and conflict in Korean development, 1960–1990.State University of New York Press, Albany.
Lawrence, R.Z., 1991. Efficient or exclusionist? The import behavior of Japanese corporate groups.Brooking Papers on Economic Activity, pp. 311–341.
Lincoln, J.R., Gerlach, M.L., Ahmadjian, C.L., 1994. Keiretsu networks and corporate performance inJapan. Manuscript, Haas School of Business, University of California at Berkeley, December.
Martins, J.O., 1992. Export behaviour with differentiated products: exports of Korea, Taiwan and Japanto the US domestic market. In: Dagenais, M.G., Muet, D.A. (Eds.), International Trade Modelling.Chapman and Hall.
Numazaki, I., 1986. Networks of Taiwanese big business: a preliminary analysis. Modern China 12,487–534.
Numazaki, I., 1991. The role of personal networks in the making of Taiwan’s guanxiqiye, (relatedenterprises). In: Hamilton, G.G. (Ed.), Business Networks and Economic Development in East andSoutheast Asia. Center of Asian Studies, University of Hong Kong, pp. 77–93.
Orru, M., Hamilton, G.G., Suzuki, M., 1990. Patterns of inter-firm control in Japanese business.Organizational Studies 10 (4), 549–574.
Orru, M., Biggart, N.W., Hamilton, G.G., 1991. Organizational isomorphism in East Asia: broadeningthe new institutionalism. In: Powell, W.W., DiMaggio, P.J. (Eds.), The New Institutionalism inOrganizational Analysis. Univ. of Chicago Press, Chicago, pp. 361–389.
Rauch, J., 1997. Networks versus markets in international trade. Journal of International Economics, inpress.
Rodrik, D., 1993. Industrial organization and product quality: evidence from South Korean andTaiwanese exports. In: Krugman, P., Smith, A. (Eds.), Empirical Studies of Strategic Trade Policy.Univ. of Chicago and NBER, Chicago, pp. 195–210.
Sato, K., 1976. The ideal log-change index number. Review of Economics and Statistics 58, 223–228.Saxonhouse, G., 1993. What does Japanese trade structure tell us about Japanese trade policy?. Journal
of Economic Perspectives 7 (3), 21–43.Steers, R.M., Shi, Y.K., Ungson, G., 1989. The chaebol: Korea’s new industrial might. Harper and
Row, New York.US Bureau of the Census, 1978–1988. US general imports for consumption, schedule a, commodity by
country, FT135. US Department of Commerce, Government Printing Office, Washington, DC.Vartia, Y.O., 1976. Ideal log-change index numbers. Scandinavian Journal of Statistics 3, 121–126.Vancil, R.F., 1978. Decentralization: managerial ambiguity by design. Dow Jones-Irwin, Homewood,
IL.Wade, R., 1990. Governing the Market. Princeton University Press, Princeton.
Yang, T.-H., 1993. Industrial structure and trade patterns: evidence from South Korea and Taiwan.Ph.D. dissertation, University of California, Davis.
Zeile, W., 1991. Industrial policy and organizational efficiency: the Korean chaebol examined. In:Hamilton, G.G. (Ed.), Business Networks and Economic Development in East and Southeast Asia.Center of Asian Studies, Univ. of Hong Kong, pp. 300–329.
