Journal of International Economics 76 (2008) 403–415

Contents lists available at ScienceDirect

Journal of International Economics

j ourna l homepage: www.e lsev ie r.com/ locate /econbase

Erratum

Firm heterogeneity and location choice of Taiwanese multinationals

Bee Yan Awa,⁎, Yi Lee b

a The Pennsylvania State University, United Statesb Chung-Hua Institution for Economic Research (CIER), Taiwan

a r t i c l e i n f o

DOIs of original article: 10.1016/j.jinteco.2008.02.0⁎ Corresponding author.

E-mail address: byr@psu.edu (B.Y. Aw).1 Sales of multinational firms exceeded that of expo2 An exception is Debaere's (2004) discussion on FD

0022-1996/$ – see front matter © 2008 Elsevier B.V.doi:10.1016/j.jinteco.2008.06.005

a b s t r a c t

Article history:Received 4 July 2007Received in revised form 24 February 2008Accepted 25 February 2008

This paper examines the extent to which production location decisions of Taiwanesemultinationals reflect underlying patterns of firm productivity. In our theoretical model,heterogeneous firms in a middle-income country decide on the optimal production locationsfor serving three geographically separate markets: domestic, foreign high-income and foreignlow-income. The model shows that the equilibrium decision of a firm depends on the fixedinvestment costs of establishing foreign subsidiaries, production costs, transportation costs,market size and its own productivity level.Using firm-level data in 2000, Taiwanese electronics firms are divided into four differentcategories: non-FDI, investors in China only, investors in the U.S. only, investors in both Chinaand the U.S. We use a multinomial logit model to link firms' location choices with theirproductivity, controlling for country, industry and other firm characteristics. Our empiricalresults are consistent with the predictions of the theoretical model. We show that moreproductive firms engage in outward FDI, with the most productive ones investing in both Chinaand the U.S. We also provide evidence indicating that Taiwanese multinationals investing onlyin the U.S. are more productive than those investing exclusively in China due to smaller fixedinvestment costs in China relative to the U.S.

© 2008 Elsevier B.V. All rights reserved.

Keywords:Outward FDIFirm heterogeneityLocation choices

1. Introduction

In the decade of the 1990s, worldwide sales of multinational firms expanded at rates that outpaced the rapid growth of trade inmanufactured goods.1 By 2001, the gross product of foreign affiliates accounted for more than 11% of world GDP. This rapidexpansion has triggered much research into the factors underlying the investment patterns of multinational firms. However, thebulk of recent research on outward foreign direct investment (FDI) has been at the aggregate level and restricted to developedcountries (DCs) like the U.S. and Japan.2 This paper uses micro-level data for Taiwan, a developing country that has experiencedextraordinary growth in outward FDI since 1990, to shed light on the extent to which production location decisions of Taiwanesemultinationals reflect underlying patterns of firm productivity.

A rapidly growing area of research in the literature on FDI in recent years has been in incorporating firm heterogeneity withinindustries into simple models of international investments. In particular, Helpman et al. (2004) focus on heterogeneous firmsserving consumers in foreign markets through exports or horizontal FDI. They model a firm's decision between exports and FDI asdriven by the proximity-concentration trade-off whereby FDI incurs higher fixed costs but involves lower transportation costsrelative to the export activity. Their model predicts that the least productive firms serve only the domestic market, the relativelymore productive firms export and the most productive firms engage in FDI. The authors provide empirical evidence at the industry

02, 10.1016/j.jinteco.2008.06.004.

rts by about 7% per year from 1990 to 2000 (UNCTAD, 2002).I in Newly Industrialized Economies (NIEs) in the 1990s.

All rights reserved.

404 B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

level to support the sorting pattern predicted by their theoretical model. Yeaple (2007) takes the Helpman et al. paper one stepfurther by incorporating firm heterogeneity as well as geographic sorting into the structure of U.S. multinational activity acrossindustries and countries. He shows that more productive firms are more likely to invest abroad, but firm productivity plays asmaller role when the destination country is large.3

This paper is based on a very recent theoretical paper by Grossman et al. (2006) in which the authors develop a complementarystrategymodel incorporatingfirmheterogeneity as a determinant of horizontal and vertical FDI. They show that evenwithin the sameindustrywhere eachfirm faces the samemarket size,fixed costs and transportation costs,firmsdiffering inproductivity have differentoptimal strategies. In their model, a firm performs intermediate stages of production in a country with low production costs andsubsequent stages (such as assembly) in another country close to the finalmarket to save on transportation costs. Theirmodel has twoidentical Northern countries and one Southern country with low production costs and small market size. Each firm chooses theproduction sites for intermediate and assembly stages in serving the global market. Grossman et al. conclude that the least productivefirms produce in the home market, firms engaging in FDI are more productive, and the most productive firms will move bothintermediate and assembly stages into the South. Integration strategies dependon the scale of trading costs and the relative fixed costsof the intermediate and assembly stages. In the casewhere trading/transportation costs offinal goods are low,firmswill not produce inthe North. As the transportation costs on final goods increase, firms are going to spread out assembly into all final markets. Finally,when there are transportation costs on intermediates, firms prefer to combine these two stages in the same country.

We develop a modification of the Grossman et al. (2006) framework to explain integration strategies of multinational firms,headquartered in a middle-income country, choosing production locations in developed countries (North) or developing countries(South) for their global sales. Existingmodels which focus on FDI flows from the North to the South or between Northern countriescannot adequately explain the outward FDI activities of multinational firms in middle-income countries such as Taiwan and SouthKorea, where wages are sandwiched between those in the North and South.4 We introduce a simple one-stage production modelwhere firms investing in and serving the Northern market are able to save on transportation and trading costs while thoseinvesting in and selling in the market in the South conserve on both production and trading costs. This model captures bothmarket-seeking and resource-seeking incentives for multinational firms. In particular, it focuses attention on the trade-off in thecosts associated with domestic production for export versus moving the production overseas to locations with varying set-upcosts. Our theoretical model shows that a firm's location decision depends on industry characteristics, such as fixed investmentcosts of building a foreign plant, market size, transportation and trading costs, as well as firm productivity.

Given that fixed investment costs and product substitutability are likely to differ most significantly across industries, weestimate the location decisionmodel separately for the Taiwanese Computer and Telecommunications Equipment industry and theelectronic Parts and Components industry within the electronics subsector. The empirical findings of the paper are consistent withthe predictions of our theoretical model. We show that more productive firms engage in outward FDI, with the most productiveones investing in multiple countries. We also provide evidence indicating that Taiwanese multinationals investing only in the U.S.are more productive than those investing exclusively in China due primarily to smaller fixed investment costs in China relative tothe U.S.

This paper makes two main contributions to the literature. First, we develop a three-country model that accounts for theinterdependence between host country and other final consumption countries. Our model resembles those of Ekholm et al. (2007)and Helpman et al. (2004). The key difference is that we explicitly model the effects of firm heterogeneity and the effect of countrycharacteristics on firms' location choices. By introducing firm heterogeneity into Ekholm et al.'s framework, we allow firms withdifferent productivity levels to choose different production locations based on factor price differentials, fixed investment costs andmarket size across countries. In contrast to Helpman et al. (2004), we introduce the strategy of exporting from a third country.

Second, unlike existing empirical work which primarily focuses on investments flowing from North to South countries, oramong countries in the North, we introduce a middle-income country that is just as likely to invest in the South as it is to invest inthe North. More specifically, by simplifying Grossman et al.'s (2006) two-stage production model to a one-stage model, weconcentrate on the production destinations of firms from Taiwan to China and the U.S. This model allows us to take full advantageof the unique information in the data set on the destination of outward FDI among Taiwanese firms. GivenTaiwan's middle-incomestatus, one of our objectives here is to provide insights into the economic rationale for the growing trend towards outward FDIamong firms in a rapidly growing developing country.

Section 2 presents some background of Taiwanese FDI in the 1990s and the data used for the empirical estimation. Section 3develops a theoretical model for firms' production location choices. The empirical counterpart to the theoretical model ispresented in Section 4. Section 5 reports the estimation results. The summary and conclusions are presented in the final section.

2. Data description and the pattern of Taiwanese outward FDI

This paper is based on plant-level data of the Taiwanese manufacturing sector in the year 2000. This data was collected by theMinistry of Economic Affairs (MOEA) in Taiwan. In addition to reporting revenue, total employment, and R&D expenditures, thedata set also includes a unique firm identification number for each plant so that we can identify the owner of each plant. While this

3 Other recent papers include Girma et al. (2005) on U.K. firms and Head and Ries (2003) on Japanese firms.4 Debaere (2004) illustrates how explicitly accounting for a middle income country (South Korea in his case) in the decision of Korean multinationals involves

different incentives and yields interesting results regarding the effects of the reallocation of their activities between the labor abundant South and capitalabundant North.

Table 1Distribution of sales by Taiwanese affiliates (percent), by destination country (2003)

Host country Firms (number) Back to Taiwan Local sales Export to third country

All ManufacturesU.S. 134 9.14 74.47 16.39China 1172 17.59 47.27 35.14Europe + Japan 31 11.71 62.24 26.05NIEs 79 20.47 47.03 32.50S.E. Asia 171 12.20 46.27 41.53

1. Computer and telecommunications equipmentU.S. 49 2.04 80.98 16.98China 121 21.55 38.56 39.89Europe + Japan 10 13.50 57.20 29.30NIEs 15 10.26 42.47 47.27S.E. Asia 7 24.29 55.71 20.00

2. Parts and componentsU.S. 34 8.21 81.18 10.61China 190 29.05 41.53 29.42Europe+Japan 5 14.00 49.00 37.00NIEs 25 24.32 41.80 33.88S.E. Asia 11 16.18 37.64 46.18

Source: Outward FDI Survey in Manufacturing, the Ministry of Economic Affairs (2004).NIEs: Hong Kong and Singapore. South Korea is excluded since the MOEA does not collect data for firms undertaking FDI in South Korea. South Korea accounts forless than 7% of Taiwan's total value of FDI stock in the NIEs.

405B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

data is collected annually, the data for the year 2000 is unique in that theMOEA also collected information on a plant's outward FDIby country of destination.

Since the FDI decision is made at the firm-level, we constructed firm-level data by using the ownership information in the dataset to sum up the values of the variables across plants.5 When a firm owns multiple plants in different industries, we define themain industry of the firm based upon the industry of the plant with the largest revenue share.

The top section of Table 1 presents some stylized facts of a sample of Taiwanesemultinational firms that engage in outward FDI.This sample comprises 75% of all Taiwanese multinationals in the manufacturing sector. It summarizes the sales of manufacturedgoods by foreign affiliates of Taiwanese multinationals in different host countries. Affiliates' sales are broken down into exportsales back to Taiwan, local sales in the host country, and export sales to third countries. One striking feature of Taiwanese affiliatesin the U.S. is that 74% of their output is sold in the U.S. market, possibly indicating the importance of conserving transportationcosts. In contrast, affiliates in other Asian countries engage in both local sales and export to third countries, suggesting thatconserving both transportation costs and production costs play important roles. These stylized facts are consistent with ourtheoretical model which introduces an export-platform strategy for heterogeneous firms.

Within the Taiwanesemanufacturing sector, the electrical and electronics subsector constitutes the largest share of outward FDIthroughout the decade of the nineties, replacing textiles and plastics in the previous decade. By 2000, the electrical and electronicssubsector accounted for 40% of the stock of outward FDI in the manufacturing sector and 40% of total export orders. In the face ofintense international competition in the subsector and the absence of cutting-edge skills and their own brands, Taiwanese firmspurposefullymoved their operations overseas.Within the electronics subsector, between 77% and 89% of Taiwanesemultinationalsoperate in the same two-digit industry as the parent firms in Taiwan.6 Among FDI types, horizontal FDI is much more prevalentthan vertical FDI. For instance, more than four-fifths of firms in both the Computer and Telecommunications Equipment industryand the Parts and Components industry undertake horizontal FDI rather than vertical FDI.7

The top two overseas investment destinations for Taiwanese manufacturing firms in the 1990s are China and the U.S.8 In termsof single country destinations, China attracts more Taiwanese FDI firms than any other country with the U.S. a distant second andthese firms are well represented across all industries with the bulk concentrated in Machinery and Electronics industries. Amongadvanced countries, the U.S. is the distant front runner with more than half of the FDI firms in the U.S. in the Computer andTelecommunications and Parts and Components industries. These two industries have the most number of firms among the smallgroup of firms with foreign affiliates in both China and the U.S. Given our interest in firms with investments in multiple locations,our industry choice for our empirical estimation was limited to the Computer and Telecommunications and the Parts andComponents industries.

5 Only 5% of all manufacturing firms are multi-plant firms. Around 10% of the Computer and Telecommunications Equipment industry, and 8% of the electronicParts and Components industry are multi-plant firms.

6 In contrast to the parent firm, information on characteristics of affiliates is limited. In 2000, of the 48 and 52 parent firms in the Computer andTelecommunications industry investing in the U.S. and China, respectively, 77% and 85% of them have affiliates in the electronics and electrical subsector. Of the29 and 129 parent firms in the Parts and Components industry investing in the U.S. and China, respectively, 83% and 89% of them have affiliates in the electronicsand electrical subsector.

7 These figures are from Outward FDI Survey in Manufacturing, MOEA (2001).8 The MOEA started to record Taiwanese outward FDI in China in 1992.

Table 2Descriptive statistics of firms in the electronics industry by investment location, 2000

Computer and telecommunications equipment Parts and components

Firm Characteristics Domestic Non-FDI (Exporter) China U.S. Both Domestic Non-FDI (Exporter) China U.S. Both

Firms (number) 868 473 73 40 16 1591 713 130 34 7Labor Productivity (NT$ per worker) − .29 .25 .36 .61 .77 − .23 .32 .47 1.13 .89Size (no. of workers) 2.75 3.55 4.44 4.85 5.12 2.65 3.83 4.25 5.46 5.47R&D Intensity (% of revenue) 1.57 4.30 3.34 8.61 8.15 0.60 1.91 2.01 3.62 5.54Foreign Technology Purchasing Intensity(% of revenue)

0.06 0.08 0.07 0.34 0.46 0.02 0.2 0.03 0.30 0.3

Age (years) 3.98 4.59 5.20 3.98 4.63 5.56 6.63 8.15 4.24 7.57

Notes: Labor productivity is measured as the mean corrected total revenue and net inventory change divided by the mean corrected total employment.R&D Intensity is defined as expenditures on R&D and Domestic Technology Purchases as a fraction of total firm revenue.

406 B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

The lower section of Table 1 summarizes the distribution of the percentage of total sales in the Computer and Telecommunicationsand Parts and Components industries by foreign affiliates of Taiwanese multinationals in different host countries. The pattern isconsistentwithourearlier observation at themore aggregate level. Over 80%of total output producedby Taiwanese affiliates in theU.S.is sold in the U.S. market while affiliates in China and other Asian countries engage in both local sales and export to third countries.

This paper focuses on horizontal FDI investments in the Computer and Telecommunications and Parts and Componentsindustries by Taiwanese multinationals in China and the U.S. We classify firms' investment types into four categories: non-FDI, FDIin China, FDI in the U.S., and FDI in both the U.S. and China. Exporting firmswithout any foreign affiliate are included in the non-FDIgroup.9 Firms belonging to the second and third groups go abroad only to China or the U.S. The last category includes Taiwanesemultinationals active only in both the U.S. and China. In order to focus on firms' location choices between China and the U.S., weignore firms that also invested in other countries besides the U.S. and China.10

Table 2 provides some descriptive statistics of firms in the two electronics industries by different destination countries. Thedetails on the construction of these variables are reported in Appendix A. Table 2 indicates that FDI firms are more productive,larger, older, more R&D-intensive and engage in more intensive purchases of foreign technology relative to non-FDI firms.However, among FDI firms, firms investing only in China on average have the lowest values for all these performance indicatorsexcept for age. Firms investing in China average 5.2 years and 8.2 years in the Computer and Telecommunications and Parts andComponents industries respectively, while those investing in the U.S. in both industries average about 4 years.

3. The model

We develop a simple model in which firms engaging in production activities, face decisions regarding where to locate theirplant in order to sell their products globally. Firms serve each market by either producing in their foreign subsidiary or exportingfrom another country. A firm pays a fixed investment cost for building a plant outside the home country. If it chooses to export,there is an iceberg transportation cost. Firmsmake their production location decisions by comparing their variable costs with fixedinvestment costs. To capture the fact that the bulk of Taiwanese multinational firms conduct FDI in China and/or the U.S., weconstruct a three-country model to analyze the determinants of multinational firms' location choices.

Firms in the Home market are willing to offer final goods in three countries: Home (H), North (N) and South (S). Each firmproduces a single differentiated product.11 Consumers in all countries have the same CES utility function in the consumption ofproduct i, qi, such that:

9 We10 This11 Mudefinedperspecamongfollowinafter un

U ¼Xi¼1

nqai

!1a

; 0bab1 ð1Þ

th this utility function, we can derive the demand function of product i for consumers in country k as:

Wi

qik ¼ Ykp�rik where r ¼ 1

1� aN1; Yk ¼

EkP~k

; P~k ¼

Xnki¼1

p1�rik

!ð2Þ

re the country index k=N, H, S represents North, Home and South, σN1 is the elasticity of substitution between any pair of

Hegoods, Ek is the total expenditure in country k and P k is the price index for all products in country k.

exclude pure domestic firms since we are primarily interested in firms participating in foreign markets.results in a loss of only 6.9% of firms in the electronics subsector.

ltiproduct production is generally not prevalent in the Taiwanese manufacturing sector. Specifically, in the two industries under study, multiproduct firms,at the 3-digit level, comprise about 15 percent of the observations. Baldwin and Ottaviano (2001) examine the export versus FDI decision from thetive of a multiproduct firm. They show that multiproduct firms have an additional incentive to locate production abroad to reduce intra-firm competitiontheir products (cannibalization effect). Their model implies that in the presence of multiproduct production, parent firms will reduce their product lineg FDI. In the Taiwanese electronics industry, we find that less than 8% of Taiwanese multinationals reduce the number of products in their product linedertaking FDI in 2000.

407B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

Countries differ in their market sizes Ek, factor prices wk and fixed investment costs fk. We assume fH=0. Firms areheterogeneous in their productivity level ρi, which is firm-specific such that each subsidiary belonging to a firm is endowed withthe same productivity level in different production locations. We also assume that production only requires labor input in theproduction function, qi=ρili. This production technology incurs a constant marginal production cost,MCik ¼ wk

qi , of product i, whichis produced by a subsidiary in country kwith productivity level, ρi. The average hourly wage in Taiwan is sandwiched between thatin the U.S. at the higher end and China at the lower end.12 These stylized figures are consistent with the basic premise of our modelthat the wage level is highest in the North, lowest in the South and intermediate in Home, i.e. wNNwHNwS. If a firm chooses toexport, it incurs a symmetric transportation cost, tN1.13

The optimal pricing rule for a CES-induced demand function is Pik ¼ Cika , where Cik=MCik if firm i serves market k from its local

subsidiary and Cik=MCijt if this firm produces in country j and then exports final goods to market k. Solving the profit, πik, for firm iserving country k, we obtain

12 Acc13 If trlong asThese twTaiwan.14 ThisS countpattern15 The16 Thuin the Hto both

pik ¼ pik � Cikð Þqik � fk ¼ 1� að ÞYkC1�rik 1=að Þ1�r�fk ¼ Bk w~kð Þ1�rhi � fk ð3Þ

where Bk≡ (1−α)Yk(1/α)1−σ and θi≡ρiσ−1. If firm i produces and serves in country k, then w~k=wk. If firm i produces in country j

and exports to country k, then w~k=wjt.To simplify notations in the model, we normalize the wage in Home to one (wH=1) and re-scale foreign market sizes relative to

the Home market size by BH=B, BS=βSB and BN=βNB. We define (x, y, z) as the choice set from serving the Home, North and Southmarkets in production locations x, y and z, respectively. For example, (H, H, S) describes a firmwith plants in Home and South. Thisfirm sells final goods to the Home and North markets from the subsidiary at Home and to the South market from the plant locatedin the South.

If goods selling in each market can be produced from any of these three countries, there will be 27 (3×3×3) possible locationcombinations. Many of the combinations will not be chosen in the model. Given that wNNwHNwS, the presence of symmetrictransportation cost and the fixed cost of establishing a plant in a foreign country, we know that the Home and South markets willnever be served by a plant in the North.14 That is, we can eliminate 15 of the 27 strategies and consider only 12 strategies(2×3×2).15 The existence of a local plant in any of the three markets implies that it is most profitable for the local, rather thanforeign plants to serve those markets. In particular, since our observations are of Home plants, the Home market is best served bythese plants.16 In addition, a firmwith a plant in the South will use this plant, not one in Home, to serve the Southmarket, ensuringthat (H, S, S) will always dominate (H, S, H).

In sum, we can eliminate 7 of the above 12 strategies andwill need to only consider the following 5 strategies: (H,H,H), (H,N,H),(H,N,S), (H,S,S) and (H,H,S). Finally, the strategy (H,H,S) is dominated by (H,S,S) since, in the absence of a plant in the North, it ischeaper to serve the North market from the South rather than Home given that wages are lower in the South and transportationcosts are symmetric.

The profit functions of the remaining 4 strategies are as follows:

pHHH ¼ Bh 1þ bNt1�r þ bSt

1�r� � ð4Þ

h i

pHSS ¼ Bh 1þ bN wStð Þ1�rþbS wSð Þ1�r � fS ð5Þ

pHNH ¼ Bh 1þ bN wNð Þ1�rþbSt1�r

h i� fN ð6Þ

pHNS ¼ Bh 1þ bN wNð Þ1�rþbS wSð Þ1�rh i

� fN þ fSð Þ ð7Þ

ll four strategies, the Home country produces and sells its output in the local market. In the first strategy (Eq. (4)) the Home

In acountry also exports the final goods to both the North and the South. The second strategy (Eq. (5)) combines elements of“horizontal FDI” and “export-platform FDI”: firms produce in the South to serve the local (South) market and export part of theirproduction to the North. The key motivation for foreign investment by the Home country is the low wages in the South. The thirdstrategy (Eq. (6)) describes the market-access incentive for a firm doing business in the North. A firm builds a plant in the North toconserve on transportation costs rather than production costs. The final strategy (Eq. (7)) is one of complete FDI, whereby the final

ording to IMD (2000), the total hourly compensation for manufacturing in 1999 is $18.56 in the U.S., $5.45 in Taiwan and $2.11 in China.ansportation costs are not symmetric, it is clear from the profit function that we obtain the same result as in the symmetric transportation cost case aswNtNSN tHS and tHNNwStNS, where tHN, tHS, and tNS represent the transportation costs between North and Home, South and Home, and North and South.o conditions imply that serving China from Taiwan is more profitable than from the U.S. and serving the U.S. market from China is cheaper than from

would not be the case without modeling a middle income country with wages sandwiched between N and S. For example, in models with two N and oneries where wages are symmetric in H and N, then πHNN equals πHNH and Taiwanese multinationals (H) will build a plant in N to serve the S market, ainconsistent with our observations reflected in Table 1.12 strategies are (H,H,H), (H,H,S), (H,N,H), (H,N,S), (H,S,H), (H,S,S), (S,H,H,), (S,H,S), (S,N,H), (S,N,S), (S,S,H), and (S,S,S).s, the starting point in our paper is to take as given the existence in the Home country of local plants. In theory it is possible for firms not to produce at allome country (i.e. if wages in the South are low enough (wStb1) then a firm would choose the strategy (SSS), moving its plant to the South and exportingthe North and the Home markets.).

Fig. 1. Profit function of different strategies.

408 B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

goods are produced and sold in each of the three markets. The second and third strategies are consistent with the stylized facts,described in Table 1, that Taiwanese firms invest in China for local and third-country sales and invest in the U.S. for access to theconsumers in that market.

We next compare the operating profits attainable for a firm with the measure of productivity, θ, from Eqs. (4)–(7). Fig. 1 showsthe profits attainable for different levels of productivity, θ: for complete Home production and strategy HSS in Fig. 1a and for strategiesHNH and HNS in Fig. 1b. Fig. 1a depicts the profit functions reflected underlying Eqs. (4) and (5). The steeper πHSS relative to πHHHreflects lower unit production and transportation costs involved for firms which produce in the South and export part of theirproduction to the North. In this case, more productive firms are in the position to overcome the fixed investment costs, fS, in theSouth and take advantage of higher returns generated by the lower unit cost of production.17 In Fig. 1a, firms with productivitylevels lower than θ(HHH, HSS) prefer complete Home production and firms with productivity level higher than θ(HHH, HSS) conductFDI in South. This pattern is consistent with the Helpman et al. (2004) model that exporting firms are less productive than FDI firms.

If we compare Eqs. (6) and (7), πHNS has a steeper slope than πHNH since tNwS. Fig. 1b indicates that the most productive firmsengage in FDI in both North and South and firms with a productivity level lower than θ(HHH, HNS) undertake FDI only in North.

In order for the co-existence of all four location choices, we need to combine Fig. 1a and b. Fig. 1a combines the two strategiesreflected in Eqs. (4) and (5) while Fig. 1b combines the choices in Eqs. (6) and (7). In combining the two figures to identify theconditions necessary for all four strategies to uniquely co-exist would require further restrictions on the relative magnitudes oftransport costs, relative wages, fixed costs and market sizes.

Taking the strategies underlying Eqs. (5) and (7), it is clear that as long as tNwNwS, πHNS will be steeper than πHSS. That is, for the

(H, N, S) strategy to be more profitable than the (H,S,S) strategy, transport costs have to be larger than the wage differentialbetween the North and the South. If tbwN

wS, then πHNS will be lower than πHSS at every productivity level, θ, and (H, N, S) will never

be the optimal strategy for firms.18 In our data, the presence of Taiwanese multinationals in both China and the U.S. is thusconsistent with the assumption that tN wN

wS. Under this assumption, we readily have the complete ranking on the cost per unit:

wSb1bwNbwStb t. However, these assumptions are insufficient to determine the relative profitability reflected in Eqs. (5) and(6). For πHNH to be steeper than πHSS, the relative market size between South and North, bu bS

bN, must be less than a threshold

value, b̂uw1�rN � wStð Þ1�r

w1�rS �t1�r . In this case, the South market must be sufficiently small. Similarly, if the South market is sufficiently large,

βN β̂, then the slope of πHSS will be steeper than πHNH and the strategy (H, S, S) is more profitable than strategy (H, N, H) in theabsence of fixed investment costs.

In summary, the profitability of following a strategy implied by horizontal FDI plus export-platform in the South instead ofthat implied by northern market-access depends crucially on relative market size between the South and North. Wewill discusseach condition in turn.

1. Large Market Size in South (βN β̂): When the market in the South is large enough (βN β̂, β̂b1) compared to the North, Fig. 2outlines the conditions that must hold for all four strategies underlying Eqs. (4)–(7) to co-exist. In this case, it must be true thatπHNHbπHSS at the productivity level, θ(HSS, HNS), and πHNHNπHSS at the productivity level, θ(HHH, HSS). This implies that fSbδ1 fN

17 Note that in the absence of fS, the line represented by πHSS will go through the origin and it will always be more profitable to serve the South and Northmarkets with Southern, rather than Home, plants.18 Moreover, if tbwN (the wage differentials between the North and Home is greater than transportation costs), then the (H, H, H) strategy is more profitablethan the (H,N,H) strategy and there are only two possible strategies, (H, H, H) and (H, S, S). In this case, firms undertaking FDI in the South are more productivethan non-FDI firms. If wStbwNb t then the optimal outcome will depend on the scale of fS and fN For instance, if fSb fN, then πHNHbπHSS at all θ and only (H, H, H)and (H, S, S) will be the equilibrium outcome.

Fig. 2. Profit function, high trading cost and large south market.

409B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

and fSNδ2fN, where d1 ¼ w1�rS �t1�rð Þb

w1�rN � wStð Þ1�r and d2 ¼ wStð Þ1�r�t1�rþb w1�r

S �t1�rð Þw1�r

N �t1�r . Given that δ2N1, it implies that fSN fN.19 In this case, Fig. 2

shows that firms with the lowest productivity (below θ(HHH, HNH)) will concentrate on Home production while those with thehighest productivity (more than θ(HSS, HNS)) will undertake investments in both countries. Among firms with productivitylevels in the intermediate range, the relative size of the country-specific fixed costs determines the orderings of productivitylevels of South versus North investors. As long as fSNδ2fN firms investing in the North will be relatively less productive(between θ(HHH, HNH) and θ(HNH, HSS) than those investing in the South (between θ(HNH, HSS) and θ(HSS, HNS)).

2. Small Market Size in South (βbβ̂): In order for all four strategies in Eqs. (4)–(7) to co-exist it must be the case that πHNHNπHSS atthe productivity level, θ(HSS, HNS) and πHNHbπHSS at the productivity level θ(HHH, HSS). These two conditions imply, respectively,

that fSNδ1fN and fSbδ2fN, where d1 ¼ w1�rS �t1�rð Þb

w1�rN � wStð Þ1�r and d2 ¼ wStð Þ1�r�t1�rþb w1�r

S �t1�rð Þw1�r

N �t1�r . Since δ2b1, it must be the case that fSb fN.20

The above case of smallmarket size in the South is consistentwithGDP of $9.255 trillion in theU.S. and $4.8 trillion in China accordingto theWorld Factbook2000 (CIA).21Moreover, for Taiwanesemultinationals,fixed costs of investments in the South (China) are likely tobelower than those in the North (U.S.) for several reasons: China and Taiwan share the same language and culture, the close physicalproximity to each other and the larger scale of investments that Taiwanesefirms have in China.Moreover, special economic zones and taxincentives offered by the Chinese government to foreign investors lower the fixed costs of investing in China relative to the U.S.22 Fig. 3depicts the operating profits in the case where the market size in the South is smaller than the North. The least productive firms withproductivity levels less thanθ(HHH,HSS),willfind itmost profitable toproduce all goods atHomeandexport to both the South andNorth.In contrast, the most productive firms with productivity levels in excess of θ(HHH, HNS) will undertake FDI in both the North and South.These rankings mirror those for the large South market case. The difference here is among firms with productivity levels in theintermediate range. If fixed investment costs in the South are smaller than those in the North, firms investing in the South are lessproductive than those investing in the North. More specifically in Fig. 3, those investing in the South have productivity levels betweenθ(HHH, HSS) and θ(HSS, HNH) while firms investing in the North have higher levels of productivity, ranging between θ(HSS, HNH)andθ(HNH, HNS). In the figure, it is clear that raising the fixed cost of investing in the South market (downward shift in πHSS) will reducethe number of firms investing in the South, with themore productive ones in this group investing in the North and the remaining lowerproductivity firms producing domestically in order to export to the South.

19 If fSb fN, then strategy HNH will not be the optimal choice for firms. Large South market size favors strategy HSS, so firms prefer strategy HSS to HNH until thefixed investment cost in the South becomes high enough, fSNδ2fN. The high fixed investment cost in the South will force less productive firms switch fromstrategy HSS to HNH since HSS becomes less profitable than HNH for them now.20 If fSN fN, then strategy HSS will not be the optimal choice for firms. Small South market size favors strategy HNH, so firms prefer strategy HNH to HSS until thefixed investment cost in the North becomes high enough, fNN

fSd2. The high fixed investment cost in the North will force less productive firms switch from strategy

HNH to HSS since HNH becomes less profitable than HSS for them now.21 More specifically, with a small market size in the South, we will have the following ranking of fixed investment costs: fn+ fsN fnN fs which ensures that all fourtypes (HHH, HSS, HNH and HNS) coexist.22 In Yeaple (2006), distance, border, language, openness, corporate tax and social infrastructure are used as proxies for fixed costs. The maximum corporate taxin the U.S. is 35% (University of Michigan World Tax Database) in contrast to the 15% and 30% in the economic special zone and mid-west regions in China,respectively (Kao, 2005).

Fig. 3. Profit function, high trading cost and small south market.

410 B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

In summary, given the same fixed investment cost, more productive firms will choose FDI.23 With small wage differentialsand sufficiently high transportation costs, i.e., tNwN

wS, the most productive firms will invest in all countries, while the least

productive ones will stay in the home country.24 This result is independent of the market sizes and the scale of the fixed costsin the South and North. These factors are only important among firms with productivity levels in the intermediate range andinvesting in single countries. For these firms, the smaller fixed costs and market size of the South relative to the North suggestthat those that invest in the North will have higher productivity than those investing in the South. Thus, our theoreticalmodel predicts that Taiwanese multinationals investing in both the U.S. and China are most productive, followed inproductivity levels by those that invest only in the U.S., only in China and non-FDI firms. This clear sorting of FDI locationdecisions by firms' productivities provide the principal building blocks for the empirical specification presented in the nextsection.

4. Empirical specifications

The theoretical model presented in Section 3 predicts that the least productive firmswill produce at Home and export to foreignmarkets while themost productive firms undertake FDI in both the U.S. and China. The sorting of firms doing business in the U.S. orChina on the basis of their productivity depends on the scale of the fixed investment costs, fS and fN.

A firm's operating profit function for each strategy is specified as:

23 This24 The25 Thereject t

pit ¼ at þ btZi þ eit ð8Þ

where t=1, 2, 3, 4 and π1 represents firm i's profits for type HHH, π2 for type HSS, π3 for type HNH and π4 for type HNS. A firm'sprofit depends on the type-specific parameter, αt, and a vector of firm characteristics which determine the firm's productionefficiency, Zi. Each firm will choose the locations where its operating profit is maximized. Since we only observe each firm'slocation choices, rather than its profit on overseas operation, a multinomial logit (MNL) model25 is used to estimate the probabilitythat a firm belongs to one of the four strategy types, HHH, HSS, HNH and HNS:

Pti ¼

exp at þ btZi½ �P4k¼1 exp ak þ bkZi½ �

ð9Þ

is consistent with Helpman et al.'s (2004) model where FDI firms are more productive than exporting firms.se implications are consistent with Yeaple's (2007) result that the number of countries that firms have FDI increases with their productivity.multinomial logit model assumes that the ratio of the probabilities of any two alternatives is independent from the remaining choices (IIA). We cannot

he IIA assumption using a Hausman specification test.

26 Howempiric27 Kiyo28 Duntechnolinnovat29 Helinternat

411B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

t=1,2,3,4 and Pit is the probability that firm i belongs to type t. The firm characteristics included in Zi are labor productivity,

where

firm size, R&D intensity, foreign technology purchases. To identify parameters (α, β), we normalize the profit for type HHH to zeroand the estimated MNL model becomes:

P1i ¼ 1

1þP4k¼2 exp a~k þ b

~k Zi

� �

Pti ¼

exp a~t þb~t Zi

� �1þP4

k¼2 exp a~k þb~k Zi

� �ð10Þ

t=2, 3, 4 and α~k=(αk−α1) and β~k=(βk−β1).

whereThe constant term in each regression captures country-wide characteristics which are invariant across firms within the same

investment type. In our model, it represents the scale of the fixed investment costs for different investment types. We use laborproductivity rather than total factor productivity as the measure of a firm's productivity since information on capital flow or stockis not available in this data set.26 In order to distinguish the size effect from productivity, we include total employment as themeasure of size in the regression. Size is also an indicator for firms' capabilities to overcome investment barriers since larger firmshave more resources or access to more resources to invest abroad relative to their smaller counterparts. Since anecdotal evidencesuggests that firms first enter global markets by exporting and then becomemultinationals later, we include age as another controlvariable.27

While the theoretical model emphasizes firm productivity as a determinant of location choices by multinational firms, otherfactors have also been identified as important in the literature.28 In this paper, we use an input measure of technologicalknowledge, R&D and domestic technology purchasing intensity, as the common indicator of firms' capabilities.29 Firms'expenditures on their technology purchases abroad are also an important technology-sourcing strategy in LDCs. Firms in LDCstypically lack sufficient knowledge in the production of high-technology goods. In their constant quest to upgrade the quality oftheir products, firms in LDCs rely on technology imports from countries at the technology frontier. Basant and Fikkert (1996) findhigher returns in terms of productivity from technology purchasing than from firms' own R&D in India. Therefore, in addition tofirms' own investments on R&D, we also include foreign technology purchases as a proxy for firms' capabilities.

5. Empirical results

In Table 3 we report the results of estimating Eq. (10) using the MNL technique. The coefficients show the effect of each variableon the likelihood of a firm belonging to that group relative to the base group of non-FDI firms. Recall that the constant term in eachregression represents the scale of the fixed investment costs for different investment types. The negative constant terms point tothe higher costs of undertaking FDI relative to non-FDI, conditional on all firm characteristics, Zi. For both industries, the absolutevalues of the constant terms are highest for the group investing in both countries, intermediate for firms investing in the U.S., andsmallest for firms investing in China. This ordering implies that the fixed investment costs are higher in the U.S. than in China, aresult consistent with the observed proximity and similarities between Taiwan and China relative to the U.S. and the case of thesmall market size in the South in Fig. 3 of the theoretical model in Section 3.

The first three columns of Table 3 show the effect of each variable on the likelihood of a firm belonging to that group relative tothe base group of non-FDI firms. The difference in the coefficients of the first and second (third) columns reflects the effects of firmcharacteristics on the probability that a firm invests in the U.S. (both the U.S. and China) relative to China. Similarly, the differencein the coefficients in the second and third column reflects the effects of firm characteristics on the probability that a firm invests inboth the U.S. and China relative to only the U.S. These differences and their statistical significance are reported in the last threecolumns in Table 3. The positive and significant coefficients on the labor productivity and firm size variables in the first threecolumns in both industries are consistent with other empirical research showing that more productive and larger firms are in abetter position to overcome any disadvantage of operating abroad.

More importantly, the figures in the last three columns of Table 3 show that the ranking of the coefficients is consistent with ourtheoretical model which predicts how firms with differing productivities choose different types of FDI. In the Computer andTelecommunications industry, relative to firms that invest only in China, themost productive firms aremore likely to invest in both

ever, we have capital stock information on the firms that also appear in 1999. In the section of the paper where we perform some robustness checks of theal results, wewill rerun the regressions using this information on capital to construct a total factor productivity index to replace the labor productivity measure.ta and Urata (2007) document that Japanese firms are large exporters before becoming multinationals.ning (1981) presumes that a firm's FDI decision is based on organization, location and internalization advantages such that intangible assets, such asogy capabilities and reputation can outweigh the disadvantages of operating abroad. Criscuolo et al. (2005) find that globally engaged firms in the UKe more by hiring more researchers and acquiring knowledge through their international activities.pman et al. (2004), Yeaple (2007), Belderbos and Sleuwaegen (1996) and Brainard (1997) use R&D intensity to proxy for firm-specific intangible assets andionalization edge that firms use to compete with foreign rivals.

Table 3Multinomial logit regression of investment location choice, 2000

Independentvariables

China USA Both Difference in Coefficientsbetween USA and China

Difference in Coefficientsbetween Both and China

Difference in Coefficientsbetween Both and USA

Computer and telecommunications equipmentConstant −4.29 (0.51)a −6.41 (0.75)a −8.40 (1.20)a −2.12 (0.84)b −4.11 (1.24)a −1.99 (1.31)Labor Productivity 0.03 (0.16) 0.45 (0.22)b 0.60 (0.33)c 0.42 (0.25)c 0.57 (0.34)c 0.15 (0.37)Size 0.60 (0.11)a 0.85 (0.15)a 0.99 (0.22)a 0.25 (0.16) 0.39 (0.23)c 0.14 (0.24)R&D Intensity −1.62 (1.88) 3.31 (1.11)a 3.93 (1.03)b 4.93 (2.11)b 5.55 (2.42)b 0.62 (1.72)Foreign technology −7.83 26.09 29.86 33.92 37.69 3.77Purchase intensity (28.39) (15.19)c (18.73) (29.73) (31.56) (19.11)Age 0.08 (0.13) −0.08 (0.17) 0.09 (0.26) −0.16 (0.20) 0.01 (0.28) 0.17 (0.30)Observations 73 40 16Sample 602Likelihood −383.75

Parts and componentsConstant −3.20 (0.36)a −7.03 (0.79)a −9.55 (1.78)a −3.83 (0.84)a −6.35 (1.80)a −2.52 (1.91)Labor productivity 0.26 (0.12)b 0.90 (0.20)a 0.79 (0.42)c 0.64 (0.22)a 0.53 (0.43) −0.11 (0.45)Size 0.27 (0.07)a 0.72 (0.13)a 0.83 (0.28)a 0.45 (0.14)a 0.56 (0.28)b 0.11 (0.30)R&D Intensity 0.95 (1.70) 3.95 (2.35)c 6.25 (3.14)b 3.00 (2.73) 5.30 (3.40) 2.30 (3.69)Foreign technology −88.66 6.94 −21.59 95.60 67.07 −28.53Purchase intensity (46.45)c (2.87)b (39.20) (46.51)b (60.07) (39.23)Age 0.22 (0.09)b −0.12 (0.19) 0.27 (0.36) −0.34 (0.20)c 0.05 (0.36) 0.39 (0.39)Observations 130 34 7Sample 884Likelihood −493.60

Notes: Standard errors of the difference are in parentheses. ‘a’ represents significance at the 0.01 level, ‘b’ for significance at the 0.05 level, ‘c’ for significance at the0.1 level. R&D Intensity is defined as expenditures on R&D and Domestic Technology Purchases as a fraction of total firm revenue.

412 B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

U.S. and China (.57) followed by firms that invest only in the U.S. (.42).30 An identical ranking of the investment categories areobtained for the variable representing firm size. This pattern is consistent with the key prediction of our theoretical model thatinvestors in multiple destinations have the largest productivity and size premium. In the Parts and Components industry, relativeto firms that invest only in China, the coefficient on labor productivity is highest for firms investing only in the U.S. (.64) and notstatistically significant for those investing in both countries. This result appears to be inconsistent with the theoretical predictionthat the highest coefficient should be for firms investing in both the U.S. and China. One explanation is that in the Parts andComponents industry, there are only a handful (3%) of Taiwanese multinational firms investing in both China and U.S. The largercoefficient of the group investing only in the U.S relative to the group investing only in Chinamay be related to the difference in theproduct mix of firms investing in the two countries within the Parts and Components industry.31 Firms investing only in the U.S.focus on semi-conductors (57%) while more than 70% of those investing only in China specialize in the production of passivecomponent and power supplies. Fig. 4 presents a box plot of productivity for products within the electronic Parts and Componentsindustry. The full range of the box represents the inter-quartile range of labor productivity and the vertical linewithin the box is themedian value of labor productivity for each product category. Overall, firms producing semi-conductors are more productive thanthose producing passive component and power supplies. Thus, the stronger contribution of labor productivity to the likelihood offirms investing in the U.S. may be explained by the heterogeneity in themix of products within the two industries. The positive andsignificant coefficients on the firm size variable are similar to those in the computer industry and confirm the standard result thatlarger firm size increases the likelihood of a firm engaging in FDI, with investors in multiple destinations having the largest sizepremium, followed by those investing only in the U.S. and only in China.

The coefficients on R&D intensity in the first three columns of the table are positive and significant for the group investing in theU.S. as well as firms investing in both China and the U.S. and statistically insignificant in both industries for the group investing onlyin China. The last column shows that there is no statistical difference in the R&D intensity of firms investing only in the U.S. or boththe U.S. and China. Columns four and five indicate that relative to investors in China, firms investing in both the U.S. and China aremost R&D-intensive followed by those that invest only in the U.S. These coefficients are only statistically significant for theComputer and Telecommunications industry. Together, these results suggest that firms in this industry that invest in the U.S.,unlike their counterparts that invest in China, may face stiff competition with high quality products in the local U.S. market andhave greater incentive to accumulate intangible assets created by R&D. In contrast, the coefficients for R&D intensity in the lastthree columns of the table are not statistically significant in the Parts and Components industry where, instead, the coefficient forforeign technology purchasing intensity for firms investing in the U.S. relative to investors in China is positive and statisticallysignificant. These patterns are not surprising for two reasons. First, the Computer and Telecommunication industry is much moresophisticated technologically than the Parts and Components industry. Second, within both industries, firms that invest in the U.S.,

30 By taking the exponent of the coefficient on labor productivity, we obtain the odds ratio for the various location choices. In particular, a one standarddeviation increase in the labor productivity raises the odds of investing in the U.S. instead of China by 52%, i.e., [exp(0.42) -1]×100%, and raises the odds ofinvesting in both China and the U.S. instead of investing only in China by 76%.31 Bernard et al. (2006) emphasize the different production techniques of products even within a narrowly defined industry such that firms differing inproductivity will endogenously self-select into various products.

Fig. 4. Technology intensity of product.

413B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

in contrast to their counterparts that invest in China, tend to concentrate on higher quality and more technologically advancedproducts and have greater incentives to seek new technology through either R&D activities or purchasing technology fromabroad.32 Finally, the age variable is only negative and significant in the Parts and Components industry for firms investing in the U.S. relative to investors in China.

In Table 4 we standardized the coefficients in the regressions in order to assess the importance of firm productivity relative tothe other factors in affecting a firm's choice of location. To do this, we take the deviation of the value of each independent variablefrom the sample mean and divide this by the sample standard deviation. Firm size is the most important factor for Taiwanesemultinationals in both industries. In the Parts and Components industry, labor productivity is a close second followed by R&Dintensity. In the Computer and Telecommunications Equipment industry, the magnitudes between labor productivity and R&Dintensity are comparable, probably reflecting the industry's greater technological sophistication.

We performed a couple of robustness checks in the estimated regressions. Ideally, the measure of productivity should be thefirm's total factor productivity to account for the effect of capital intensity in production. Capital stock data, however, was notcollected in 2000, the year that outward FDI became available. To examine the sensitivity of our results to the inclusion of capital,we merged information on firm-level capital stock in the 1999 survey data into the 2000 survey data, and constructed total factorproductivity index (TFP) at the firm-level.33We then reran the regression replacing labor productivity and size (employment) withTFP and capital stock, respectively. The results, presented in Table 5, are consistent with those presented in the original regressions.In fact, in the Parts and Components industry, once we accounted for the role of capital, the ranking of location choices is nowconsistent with the theoretical prediction that the highest coefficient should be for firms investing in both the U.S. and China.Finally, to examine the sensitivity of our results to including a broader set of countries in our model, we reran the regression withthe South region comprising of China and other Southeast Asian countries and the North comprising of the U.S., Japan and theEuropean Union. The results are essentially unchanged from those reported in Table 3.

Overall, our results provide empirical support for the theoretical predictions relating the location choices of firms to theirproductivity and country and industry characteristics. The least productive firms do not invest abroad while the most productiveinvest in multiple countries (China and the U.S.). This pattern is consistent with the theoretical model in which fixed costs play akey role in the firm's choice of production location. In locations where fixed costs are high (the U.S.), only themore productive firmschoose to invest, leaving less productive firms to invest in countries where fixed costs are significantly lower (China).34,35

6. Summary and conclusion

In this paper, we examine the link between firm-level heterogeneity and FDI location choices among Taiwanese multinationalsin the electronics subsector. The traditional theory underlying horizontal and vertical FDI is inadequate in explaining the more

32 Fosfuri and Motta (1999) suggest that foreign FDI can facilitate knowledge acquisition through intra-national technological spillovers. Their model capturesthe technology acquisition rational for the flood of foreign direct investment in Silicon Valley during the 1990s.33 TFP indexes are constructed as multilateral Tornqvist index developed by Caves et al. (1982) and modified by Aw et al. (2000).34 Belderbos and Sleuwaegen (1996) obtain a similar result with firm-specific advantages playing a less important role in Japanese firms investing in SoutheastAsian countries relative to those investing in Western industrial countries due to the relative absence of strong local competing firms in Southeast Asian countries.However, their paper ignores the relationship between the geographic structure of production and firm-specific capability, a point emphasized by Yeaple (2007).35 Another possibility is that FDI firms may only appear to have higher productivity because they have transferred their less productive activities abroad. Giventhe cross-sectional nature of our data set and without data on affiliates, we are unable to make any causality statements between productivity and firm decisionsto invest abroad. However, migration of low productivity activities abroad is likely to occur only to a low-wage country such as China rather than the U.S. Thatwould lead us to observe firms with FDI in China having higher productivity than firms with FDI in the U.S., which is not what we observe in the data.

Table 4Multinomial logit regression of investment location choice, 2000 (with Standardized Independent Variables)

Independentvariables

Computer and telecommunications Parts and components

China USA Both China USA Both

Constant −2.01 (0.14)a −3.09 (0.23)a −4.32 (0.41)a −1.94 (0.16)a −3.92 (0.30)a −5.49 (0.65)a

Labor productivity 0.02 (0.14) 0.39 (0.19)b 0.52 (0.29)c 0.22 (0.10)b 0.77 (0.17)a 0.68 (0.36)c

Size 0.78 (0.14)a 1.10 (0.19)a 1.28 (0.28)a 0.38 (0.10)a 0.99 (0.18)a 1.16 (0.39)a

R&D intensity −0.18 (0.21) 0.37 (0.13)a 0.44 (0.18)b 0.05 (0.10) 0.23 (0.14)c 0.37 (0.18)b

Foreign technology −0.06 0.19 0.21 −2.38 0.19 −0.58Purchase intensity (0.20) (0.11)c (0.14) (1.25)c (0.08)b (1.05)Age 0.08 (0.13) −0.08 (0.18) 0.09 (0.26) 0.23 (0.10)b −0.13 (0.20) 0.28 (0.38)

Notes: Standard errors of the difference are in parentheses. ‘a’ represents significance at the 0.01 level, ‘b’ for significance at the 0.05 level, and ‘c’ for significance atthe 0.1 level. Each independent variable is normalized to its mean level and then divided by its sample standard deviation.

414 B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

complex pattern of FDI observed in reality. Using a three-country model with different cost specifications, we analyze the variedmotivations underlying the decisions of investors from a middle-income country such as Taiwan to serve consumers at home, inthe U.S. and China.

In our theoretical model, firms serve each market by either producing in its foreign subsidiary or through exports. Firmsengaging in FDI incur fixed investment costs of establishing a plant in the foreign market, while exporting firms incur costs intransporting goods to the foreign markets. Given this framework, our theory predicts that the least productive firms tend to servethe foreign markets through exporting from domestic plants. The more efficient firms, on the other hand, will incur the fixed costsassociated with setting up a foreign subsidiary. As long as fixed investment costs in the South are less than in the North, Taiwanesemultinational firms investing in the North are predicted to have higher productivity than firms investing in the South. Amultinomial logit model is used to test the predictions of the theoretical model.

A broad conclusion from our estimation of the empirical model shows that the choice of production site reflects the underlyingpattern of firm productivity as predicted in our theoretical model. More productive firms are more likely to undertake overseasinvestments. The ranking of productivity levels among FDI firms depends on the scale of the fixed investment costs involved. Ourresults indicate that the fixed investment costs for Taiwanese multinationals are higher in the U.S. than in China. After controllingfor firm size, age, R&D intensity and other proxies for technological acquisition, the least productive firms produce at home andexport to foreign markets while the most productive firms set up foreign subsidiaries in both China and the U.S. The empiricalresults also highlight the importance of fixed costs in establishing a foreign subsidiary and the trade-off involved in policies thatchange the size of these fixed costs. Where these costs are higher (in the U.S.), Taiwanese multinationals producing in the U.S. arefound to have higher productivity levels than those operating in China where fixed costs are smaller.

Table 5Multinomial logit regression of investment location choice and industry with 1999 capital data

Independentvariables

China U.S. Both Difference in Coefficientsbetween USA and China

Difference in Coefficientsbetween Both and China

Difference in Coefficientsbetween Both and USA

Computer and telecommunications equipmentConstant −6.54 (0.96)a −8.09 (1.34)a −11.03 (2.06)a −1.55 (1.52) −4.49 (2.16)b −2.94 (2.27)TFP 0.60 (0.41) 1.48 (0.49)a 1.49 (0.65)b 0.88 (0.51)c 0.89 (0.69) 0.01 (0.71)ln(Capital) 0.44 (0.08)a 0.51 (0.11)a 0.68 (0.16)a 0.07 (0.12) 0.24 (0.17) 0.17 (0.18)R&D Intensity −3.24 (2.95) 3.94 (1.47)a 4.17 (1.93)b 7.18 (3.17)b 7.41 (3.41)b 0.23 (2.03)Foreign Technology −8.07 13.95 18.45 22.02 26.52 4.50Purchase Intensity (28.81) (17.06) (21.52) (30.77) (33.04) (21.37)Age 0.10 (0.14) −0.14 (0.20) −0.07 (0.28) −0.24 (0.23) −0.17 (0.30) 0.07 (0.32)Observations 61 32 15Sample 496Likelihood −324.89

Parts and componentsConstant −4.40 (0.67)a −10.98 (1.48)a −15.32 (3.44)a −6.58 (1.58)a −10.92 (3.46)a −4.34 (3.67)TFP 1.48 (0.34)a 1.68 (0.66)b 2.99 (1.01)a 0.20 (0.71)a 1.51 (1.01) 1.31 (1.14)ln(Capital) 0.24 (0.06)a 0.66 (0.11)a 0.83 (0.24)a 0.42 (0.12)b 0.59 (0.24)a 0.17 (0.25)R&D intensity 1.54 (1.95) 2.38 (3.16) 7.48 (3.32)b 0.84 (3.56) 5.94 (3.67)c 5.10 (4.35)Foreign technology −101.91 14.13 −26.54 116.04 75.37 −40.67Purchase Intensity (49.48)b (3.26)a (48.08) (49.53)a (67.45) (48.11)Age 0.16 (0.10) −0.12 (0.22) 0.35 (0.43) −0.28 (0.24) 0.19 (0.43) 0.47 (0.47)Observations 120 26 6Sample 767Likelihood −421.73

Notes: Standard errors of the difference are in parentheses. ‘a’ represents significance at the 0.01 level, ‘b’ for significance at the 0.05 level, ‘c’ for significance at the0.1 level. R&D Intensity is defined as expenditures on R&D and Domestic Technology Purchases as a fraction of total firm revenue.

415B.Y. Aw, Y. Lee / Journal of International Economics 76 (2008) 403–415

Finally, our results suggest that there is a clear trade-off in the promotion of exports versus outward FDI. For instance,policies that are successful in raising the fixed costs of investing in China (and/or the U.S.) may increase exports and the jobsassociated with these exports but only at the expense of more profitable strategies of setting up subsidiaries in China or the U.S.

Acknowledgements

We thank Mark J. Roberts and James Tybout as well as two anonymous referees and the co-editor (Bruce Blonigen) for veryhelpful comments.

Appendix A. Definition of explanatory variables

Explanatory Variables

Definition

Labor Productivity

LP ¼ lnQ � PlnQ

� �� lnL�PlnL

� �� �where

PlnQ and

PlnL are the industry mean levels of the log of total revenue

plus net inventory change and log of total employment.

Size Measured in terms of ln(total employment) Age ln(age) where age is defined as the number of years since the firm began business through the year 2000. R&D Intensity (R&D Expenditure+Domestic Technology Purchases) /Total Revenue, where R&D Expenditure is the sum of the

salaries of R&D personnel (researchers and scientists), material purchases for R&D, and R&D capital (equipmentsand buildings) expenses. Domestic Technology Purchases refer to the total payment made in the acquisition oftechnology (e.g. patent royalty) from other domestic corporations.

Foreign Technology Purchases Intensity

(Foreign Technology Purchases) /Total revenue, where Foreign Technology Purchases refer to expenses incurred bythe firm in purchasing technology (e.g. patent royalty) from foreign firms.

Source: Survey of Manufacturing Plant (MOEA), 2000.

References

Aw, Bee-Yan, Chung, Sukkyn, Roberts, Mark, 2000. Productivity and turnover in the export market: micro-level evidence from Republic of Korea and Taiwan. TheWorld Bank Economic Review 14, 65–90.

Baldwin, Richard, Ottaviano, Gianmarco, 2001. Multiproduct multinationals and reciprocal FDI dumping. Journal of International Economics 54, 429–448.Basant, Rakesh, Fikkert, Brian, 1996. The effect of R&D, foreign technology purchase, and domestic and international spillover on productivity in Indian firms. The

Review of Economics and Statistics 78 (2), 187–199.Belderbos, Rene, Sleuwaegen, Leo, 1996. Japanese firms and the decision to invest abroad: business groups and regional core networks. The Review of Economics

and Statistics 78 (2), 214–220.Bernard, Andrew, Redding, Stephen, Schott, Peter, 2006. Products and productivity. CEPR Discussion Papers 5125.Brainard, Lael, 1997. An empirical assessment of the proximity-concentration trade-off between multinational sales and trade. American Economic Review 87 (4),

520–544.Caves, Douglas, Christensen, Laurits, Diewert, Erwin, 1982. Output, Input, and Productivity Using Superlative Index Numbers. Economic Journal 92, 73–96.Criscuolo, Chiara, Haskel, Jonathan, Slaughter, Matthew, 2005. Global engagement and the innovation activities of firms. NBER Working Paper 11479.Debaere, Peter, 2004. Reversing the perspective: expansion activities of multinational corporations from middle-income countries. CEPR Discussion Papers 4435.Dunning, J.H., 1981. International Production and the Multinational Enterprises. Allen & Unwin, Lodon.Ekholm, Karolina, Forslid, Rikard, Markusen, James, 2007. Export-platform foreign direct investment. Journal of the European Economic Association 5 (4), 776–795.Fosfuri, Andrea, Motta, Massimo, 1999. Multinational without advantages. Scandinavian Journal of Economics 101 (4), 617–630.Girma, Sourafel, Kneller, Richard, Pisu, Mauro, 2005. Exports versus FDI: an empirical test. Review of World Economics 141 (2), 193–218.Grossman, Gene, Helpman, Elhanan, Szeidl, Adam, 2006. Optimal integration strategies for the multinational firm. Journal of International Economics 70 (1),

216–238.Head, Keith, Ries, John, 2003. Heterogeneity and the foreign direct investment versus exports decision of Japanese manufacturers. Journal of the Japanese and

International Economies 17 (4), 448–467.Helpman, Elhanan, Melitz, Marc, Yeaple, Stephen, 2004. Export versus FDI with heterogeneous firms. American Economic Review 94 (1), 300–316.IMD, 2000. World Competitiveness Yearbook. Institute for Management Development, Lausanne, Switzerland.Kiyota, Kozo, Urata, Shujiro, 2007. The role of multinational firms in international trade: the case of Japan. Research Seminar in International Economics. .

Discussion Paper, vol. 560. University of Michigan.Kao, Charng, 2005. The evolution of Chinese FDI policy. Trade Reform in China and Trade Relationship between China and Taiwan (in Chinese) (Wu-nan Culture

Enterprise).MOEA, 2001. Outward FDI Survey in Manufacturing. (Ministry of Economic Affairs).MOEA, 2004. Outward FDI Survey in Manufacturing. (Ministry of Economic Affairs).UNCTAD, 2002. World Investment Report: Transnational Corporations and Export Competitiveness. United Nations Conference on Trade and Development, New

York. and Geneva.University of Michigan (Ross School of Business): World Tax Database.(Accessed Oct., 2007) http://www.bus.umich.edu/otpr/otpr/default.asp.Yeaple, Stephen, 2006. How do Multinationals Sort? University of Colorado, Boulder. manuscript.Yeaple, Stephen, 2007. Firm Heterogeneity and the Structure of U.S. Multinational Enterprise: An Empirical Analysis. University of Colorado, Boulder. manuscript.