offshore outsourcing decision and capital intensity - Wiley Online Library

OFFSHORE OUTSOURCING DECISION AND CAPITAL INTENSITY: FIRM-LEVEL RELATIONSHIPS EIICHI TOMIURA, BANRI ITO and RYUHEI WAKASUGI∗

In offshore sourcing, a firm chooses outsourcing to independent suppliers or insourcing from own foreign direct investment (FDI) subsidiaries. Based on the firmlevel data on offshore make-or-buy decision covering all manufacturing industries, this paper compares averages, documents inter-firm distributions, and estimates multinomial logit models of the firm’s sourcing mode choice. As predicted by previous theoretical models, this paper directly confirms at the firm level that outsourcing firms tend to be substantially labor-intensive compared with firms in-sourcing from the same region, even after the firm’s R&D intensity, firm size, or industry is controlled for. (JEL F23, L23, L24, L14) I.

by which we can distinguish outsourcing versus intra-firm sourcing, and offshoring destinations, across all manufacturing industries. As a seminal paper on the offshore makeor-buy decision, Antràs (2003) shows that offshore outsourcing firms are more labor-intensive than firms engaged in intra-firm offshoring. This empirically testable hypothesis is derived from an incomplete contract model combined with the presumption that investment cost sharing is likely to be more difficult in labor than in physical capital. As implied by Grossman and Hart (1986), it is important to give incentives to the party who make critical investment. Then, in producing labor-intensive goods, firms choose outsourcing for providing incentives to independent contractors, who invest in labor. On the other hand, firms choose vertical integration in producing capital-intensive goods by sharing costs for investment in capital. Consequently, the capital–labor ratio K/L is the key determinant of the outsourcingversus-integration choice not only for sourcing within the same country but also for sourcing across borders. The model by Antràs (2003) is a splendid application of the Grossman–Hart firm-level make-or-buy decision model to the

INTRODUCTION

The world economy in recent years is strongly characterized by new modes of cross-border activities, such as complex integration strategies by multinational corporations and offshoring of wide varieties of tasks, just to name a few. Motivated by these changes in the real world, new theoretical models of international trade have recently formalized how firms organize their production on a global scale.1 Empirical evidence for these new theories, however, has been so far limited because of the constrained data availability. This paper fills a part of this gap by using the unique firm-level offshoring data, *The authors deeply appreciate insightful comments from the two anonymous referees and the Editor. This article is a substantially revised version of the paper presented at the University of Nottingham and Research Institute of Economy, Trade and Industry (RIETI). Valuable comments from the conference participants are acknowledged. This research was partly financed by RIETI and Grant-in-Aid for Scientific Research. The authors also appreciate the administrative support by RIETI, especially Akira Kawamoto, Kazumi Wakai and Database Group staffs. Remaining errors are those of the authors. The opinions expressed in this paper do not reflect those of RIETI or METI. Tomiura: Professor, Department of Economics, Yokohama National University and RIETI, 79-4 Tokiwa-dai, Hodogaya-ku, Yokohama City 240-8501, Japan. Phone 81-(0)45-339-3563, Fax 81-(0)45-339-3574, E-mail [email protected] Ito: Assistant Professor, Department of Economics, Senshu University and RIETI, 2-1-1 Higashimita, Tama-ku, Kanagawa, Japan. Phone 81-(0)44-900-7988, Fax 81(0)44-900-7849, E-mail [email protected] Wakasugi : Professor, Institute of Economic Research, Kyoto University and RIETI, Yoshida-Honmachi, Sakyo-ku, Kyoto, Japan. Phone 81-(0)75-753-7135, Fax 81-(0)75753-7138, E-mail [email protected]

ABBREVIATIONS ASEAN: Association of South-East Asian Nations CDF: Cumulative Distribution Function FDI: Foreign Direct Investment IIA: Independence of Irrelevant Alternatives RIETI: Research Institute of Economy, Trade and Industry ROW: Rest of the World

1. See Helpman (2006) and Antràs and Rossi-Hansberg (2008) for a survey. 364 Economic Inquiry (ISSN 0095-2583) Vol. 49, No. 2, April 2011, 364–378

doi:10.1111/j.1465-7295.2010.00316.x Online Early publication July 16, 2010 © 2010 Western Economic Association International

TOMIURA, ITO & WAKASUGI: OFFSHORE OUTSOURCING

offshoring context, although the latter part of his paper further combines it with Helpman–Krugman international trade model to analyze sector-level aggregated intra-firm import shares. Although many theoretical firm heterogeneity models of international trade focus on productivity, the investigation of the relationships with K/L is critical for understanding offshoring in the real world performed by firms with remarkably different factor intensities, and for testing the empirical relevance of the Grossman–Hart mechanism in offshore outsourcingversus-integration choice. Several papers have investigated this issue based on various data sources. For example, Antràs (2003) provides evidence favorable to his theoretical prediction from aggregated data on 23 U.S. sectors. Yeaple (2006) revisits it by using data on 51 U.S. industries with additional sector-level control variables (e.g., R&D intensity). Nunn and Trefler (2008) bring U.S. intrafirm trade share data at detailed product level, but their K/L remains at aggregated sectorlevel (370 industries identified).2 As the original theoretical prediction is derived from the recently developed literature on firm heterogeneity in international trade, within-industry variations across firms should be squarely examined. This paper empirically compares offshore outsourcing firms with firms sourcing from their foreign direct investment (FDI) subsidiaries (insourcing), by exploiting the unique firm-level offshoring data. This paper is differentiated from the previous work as follows. First, offshore outsourcing firms and insourcing firms are directly identified at the firm level, instead of relying on intra-firm trade shares aggregated over heterogeneous firms. This enables us to investigate not only the K/L gap between average outsourcing firms and average in-sourcing firms but also the interfirm K/L distribution among firms choosing the same sourcing mode. This paper also estimates the multinomial choice model of the individual firm’s sourcing mode by controlling for relevant firm-specific variables, such as the firm’s R&D intensity and firm size. Our use of firm-level data will certainly enrich previous results from intrafirm trade shares (e.g., Nunn and Trefler 2008), as the data on firm characteristics are inevitably unavailable in trade statistics and have been previously replaced by sector-level averages. 2. Similarly, Bernard et al. (2008) use detailed productlevel intra-firm trade data, but their K/L is also at four-digit industry level.

365

Although standard theoretical firm heterogeneity models of trade, such as Antràs and Helpman (2004), consider productivity heterogeneous at the firm level but assume K/L determined by industry factors, K/L actually differs substantially across firms even within the same industry.3 For example, firms at low K/L quartile in iron and steel (the most capital-intensive industry) are as labor-intensive as average firms in wooden products or textiles. Firms at high quartile in apparel (the most labor-intensive industry) are nearly as capital-intensive as firms at low quartiles in general machinery. K/L varies more across firms within each industry, compared with cross-industry variations of industries’ average K/L.4 As K/L is the key variable in the firm-level Grossman–Hart type theoretical model by Antràs (2003), and as we cannot ignore thus enormous within-industry K/L variations, we exploit firm-level data on K/L in our empirical analysis of offshore outsourcing versus in-sourcing choice. Although it is less extensive than in previous aggregated data, the coverage of our data set remains substantial. Virtually all large- or medium-sized Japanese firms in all manufacturing industries are surveyed. Our results should be regarded as a reliable representation of the whole manufacturing because very few smallsized firms are involved in offshoring. Major destinations for Japanese offshoring firms are also identified in our firm-level data. As firms expand their offshoring locations around the globe, cross-regional variations in the enforcement of contracts and the quality of legal system should become increasingly critical. By comparing outsourcing firms versus in-sourcing firms among firms offshoring to the same destination, this paper can isolate the effects of capital intensity on the offshore outsourcingversus-integration decision from the effects of cross-regional variations in institutional factors.5 The rest of this paper is organized as follows. 3. See Appendix Table A1. The definition and coverage of our firm-level data will be explained in the next section. The two-digit classification is the most detailed level available for us consistently over classification revisions, as we will later use data over more than 5 years. 4. MacKay and Phillips (2005) and Spaliara (2009) have also confirmed that firm-specific factors, rather than industry characteristics, are important for explaining inter-firm K/L variations. 5. Nunn and Trefler (2008) and Bernard et al. (2008) compare intra-firm import shares across destinations based on product-level trade data. Their approach and ours should be viewed complementary for the same research purpose, given the data limitation in each case.

366

ECONOMIC INQUIRY

Section II describes the data, and summarizes basic statistics. Section III reports firm-level empirical results on the relation between offshore make-or-buy choice and capital intensity. Section IV adds concluding comments. II.

DATA DESCRIPTION

A. Description of the Survey This paper derives firm-level data from a unique survey linked with official statistics. The questionnaire of the survey was sent to 14,062 firms in Japan, of which 39% returned their answer sheets.6 The population of firms for this survey is chosen as the same as those used for the previous wave of the annual national legal mandatory survey, covering all firms with 50 or more employees and capitalized at not less than 30 million yen. Consequently, these firms are virtually all large- or medium-sized firms in all manufacturing industries, and thus should be regarded as reasonably reliable in deriving implications to the whole manufacturing.7 Because many, although not all, previously available firm-level data sets on offshore outsourcing include only a limited number of firms and are not designed to cover the entire manufacturing, this survey has an advantage in its coverage.8 Some previous studies have actually used micro data with comprehensive coverage, but used different definitions of offshoring, as explained below. As this survey was conducted only once, all the offshoring data reported in this paper are at the year 2006. The “offshoring,” or offshore sourcing, is defined by contracting-out to other firms9 located overseas10 based on explicit contracts specifying specifications or other dimensions of the offshored tasks. Offshoring of both 6. The survey was conducted by Japan’s Research Institute of Economy, Trade and Industry (RIETI) for our research project. For details of the survey, see Ito, Tomiura, and Wakasugi (2007). 7. The questionnaire was sent to all firms surveyed by The Basic Survey of Business Structure and Activities (Kigyo Katsudo Kihon Chosa, in Japanese). 8. For example, Görg and Hanley (2005) cover 652 plants within Irish electronics industries. Swenson (2000) focuses on firms located in U.S. foreign trade zones. Intrafirm trade share data, which have been used by Nunn and Trefler (2008) and Bernard et al. (2008), are detailed at both in product- and destination-dimensions, but contain no data on firm-characteristics. 9. Imports from branch offices/factories are not included as they are not independent legal entities. 10. Offshoring in this paper includes both arm’s-length sourcing from independent suppliers and intra-firm sourcing from subsidiaries, as long as transactions are cross-border.

production and service tasks are covered. When a firm purchases standardized goods/services readily available at marketplace, such a case is not counted as offshoring in this survey. This exclusive definition is different from those adopted in previous studies such as Head and Ries (2002), but appropriate for investigating the empirical implications of make-or-buy decision models in the theory of the firm, which emphasizes relation-specific investment.11 Although no quantitative data are available on how much each firm is offshoring, the survey distinguishes outsourcing versus insourcing. “In-sourcing” is defined by offshoring to own offshore majority-owned subsidiaries, while “outsourcing” is defined as offshoring to all other independent legal entities, including local firms or subsidiaries owned by other multinationals. Although 10% threshold is often used in FDI studies, the majority ownership is practically central and conceptually critical in discussing a controlling stake.12 The comparison of capital intensity between offshore outsourcing firms and in-sourcing firms is the main target of this paper. The survey also has disaggregated information on the destination of offshoring. This paper identifies the following four regions: (a) China, (b) Association of South-East Asian Nations (ASEAN), (c) North (United States and European countries), and (d) ROW (the rest of the world).13 By using this distinction of geographical locations, we can at least partly adjust cross-regional differences in contracting environment. As will be reported later, China 11. For example, López (2006) covers all plants with more than ten employees in Colombia but examines all intermediate imports. Görg, Hanley, and Strobl (2008) cover substantial share of plants with not less than 20 employees in Irish manufacturing but examine imports of raw materials, components and service inputs. Hijzen, Inui, and Todo (2008) use, the same Japanese official statistics, which we link to our offshoring survey, but concentrate on production subcontracting in the early 1990s. Head and Ries (2002) examine overseas employment of Japanese publicly traded companies. 12. As a support for the focus on majority-owned FDI, Desai, Foley, and Hines (2002) find that imports from minority-owned affiliates are much smaller than those from majority/wholly owned subsidiaries in the United States. Nunn and Trefler (2008) also confirm that “for a very large proportion of ownership positions in the BEA data, once the position is more than 10%, it is also more than 50%” (p. 21). 13. Hong Kong and Taiwan are included in China. ASEAN is composed of ten countries. India, Middle East, Latin America, and Australia are included in ROW, while Eastern Europe or Canada is categorized as a part of “North.”


and ASEAN are the major offshoring destinations for Japanese firms. Institutional variations in contracting environments are likely to be negligible among countries within North, while those among ASEAN countries are also supposed to become slimmer as they are coordinating various institutions in the process of regional integration. Consequently, while we cannot identify legal differences across individual countries in our data, the comparison of outsourcing firms versus in-sourcing firms among firms offshoring to the same destination region will contribute to isolating the effect of capital intensity from that of contracting environment. B. Summary Statistics This section summarizes the allocation of firms across different sourcing modes based on the firm-level data. Before investigating the contrast between offshore outsourcing and in-sourcing, it is useful to know how few firms are active in offshoring. According to our survey as reported in Ito, Tomiura, and Wakasugi (2007), merely 21% of the surveyed firms are offshoring. This participation rate may sound extremely low in a sample covering both production and service offshoring among large- or medium-sized firms in a developed country, but is in line with previous findings from a different source.14 One possible interpretation could be non-negligible entry costs for offshore sourcing, as formalized by theoretical models (e.g., Antràs and Helpman 2004). Table 1 displays the percentages of outsourcing firms and in-sourcing firms based on the binary Yes-or-No question. The percentage in this table is among offshoring firms. The sum TABLE 1 Percentages among Offshoring Firms China ASEAN North ROW

Outsourcing

In-sourcing

54.25 21.96 13.89 17.22

39.40 18.73 9.58 6.24

Note: Shown is the percentage of firms among all offshoring firms. 14. Tomiura (2007) reports that only 3% of the firms are outsourcing production offshore at 1998 among 118,300 surveyed firms, of which nearly 80% are with less than 50 workers.

367

of the eight cells naturally exceeds 100 as some offshoring firms are procuring from multiple sources. Several points must be noted in Table 1. First, more than half of the firms are involved in outsourcing to China. The prevalence of outsourcing to China is as expected, given the Japan–China differences in production costs and their geographical proximity, and is also in line with a previous report on Chinese trade by Feenstra and Hanson (2005).15 Second, China is the most attractive destination for Japanese offshoring, followed by ASEAN, which is in turn followed by North. This ordering should at least be partly influenced by inter-regional differences in production costs. Third, in every region, more firms are involved in outsourcing compared with in-sourcing. To separate overlaps because of the firms engaged in both sourcing modes, Table 2 disaggregates offshoring firms into the following three mutually exclusive categories: (a) The firms engaged both in in-sourcing and outsourcing (In & Out), (b) The firms engaged in in-sourcing but not outsourcing (In-only), and (c) The firms engaged in outsourcing but not in-sourcing (Out-only).16 Out-only is the most frequent choice in every region, followed by Inonly, while In & Out is the least popular choice. More than 80% of the firms choose in-sourcing or outsourcing but not both irrespective of the destination. This result is in line with a previous report from a different source.17 Once again, we confirm that the share of outsourcing firms is higher than that of in-sourcing firms in any region, even if In & Out firms are excluded.18 Rather than investigating peculiarities of individual regions, this paper instead focuses on the 15. Feenstra and Hanson (2005) find that the share of “Chinese-owned factories” is dominant in China’s pureassembly processing exports. 16. Some may be in-sourcing from one region simultaneously with outsourcing in another region. 17. Feenstra and Hanson (2005) find that more than 80% of Chinese processing exports choose only one mode in their four ownership/control regimes (foreign vs. Chinese in the factory ownership and in the control of input purchases). 18. Related with this finding of cross-regional invariance of outsourcing share among offshoring, Antràs and Helpman (2008) suggest that limited contractibility may rather raise the share of outsourcing firms in South when contract incompleteness is especially severe for inputs provided by local suppliers, who should be given stronger incentives to mitigate underinvestment problems. Some other factors, not captured by K/L, might affect the makeor-buy decision in China. For example, according to 2006 White Paper on Small and Medium Enterprises by METI, many Japanese firms cite the bill collection difficulty as a serious problem in China, not in other major offshore locations.

368

ECONOMIC INQUIRY

TABLE 2 Percentages of Offshoring Firms within Each Region China ASEAN North ROW

In & Out

In-only

Out-only

Total

17.26 11.31 15.91 4.62

31.92 39.76 30.68 21.54

50.82 48.93 53.41 73.85

100 100 100 100

Note: Shown is the percentage of firms among the firms sourcing from the respective region.

contrast between outsourcing firms versus insourcing firms among the firms offshoring to the same destination. The outsourcing-versusintegration comparison within the same region can alleviate the potentially serious problems stemming from unobservable cross-regional differences in contracting environment. III.

EMPIRICAL RESULTS

A. Comparisons of Averages Table 3 compares in-sourcing firms with outsourcing firms in terms of capital intensity and other firm-level characteristics related with the firm’s make-or-buy decision. Among various firm-specific variables, R&D intensity is likely to have a critical effect on integration decision. As R&D activities tend to be difficult to monitor, and as the contingencies are difficult to be spelled out for uncertain R&D outcomes, the contract incompleteness problem is supposed to be particularly serious in R&D outsourcing. Firm size is also important in examining offshore outsourcing, as large firms are likely to find suppliers abroad relatively easy because TABLE 3 Differences between Offshore Outsourcing Firms and In-sourcing Firms China Capital intensity K/L R&D intensity Firm size

ASEAN

North

ROW

28.74 42.22 48.53 58.36 (2.9843) (3.1201) (2.7095) (3.0209) 1.88 12.03 23.94 11.69 (0.1357) (0.5441) (0.8910) (0.3305) 37.22 63.75 82.34 106.64 (4.1913) (4.3621) (3.8767) (4.0186)

Note: Shown is the percentage logarithm difference (average in-sourcing firm minus average outsourcing firm). Firms with respective data available are averaged. Figures in parentheses are t-statistics.

of their high reputation, strong headquarter functions, extended overseas networks, or strong bargaining positions. Alternatively, large-sized firms may prefer integrated sourcing based on their rich internal resources within multinationals. Firm size is also likely to be correlated with, and thus is often regarded as a practical proxy for the firm-specific productivity, which has been examined intensively in the firm heterogeneity models of international trade, such as Antràs and Helpman (2004). In any case, the investigations of these variables are worthwhile for our purpose. This paper adopts the standard definitions for these variables as follows. The capital intensity K/L is defined by the book value of machine and equipment divided by the number of workers (regular employees).19 The R&D intensity is R&D expenditure divided by sales. The firm size is measured in terms of the number of workers. The firm-level data for these variables are derived from the national mandatory statistics. On the other hand, skill intensity is unavailable in our data set, because employment or wage data are not disaggregated by skills, occupations, or educational attainments in Japan’s industrial/corporate statistics.20 The percentage difference in logarithm (average in-sourcing firm minus average outsourcing firm) is displayed in Table 3. All the firms of which the respective data are available are averaged. Shown in parentheses are t-statistics for the mean difference between two groups of firms. As the first point to note in Table 3, offshore in-sourcing firms are on average more capitalintensive than offshore outsourcing firms. The gap between them is statistically significant and sizable in any region, ranging from 29% to 58%. This confirms the established sectorlevel evidence, such as Antràs (2003) and Yeaple (2006). Second, the average offshore insourcing firms appear to have higher R&D intensity and to be larger in size compared with the average offshore outsourcing firms. This finding is consistent with our prior, as large R&D-intensive firms tend to have their own FDI subsidiaries and prefer intra-firm sourcing 19. Although tangible fixed assets are usually used, the asset value of machine and equipment is a better measure for constructing K/L, as volatile values of land and plant constructions are excluded. 20. Skill intensity is often included in previous research but found statistically insignificant in the regressions of intra-firm trade shares by Antràs (2003) and in most cases by Yeaple (2006).


because of the internalization gains. Although the gap tends to be relatively narrow and statistically insignificant in R&D intensity, the gap in terms of the firm size is substantial (37%–107%) and significant at any conventional level. We will separate the effect of capital intensity on the firm’s offshore outsourcingversus-integration decision from the R&D or size effects in Section III.C. B. Inter-firm Distributions Although the previous investigations depend solely on average values, we cannot neglect the inter-firm variations among the firms choosing the same sourcing option. Figure 1 displays the cumulative distributions with logarithm capital intensity on the horizontal axis. The upper panel compares In-only firms with Out-only firms in offshoring to China, while the lower panel compares those in the case of North. The graphs for other regions are omitted due to the space constraint, but qualitatively similar. Each graph can be regarded as an empirical counterpart of cumulative distribution function (CDF), and reveals rich distributional information, previously unavailable in average comparisons. In addition to visually inspecting distribution curves, it is informative to calculate the following Kolmogorov–Smirnov test statistics δ for the first-order stochastic dominance21 : δ = nV nO /(nV + nO ) max (1) 1≤i≤N

{FO (Zi ) − FV (Zi )}. The CDF is denoted by F . The suffix V or O corresponds to In-only firms or Out-only firms, respectively. Let Z be the log capital–labor ratio, which is assumed to be randomly distributed. The number of In-only firms and that of Out-only firms are expressed by nV and nO . This statistics has been repeatedly used for testing the productivity premium of exporters relative to non-exporters in established literature, such as Delgado, Farinas, and Ruano (2002). Several noteworthy findings emerge from the inter-firm distributions in our sample. First of all, the Kolmogorov–Smirnov statistics (1) is statistically significant in our sample at the conventional 5% significance levels (1.35 21. We need not distinguish one-sided and two-sided tests in our case because two cumulative distribution curves do not intersect for almost all relevant ranges.

369

for North and 1.68 for China).22 Consequently, the capital intensity distribution of in-sourcing firms FV stochastically dominates that of outsourcing firms FO not only in North but also in China.23 The graphs in Figure 1 visually confirm this test statistics, as the K/L distribution curve of outsourcing firms is located evidently to the left of that of in-sourcing firms. As another point, the four cumulative distribution curves in Figure 1 reveal that more firms choose to integrate in China than in North for any relevant K/L over the entire distribution. As firms with wider ranges of capital intensities (including relatively low K/L) choose to integrate, the average capital–labor ratio of insourcing firms in China should be lower than that in North. This observation is consistent with the transaction-cost interpretation, as China in our regional classification is supposed to have the least developed contracting environment. In sum, this section confirms that our previous result from the mean comparisons is robust. Offshore outsourcing firms tend to be more labor-intensive than in-sourcing firms in each destination, even if we consider inter-firm variations among the firms choosing the same sourcing mode. C. Firm-level Estimations of the Multinomial Choice Model This paper next estimates the firm’s choice model directly at the firm level. Previous sections of this paper and previous literature have aggregated individual offshoring cases either up to groups of firms or to product categories, thus making direct examinations of the firm’s choice impossible. The typical regression used in previous studies is as follows: (2) MV /(MV + MO ) = γ0 + γ1 ln(K/L) + γ2 ln(R & D/Sales) + γ3 Z + ε The share of intra-firm trade in total imports (MV for intra-firm imports and MO for arm’slength imports) is the dependent variable, while included on the right-hand side of the regression are capital–labor ratio, R&D intensity, and/or 22. Practical references for critical values are 1.22 for 5% and 1.51 for 1%. See Barrett and Donald (2003). 23. The gap appears different across regions (the maximal vertical difference between the two curves = 0.231 in North, 0.127 in China), but is statistically significant in both regions.

370

ECONOMIC INQUIRY

FIGURE 1 Cumulative Distributions China 1 0.9 0.8 0.7 0.6 In

0.5 0.4 Out

0.3 0.2 0.1 0

-5

-4

-3

-2

-1

0

1

2

3

4

2

3

4

lnK/L North 1 0.9 0.8 0.7 0.6 0.5

In

0.4 Out 0.3 0.2 0.1 0 -5

-4

-3

-2

-1

0

1

lnK/L

skill intensity along with other controls Z all at the sector level. The error term is denoted by ε. As previous studies depend on aggregated data, the number of industries in their regressions was limited at 23 in Antràs (2003), 51 in Yeaple (2006), and at most 370 in Nunn and

Trefler (2008).24 In all these previous studies, both right-hand side and left-hand side variables 24. Although they also use detailed product-level intrafirm trade data, Bernard et al. (2008) still depend on aggregated SIC four-digit capital-, R&D-, or skill-intensity variables.


are inevitably affected by aggregations over heterogeneous firms within the same industry.25 On the other hand, this paper directly links the probability of the individual firm’s sourcing mode decision with firm characteristics. We estimate the following multinomial logit model26 : (3)

Pr(si = j |x) = exp xi βj

1+

3

exp (xi βh ) .

h=1

The dependent variable is the response probability of the firm i’s offshore sourcing choice s from the following four disjoint categories (In-only, Out-only, In & Out, or Non-offshoring) (j = 0, 1, 2, 3) in each destination.27 Although theoretical models consider the make-or-buy decision, firms normally accomplish multiple tasks, each of which can be formalized by the binary choice. Although previous empirical work based on aggregated industry- or productlevel trade data depend on intra-firm import shares, this paper explicitly distinguishes these four mutually exclusive sourcing modes of individual firms.28 The vector of explanatory variables in Equation (3) is summarized by x, which include the capital–labor ratio, R&D intensity, the firm size (all in logarithms), and/or 24 industry dummies.29 To avoid simultaneity with the offshoring choice at 2006 captured by our survey, all the variables on the right-hand side of x are derived from the official statistics on the previous year 2005.30 All the firms with survey responses and basic firm-level data available (4,721 firms) are included in all cases of our 25. This paper, following Antràs (2003) and Yeaple (2006), focuses on the hypothesis by Antràs (2003), while Nunn and Trefler (2008) and Bernard et al. (2008), based on detailed product-level data, examine the hypothesis by a more general model (allowing domestic sourcing and varying contractibility across inputs) by Antràs and Helpman (2008). 26. The next section will check whether or not the assumption of IIA in the multinomial logit model affects our results. 27. The numerator on the right-hand side of Equation (3) is one for j = 0 because probabilities sum to one. 28. While the regression of import share is useful to directly compare our results with previous literature, no data on the value of offshoring is available in our survey. Furthermore, at the firm level, the multinomial choice model is more appropriate for describing the individual firm’s decision. 29. We will add other explanatory variables in the next section to check the robustness of the results from the specification with these three variables. 30. In the next section, we will use the variables at 2000 as instrumental variables to check the robustness of our results.

371

estimations.31 Compared with previous studies, our sample size and our direct access to firmlevel offshoring data have an advantage in capturing inter-firm variations within each industry. Although the two-country framework is standard in theoretical trade models, there are many heterogeneous offshoring locations in the real world. Besides, tasks offshored to different destinations are likely to vary drastically in factor content, especially for tasks offshored to South compared with those to North.32 Considering this problem, Nunn and Trefler (2008) define their intra-firm trade share variable separately for each source country as a distinct productcountry-specific variable, rather than defining it for each product common across all sourcing countries. Inspired by their approach, this paper compares outsourcing firms versus insourcing firms among the firms offshoring to the same destination. In other words, our multinomial logit model considers the choice of outsourcing to a given region, in-sourcing to that region, both outsourcing and in-sourcing to that region, or not offshoring to that region. The last category could include firms actively offshoring to other destinations.33 This estimation assumes that firms first choose sourcing locations and then decide make-or-buy. Although we have not tested this first stage, the comparison of firms offshoring to the same destination helps us investigate the effect of capital intensity on the offshore make-or-buy decision. If we aggregate offshoring firms across a wide range of heterogeneous destinations, the relation with capital intensity will become inevitably unclear because contracting environments are likely to critically differ across destinations, especially between South and North.34 The multinomial logit estimation results are reported in Table 4. Robust standard errors are 31. Firms without K data are excluded. R&D expenditures for firms without R&D data are set at zero. Negligible −8 10 is added to R&D–sales ratio before taking logarithm. As merely