U.S. International Trade in Other Private Services: Do Arm’s Length and Intra-Company Trade Differ? ∗ Deniz Civril ¸ Catherine L. Mann†

August, 2015

Abstract U.S. international trade in so-called ‘other private services’ (OPS) has more than tripled in the last decade to account for 13 percent of total exports and 5 percent of total imports. This paper examines the relationship between trade in OPS, service-sector capacity, technological similarity, and the role for FDI using a gravity model approach. We are particularly interested in whether these factors differentially affect affiliated trade vs. unaffiliated trade. We find that total FDI and services trade are complementary. FDI is more robustly associated with unaffiliated parties. Technological similarity is important for affiliated export. Common language is highly significant for import trade but especially for affiliated import. Distance, on the other hand, is not a key player in service trade.

∗

Corresponding author: Visiting Assistant Professor at the School of Foreign Service, Georgetown University, 37th and O Streets, N.W. Washington D.C. 20057. Email: [email protected]. Website: https://sites.google.com/site/dcivril/. † Professor, International Business School, Brandeis University

1

1

Introduction

U.S. imports and exports of so-called ‘other private services’ (OPS), which includes such services as education, finance, telecommunications, insurance, and business, professional, and technical services, are increasing rapidly. Between 1992 and 2009, OPS exports increased from $50.3 billion (30.7% of total service export) to $238.3 billion (48.2%) and OPS imports increased from $25.4 billion (24.5% of total service import) to $168.9 billion (50.2%). On the positive side, considering the U.S. current account deficit, the positive contribution of it to the balance of payment1 as a largest share of service sector in trade makes its progress worth exploring. On the negative side, from the popular press, one has the impression that the U.S. multinational corporations increasingly are setting up services affiliates in low-wage countries, for example, India, to do back-office operations, financial analysis, insurance claims processing, and software development. Popular hype and the coincident trends in the data as well as policy-related issues such as trade-agreement in service sectors have led to considerable interest in this ‘services offshoring’ phenomenon. In this paper, we look at the extent and nature of the U.S. international trade in OPS, considering in particular whether the factors driving trade through the U.S. multinationals (intra-company trade) differ from factors driving arm’s-length trade. Despite the trade in service sectors such as tourism and travel, many other services in OPS were the ‘non-traded’ activity - an activity that took place within the boundaries of the firm. It had been assumed that services could not be fragmented into stages of production and ‘out-sourced’ (outside the corporate umbrella) due to high transactions costs and because services were embodied in the core activities and products of the firm. For example, it had been assumed that services demanded close proximity between buyer and seller (loan origination at the local bank) and that services production had important economies of scope (coding of computer programs as integral to software application design). International fragmentation of the production process faced the additional hurdles posed by the local business climate (intellectual property protection or corruption), the need for cultural sensitivity (at minimum the correct language), among other barriers. All told, services were seen as not amenable to the business strategy of vertical fragmentation whether onor off-shore. Technological change, the expansion of activities of multinational enterprises as well as policy change and changes in customer and business attitudes over time, have eroded these attributes of services - transactions costs and functional integration - that heretofore made them ‘non-tradable’. But, for off-shore sourcing to take place (whether ‘out-sourcing’ at arm’s-length or intra-company), it is not just technology in the U.S. that matters. Globalization of services through trade and direct investment requires that reduced transactions costs and functional fragmentation take place in the trading partner as well. That is, globalization of services is limited unless both sides of the transaction have the key technologies that enable fragmentation. Whether that fragmentation takes place underneath the corporate umbrella (intra-company transactions) or outside the corporate umbrella (arm’s-length trade) and to what extent the transactions are international in scope are questions posed in this paper. U.S. export and import data in OPS provide us an opportunity to answer these questions because affiliated and unaffiliated trade between the U.S. and other countries are available in this category of service trade. The next section explains and analyzes the trade data in OPS and other data sets used in the 1

See Figure 10 in Appendix 7.1.

2

paper. Section 3 summarizes the literature review of the international trade in services and how this paper stands among them. Section 4 discusses the econometrics of gravity model and the implications to the OPS data. Section 5 presents the results with robustness checks and Section 6 concludes.

2

Overview of the U.S. Data on International Services Global Engagement Figure 1: Service Trade

Note: The shaded categories indicate the categories that we analyze in OPS trade.

According to UN et al. (2002) classifications, there are 4 modes of services: cross-border supply (mode 1), consumption abroad (mode 2), commercial presence (mode 3) and presence of natural persons (mode 4). Regarding commercial presence, the sales of affiliates in the market, in which they are located, are a very important part of global engagement. Although, transactions do not cross the U.S. borders, they provide the service there. In fact, BEA provides data in two categories as in Figure 1: cross-border trade and services supplied through affiliates. “Cross-border trade covers transactions between a U.S. resident and a nonresident. These transactions are recorded as U.S. exports and U.S. imports in the international transactions accounts (ITAs), or balance of payments accounts. Services supplied through affiliates cover transactions between foreign affiliates of U.S. companies and foreign residents and those between U.S. affiliates of foreign companies and U.S. residents. These transactions are not considered U.S. exports or U.S. imports in the ITAs because under the residency principle of balance-of-payments accounting, affiliates of multinational companies are considered residents of the countries where they are located rather than of the

3

country of their owners.”2 In 2008, sales of services by affiliates of U.S. parents in host markets were larger than twice the total cross border service exports. It is same for the sales of the foreign affiliates in the U.S. and total service import.3 Unfortunately, because services affiliate sales data by host market are not available at the disaggregate service category levels as in cross-border trade, we cannot incorporate them into this analysis.4 Hence, we focus only on cross-border transactions. The analysis is only for the category of Other Private Services (OPS) as affiliated and unaffiliated data by country is not available for other service categories. The paper addresses the important issue of multinationals’ structure of production through the lens of arm’s-length vs. intra-company trade in OPS, where we do have data by country and affiliation. The categories that are outlined with dots in Figure 1 are the categories we use in this study. We disregard the affiliated trade of foreign companies because the transaction by country is most of the time missing not to disclose firms. On the other hand, as we discuss later in this section, it is relatively stable for the time period we study. Table 1: Trade in Services 1992-2009 (millions of dollars) and Subcategories

Total Private Services Travel (%) Passenger fares (%) Other transportation (%) Royalties and license fees (%) Other Private Services (%)

Export 1992 2009 164,024 492,131 33.4 19.1 10.1 5.3 13.1 7.3 12.7 19.9 30.6 48.2

Import 1992 2009 103,469 347,720 37.6 21.3 10.2 7.2 22.9 12.3 4.9 9.0 24.5 50.2

Notes: Total percentage may not be equal to 100% due to rounding. Source: BEA

OPS is a sub-category of the U.S. classification system for international trade in services. Other sub-categories include Travel, Passenger Fares, Transportation, and Royalties and License Fees (film and tape rentals are included in OPS). OPS is the largest category and the fastest growing. As seen in Table 1, the share of Travel was more than OPS in 1992. But in 2009, OPS covered half of the service trade both in export and in import. The change in import is more dramatic than the one in export, which needs careful analysis to foresee the future of current account balance. While the U.S. has been experiencing huge deficit in goods, positive contribution of services has been a relief. Therefore, the drivers of export growth and import growth and their differences are important. Components of OPS include: Education, Finance, Insurance, Telecommunications, and Business, Professional, Technical services (BPT). BPT is by far the largest sub-category, with financial services a distant, but rising second for exports, and insurance a distant but large category of 2

See http://www.bea.gov/international/international services.htm for more detail. See Figure 11 and 12 in Appendix 7.1. Lennon (2009), for example, looks at if the sales of affiliates to local markets substitute for or complement cross-border service trade. For much of the existing empirical work on goods trade and foreign direct investment such as Brainard (1997), Hanson, Mataloni and Slaughter (2001), Hanson, Mataloni and Slaughter (2005), Yeaple (2003) and Borga and Zeile (2004), the question of interest has been exactly on the relationship between goods exports and these affiliate sales. 4 In fact, even at the aggregate level, there are missing data for most of the countries because of disclosure. 3

4

Table 2: Trade in Other Private Services 1997-2009 (millions of dollars) and Subcategories

Other Private Services Education (%) Financial services (%) Insurance services (%) Telecommunications (%) Business, Professional, Technical services (%)

Export 1997 2009 83,929 237,348 9.9 8.4 16.5 23.3 2.5 6.2 4.7 3.9 50.6 48.9

Import 1997 2009 43,154 174,573 3.2 3.3 13.9 9.7 13.7 32.7 19.3 4.2 48.5 48.6

Notes: Total percentage may not be equal to 100% due to rounding. Disaggregated data for the categories of OPS are available starting from 1997. Source: BEA

imports as documented in Table 2. Cross-border trade between members of multinational corporations is an important channel through which international trade takes place. For OPS, such affiliated receipts (exports) and payments (imports) represent about one third for exports and about 40 percent for imports. Figure 2 shows the shares of affiliated and unaffiliated OPS trade in total services. Furthermore, it disaggregates the affiliated trade into two: trade between U.S. parents and their affiliates vs trade between foreign parents and their affiliates. While the trade between foreign parent and their affiliates is relatively stable as a ratio, the driving force of increasing affiliated trade comes from the trade between the U.S. parents and their affiliates. Figure 3 emphasizes this point more clearly, indicating the ratio shares of the U.S. affiliated and foreign affiliated OPS trade in total OPS trade, which is about 20% for exports and 25% for imports.

5

Figure 2: Export and Import Shares of OPS in Total Services

.4 .3 .2 .1 0

0

.1

.2

.3

.4

.5

Import

.5

Export

1990

1995

2000 year

2005

2010

1990

OPS Affiliated−US

1995

2000 year

2005

2010

Affiliated Affiliated−Foreign

Figure 3: Export and Import Shares of Affiliates in OPS

.25 .2 .15 .1

.1

.15

.2

.25

.3

Import

.3

Export

1990

1995

2000 year

2005

2010

1990

Affiliated−US

1995

2000 year

Affiliated−Foreign

6

2005

2010

2009

Source: BEA

7

2009−1992 India

pines

Philip

nesia

a

Chin Indo

a

il

and

Afric

Braz

Thail

South

ntina

zuela

Arge

ysia

Chile Mala

Vene

a

o

0

.2

.4

.6

.8

2009

Mexic

Kore

bia

i Ara

l

d

Israe

ealan

Spain

Italy

ce

den

Fran

Swe

Saud

n

dom

Japa King

New Z

Unite d

alia

Austr

ny

urg

Germ a

mbo

da

Kong

Cana Luxe

ium−

Belg

Hong

s

d

rland

erlan

Switz Neth e

y

apore

Norw a Sing

−.2 ay

den

a

il

India

Philip pines

a Indo nesia

Chin

and

a

Afric Thail

South

zuela Braz

Vene

ysia ntina

Arge

Mala

Chile

Mexic o

Kore

bia

i Ara

l

nd Israe

Zeala

Spain

Italy

Fran ce

Swe

Saud

New

n

gdom

Japa d Kin

Unite

Switz erlan d Neth erlan ds Hong Kong Cana Belg da ium− Luxe mbo urg Germ any Austr alia

apore

Norw Sing

−.2

0

.2

.4

.6

Figure 4: U.S. Parent-Affiliate Trade: Share of U.S. Exports of OPS Ranked by GDP per capita, PPP (constant 2005 international $)

2009−1992

Source: BEA

Figure 5: US Parent-Affiliate Trade: Share of US Imports of OPS Ranked by GDP per capita, PPP (constant 2005 international $)

We can disaggregate total OPS trade5 for 31 individual countries.6 Graphical analysis of the international structure of the U.S. MNC trade points to some intriguing changes in the U.S. MNC behavior. Figure 4 and 5 show the ratio of U.S. intra-company trade to the sum of intra-company and arm’s-length trade for trading partner exports and trading partner imports for 1992. Also shown is how that ratio has changed over time from 1992 to 2009. Intra-company export share is higher for affiliates in richer as compared to poorer trading partners, but that share has declined for many richer countries and risen for many poorer countries. Intra-company import share is higher for relatively richer trading partners and has similar trends as in exports; the intra-company import share is lower for relatively poorer trading partners, but that share has risen dramatically since 1992. What kind of properties of these countries make such a change is the main interest of this paper. We estimate a gravity equation for OPS trade between the U.S. and 31 countries for the years 1992-2009. In addition to the OPS trade data, the following data are used7 : • Economic size: Service expenditure (constant 2005 US$) from World Development Indicator (WDI)8 • Economic cost or level of development/similarity of tastes: GDP per capita (constant 2005 US$) from WDI 5 Real OPS trade is calculated by using import and export price indices by BEA: NIPA table 4.2.4: Price Indexes for Exports and Imports of Goods and Services by Type of Product. 6 See Table 6 for the list of countries. BEA publishes information for certain countries for a long time and puts all other trade transactions under ‘Others’ categories for each region. It seems that the economic size and significance of the countries were important when they first chose these countries. The magnitude must be also important that for the latest 4 years, they also publish data for Ireland. Since we cannot identify countries within the ‘Other’ section, we assume that these trading partners have too little trade to report. On the other hand, for two countries that do report -Bermuda and Taiwan- much of the additional data needed for our econometric estimations are not available, so these two countries are dropped from the sample. Ireland has data for only last 4 years, which is also dropped. Finally, trade with ’International Organizations’ and trade that is ‘unallocated’ is subtracted from total trade in OPS. As a result, our sample of 31 countries covers somewhat more than 75 percent of OPS transactions. The reporting thresholds as well as disclosure issues present additional problems for the time series analysis, especially for the trade data of the U.S. affiliates with foreign parents. A majority of data either fall below the $0.5 million threshold or are missing values due to the disclosure issues. However, we do not focus on the rationale for trade by the U.S. affiliate with foreign parents (our focus is on trade by US parents with their affiliates abroad, and arms-length trade). For the export side, both total and US parents with their affiliates, there is no imputation. For total import, there are 8 imputation but most of the imputation is for a single year for each country. For affiliated import, there 14 imputations and in some cases 2 or 3 imputations for a single country. However, considering that there are 19 years per country and 558 data points in total, 14 imputations are relatively small. 4 of these 14 imputations are for those that fall below the $0.5 million threshold. These 4 missing values are replaced with $0.3 million, which is an average value. There is an indirect impact to the unaffiliated trade though. Unaffiliated trade is calculated by subtracting affiliated trade from total trade. For the export side, 49 values, for the import side, 38 values are imputed by interpolation due to the disclosure problem. However, note that the ratio of OPS trade between the U.S. affiliates with foreign parents are small as opposed to OPS trade between the U.S. parent with foreign affiliates, where we have data. 7 Similar to OPS data, missing values problems are solved by interpolating or plugging previous or later year values as long as there is only one missing year. Data for Belgium and Luxembourg are also summed up or aggregated properly as service data is aggregated in the BEA. 8 For import side: Services value added by WDI is used to proxy demand of the U.S. For export side: US Service Production from BEA value added by industry (http : //www.bea.gov/iT able/indexi ndustry.cf m) is used to proxy supply of the U.S. Services included are: Information, Finance, insurance, real estate, rental, and leasing, Professional and business services, Educational services, health care, and social assistance. For other countries, services value added from WDI is used for both import and export.

8

• Technology : Internet users and telephone lines (per 100 people) from WDI • U.S. FDI presence: U.S. Direct Investment Position Abroad on a Historical-Cost Basis from BEA • Bilateral trade: Bilateral trade agreement data are constructed from the website of Office of the U.S. Trade Representative9 • Distance, language, border: Taken from the two data sets provided by The Centre d’Etudes Prospectives et d’Informations Internationales (CEPII)10

3

International Trade in Services: Review of Empirical Literature

Empirical analysis of international trade in services is relatively recent, mostly because data have been limited. There are two strands of analysis, one based on the income and relative price framework and one based on the gravity model framework. The income and relative price framework (Adler (1945; 1946), Chang (1945-46; 1948)) relates the volume of exports (imports) to real foreign (domestic) income and relative prices.11 The focus of an analysis with this framework is usually on the elasticity of trade of a particular country with respect to income and relative prices so as to gauge the impact on trade of growth at home or abroad, or the impact on trade of a real currency devaluation. Considering the global imbalances, i.e. high current account deficit for some countries but high current account surplus for some others, it is especially important for policymakers to understand the external balance of a country, forecast the future as accurate as possible and address the sustainability problem. The income and relative price framework has been used to analyze U.S. trade in services in a couple of papers. Mann (2004) examines a time series panel of OPS, using unaffiliated trade disaggregated by type of OPS and by source and destination. She finds that income elasticities are larger for services exports than for services imports and that relative prices generally are not significant. Marquez (2005) focuses on time series properties (quarterly data) for the major categories of services (travel, passenger fares, transportation, and OPS). He notes that disaggregation by service category is key, finding that the income elasticity of U.S. exports with respect to income is greater than that for U.S. imports. Pain and van Welsum (2004) (for U.S. exports) and van Welsum (2004) (for U.S. imports) consider services disaggregated by major type as in Marquez (2005). These papers also consider the role for FDI in facilitating trade. The different measures of FDI such as the stock of FDI, the sales and the gross product of the U.S.-owned foreign affiliates are used. They find that the relationship between trade flows and FDI in services differs by service category, even to the extent that trade and FDI are sometimes a substitute and sometimes a complement. Relative prices are significant in the long-run when all categories are pooled. The results from the pooled sample appear to rest in particular on the travel and passenger fare categories. A second line of empirical research on international trade in services uses the gravity model framework. Developed by Tinbergen (1962) and P¨oyh¨onen (1963) to explain bilateral trade flows 9

http://www.ustr.gov/index.html http://www.cepii.fr/anglaisgraph/bdd/distances.htm 11 See Hooper, Johnson and Marquez (2000) for a concise summary of the model. 10

9

by trading partners’ economic activity and geographic distance between countries, the gravity model is a common approach to modeling bilateral trade flows among a set of countries. The pioneer for theoretical model was Anderson (1979). He used utility function to derive a well developed model. There are now multiple theoretical foundations for gravity equation, including Feenstra, Markusen and Rose (1998), Mirza and Nicoletti (2004), Helpman, Melitz and Rubinstein (2007), and Head, Mayer and Reis (2009). These theoretical underpinnings provide for various formulations of the gravity specification. Briefly, consider that Vij is the volume of trade (bilateral, imports or exports) between two countries, i and j. The volume of trade can be explained by the characteristics of the source country, Si , and the destination country, Dj , and some other factors, Rij , that facilitate or discourage trade between them. Vij = Si ∗ Dj ∗ Rij

(1)

S and D usually contain economic size, often proxied with GDP and GDP per capita (to account for intra-industry trade effects that may be associated with countries of similar incomes but varied tastes) as well as other country specific variables such as corruption level. R includes bilateral measures for trade barriers, proxied by distance, tariff rate, business regulation as well as trade facilitators, such as trade agreements, language or ethnic similarities between the source and destination markets. The log-linear version of Equation 1 has been estimated using both cross section and panel data for various country groups to explore the factors explaining trade and the policy implications related to these factors. There is by now a huge literature on the gravity model specification for trade in goods. Literature on trade in services is more limited, but recent releases of new data by the OECD and Eurostat has promoted a flurry of analysis. At this point, however, it is difficult to compare the results across the various studies because the focal interest varies.12 One of the papers that uses the OECD data set is by Gr¨ unfeld and Moxnes (2003) for 22 countries and 55 host countries with the years 1999 and 2000. They consider the relationship between services trade (aggregated over all types of services) and foreign direct investment (aggregate over all types of manufacturing and services FDI) controlling for trading partner size, as well as a trade restrictiveness in the services sectors (from the Australian Productivity Commission) and corruption in the importing country. They find a strong home bias in services (high GDP coefficients), important effects from barriers to trade as proxied either by regulations or by corruption, and that services exports and FDI are complementary which is consistent with some of the findings of Pain and van Welsum (2004). Kimura and Lee (2006) use a sub-sample of 10 ‘home’ OECD countries that trade in a broad range of services (including for example transportation) with between 27 and 47 partner countries for two years of observations (1999 and 2000). Among their objectives is to examine differences between goods trade and services trade using similar specifications, and to consider whether there is any complementarity between trade in services and trade in goods. They find that ‘distance’ is more important for services than for goods. Service exports and imports are not complements, but that goods exports and services imports are complements. Gravity model is also applied in Walsh (2006), who uses OECD database of total service imports 12

Francois and Hoekman (2010) summarize the theoretical advancement and the recent developments in service trade literature in detail.

10

and imports for four sub-sectors (travel, transport, government and other commercial), covering up 27 countries with up to 50 trading partners for the years 1999-2001. He applied the Hausman-Taylor model as an estimation strategy to deal with heterogeneity and time-invariant variables problems. While the GDP per capita and a common language are important determinants of service trade, distance is mostly not significant. Lennon, Mirza and Nicoletti (2009) use 17 OECD countries trade with 27 partners for the two year sample (1999 and 2000). They focus first on the network requirements for trade in services to take place (e.g. for trade in telecommunication services, both countries need to have domestic telecommunications capabilities) and second on the human capital requirements for trade in services. Using aggregate trade in services, they find that product market regulation in services and labor force quality affect bilateral trade in services. The results of analysis at the aggregate level are also supported at the industry level by using Air Transportation Industry data of 27 OECD countries in the 1996-97 travel season. They also improve the estimation method with Transformed Least Squares to solve the multicollinearity problem arising from exporter, importer and time fixed effects. Head et al. (2009) use Eurostat data to focus in particular on the role of distance for trade in services. The Eurostat data allows for a finer disaggregation of service type into finance and ‘other commercial services’ for example. The time series is longer than that for OECD data (1992-2006) and 65 countries are covered. However, this country coverage is limited to focus on intra-European trade, and selected non-European trading partners, such as Japan and the U.S. Some flows with India and China may be available for some years and for some types of service transactions. They analyze the importance of trade barriers in service trade such as physical distance, differences in times zones, languages and legal systems on the wage premium. A firm would be willing to pay for nearby service providers but this distance effect has been declining by time. Amiti and Wei (2004) use IMF data to establish some facts regarding the magnitude of services trade, focusing on business services, for a variety of countries. They analyze the impact of such trade on labor markets, but do not address the determinants. Some other studies only focus on the service trade of the U.S. For example, Freund and Weinhold (2002) look at the internet impact on the growth of service trade using very detailed OPS data of the U.S. for the years 1995-1999 (31 countries and 14 subcategories). They conclude that a ten percent increase in the internet variable causes a 1.7-percentage point more export of services. The other example is Lennon (2009) who explores the relationship between total services trade and FDI using bilateral data on the U.S. foreign affiliates’ sales. She finds that there is a complementary relationship, which is robust for trade in services but not trade in goods. In our analysis, we will focus on only the category of OPS trade, disaggregated by source and destination but not disaggregated by types of OPS as in Figure 1 or in Table 2. Pooling all service categories into one group can be misleading. For example, the reasons of travel trade can be totally different from the reasons of OPS trade . Marquez (2005) points out the importance of disaggregation. In the literature, as mentioned in this section, there are also papers use OPS data. So, we think that it is not misleading to use OPS trade only. Since the study is about the trade between the U.S. and other countries, it differs from other cross country analysis. It can be argued that the comparison between this study and cross section study is not appropriate. However, assuming that other developed countries behave like the U.S., we can generalize the results. Since the period we have is longer than other studies it creates an 11

opportunity to understand service trade better and give a different perspective. The findings are also important to comment on the future of current account balance of the U.S. The most important contribution of this paper is to analyze the volume of OPS trade by affiliation, which is not available for other countries. It also considers the role for FDI, along with other country characteristics that may impact the OPS trade by affiliation.

4

Econometrics of Gravity Models and Empirical Implementations for OPS Data

The early studies usually use cross section datasets for gravity model. While the ordinary least square (OLS) has been the common method to estimate the coefficients, the critics about the inconsistencies between theoretical underpinnings and estimation method lead the researchers to benefit from the richness of panel datasets. To be more clear, assume that we have a model: yijt = β0 + β1 xijt + β2 xit + β3 xjt + β4 xij + ijt

(2)

for i = 1, . . . , N , j = 1, . . . , N , i 6= j and t = 1, . . . , N . The volume of trade between two countries at time t, yijt , can be explained by the independent variables in x (which can have variations in three dimensions, ijt -such as bilateral trade-, or two dimensions, it -such as gdp- , jt, ij - such as distance) and the disturbance term ijt . Assuming that the disturbance term is independently and identically distributed with zero mean and constant variance, this equation can be estimated using OLS. However, if this assumption is violated, it has to be addressed. In gravity model, a country can export to or import from the two other countries in different amounts even though these two have similar properties in terms of GDP, distance, etc. It is possible to have multidimensional heterogeneity. So, the error structure can be decomposed into time-specific and time-invariant individual components to model the source of heterogeneity. 3-way model 2-way model

: ijt = αt + αi + αj + uijt : ijt = αt + αij + uijt

When the data consists of three dimensions (exporter, importer, time), the error structure includes fixed effects for all, which is called 3-way model. But it is also argued that each pair of countries has its own fixed character. For example, US-Canada and US-Mexico trade might have different individual fixed effects. Therefore, it is better to include pair fixed effects, which is called 2-way model. One can further discuss that country-pair effects are asymmetric, αij 6= αji . The choice of error structure and the problems arising from this heterogeneity as well as the remedies depend on the data in hand. In this study, we have panel data, so there is time component, t. On the other hand, for both export and import side, the pair country is the U.S. that has OPS trade with 31 countries. Therefore, when time dummies are used, we lose time series variables of the U.S. in the regressions. Although we would like to keep the time series variables of the U.S. by ignoring time dummies, they are very important part of the heterogeneity and control for overall change in service trade trend through years as documented in Figure 6 and 7. Their significance are very high in all our specifications. It is essential to keep these time dummies. Therefore, we create the activity variable 12

of the U.S. for each country by dividing the expenditure/production of the U.S. according to trade share of the U.S. in goods. We assume that the share of goods trade approximates the potential economic ties between a country and the U.S. For export side13 :

Si = U.S. supply for country i =

U.S. exporti ∗U.S. Service Production (3) Total U.S. export for 32 countries

For import side:

Si = U.S. demand from country i =

U.S. importi ∗ U.S. Service Expenditure Total U.S. import for 32 countries (4)

Figure 8 and 9, on the other hand, picture the heterogeneity by country. Since the U.S. is the pair country, using country fixed effects automatically dictates the 2-way model - αi−U S . When the fixed effects are randomly distributed across panels and not correlated with any other explanatory variables (Cov(xm , αk ) = 0, where m = ijt, it, jt and k = i, j, t or ij, t), the only problem in the estimation of Equation 2 will be the heteroscedasticity. The errors become serially correlated within panel because of the fixed component they have. The common way to correct this problem is to use feasible generalized least square (FGLS), which transforms the data such that one can obtain the correct standard errors and statistics. It is also called random effect model. However, Hausman test finds that αi−U S is not random in our data. Furthermore, the asymptotic properties of FGLS have poor small sample properties, especially when the cross section (N) is larger than time section (T). As T goes to infinity, fixed effect model converges to random effect. But the data we have do not have these properties with 31 countries and 18 years. Beck and Katz (1995) argue that the variance-covariance estimates of FGLS are unacceptably optimistic for this type of data. Monte Carlo experiments show that the standard errors are underestimated by %50 or more. Therefore, we are cautious about random effect model. If the assumption of the uncorrelated fixed effects with the explanatory variables does not hold, endogeneity problem arises without fixed effects. One solution to the endogeneity can be to find the reason and alleviate it with additional data. Theoretically, Anderson and van Wincoop (2003) assert that the trade cost, which is called ‘multinational resistance term’, appears in the model but it is not estimated in the model and left in the error term. This trade cost is correlated with the explanatory variables and so endogeneity is inevitable. Anderson and van Wincoop (2003) use nonlinear least square to estimate ‘multinational resistance term’ with their cross section data. They also suggest using fixed effects to solve the endogeneity problem using panel data. If one assumes that the endogeneity is the result of ‘multinational resistance term’ only, it can be proxied by a 13 U.S. Service Production is from BEA value added by industry (http : //www.bea.gov/iT able/indexi ndustry.cf m). Services included are: Information, Finance, insurance, real estate, rental, and leasing, Professional and business services, Educational services, health care, and social assistance

13

Figure 7: OPS Import by Year

19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09

19 92 19 93 19 94 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09

4

4

6

6

8

8

10

10

Figure 6: OPS Export by Year

year ln(X)

year

mean(ln(X)) by year

ln(M)

Figure 9: OPS Import by Country

30

20

10

0

4

4

6

6

8

8

10

10

Figure 8: OPS Export by Country

mean(ln(M)) by year

0

Countries Ranked by GDP per capita, PPP (constant 2005 international $) ln(X)

mean(ln(X)) by country

10 20 30 Countries Ranked by GDP per capita, PPP (constant 2005 international $) ln(M)

mean(ln(M)) by country

variable. For example, Baier and Bergstrand (2002) suggest ‘remoteness variable’.14 The papers of Carr`ere (2006) and Kimura and Lee (2006), for example, follow this approach to proxy for the term. While solving the endogeneity with ‘multinational resistance term’ is a big assumption to claim, the advantages of the data sets they use are to have multiple countries and to study goods trade. It is relatively more obvious to proxy for the resistance term for goods trade than for services trade. We do not have this option. Therefore, we use the fixed effects for countries in our regressions and consider it as our base model. On the other hand, the drawback of using fixed effects is to lose the ability to estimate coefficients of time invariant variables such as distance, common language, common border when 2-way model is used. If one ignores or justifies the 3-way model, it is possible to estimate the coefficients of country pair variables such as distance except other time-invariant and country-specific ones. However, as pointed out earlier, in each case the pair country is the U.S. in our case. When the country fixed 14

It is defined as:  X 1/1−σ N Ri = Yk (Dki )1−σ , where N is the number of countries, D is distance and Y is GDP. k=1,k6=i

.

14

effects are used with the OPS trade data, they are considered as αi,U S , in which case we cannot estimate even distance. In some papers (for example Filippini and Molini (2003), Martinez-Zarzoso and Nowak-Lehmann (2003)), estimated individual effects are regressed on known individual variables. This approach disregards the possible correlations between observed and unobserved fixed effects which we try to solve at the first place. Plumper and Troeger (2007) suggest three-stage procedure to estimate time-invariant variables in panel data with unit effects. From the first stage, one can obtain the unit effects by running fixed-effect model. The second stage breaks down the unit effects into its time invariant variables. And the third stage reestimates the model including the time-invariant variables plus the error term of second stage to account for the unexplained part of the unit effects. However, it should be noted that there are serious discussions about this method.15 Hausman and Taylor (1981) point out one other drawbacks of fixed effect estimations besides eliminating time invariant variables. It is possible that the full efficiency might not be reached since within-groups estimator does not take cross-variation between groups into consideration. The method they suggested, Hausman-Taylor Method (HTM), has been applied for gravity model in many papers.16 HTM uses the deviations of time varying variables from individuals means and exogenous variables as an instruments and apply random effect regression. It does not require any external instruments but it might be difficult to detect exogenous and endogenous variables in the model.17 Also, time series dimension of our data is smaller than the panel dimension, which makes this model inefficient.18 At the theoretical level, it has been a question if it is possible to use the models of goods trade for services trade. Although the advances in technology shade the importance of the proximity of consumers and suppliers for most of the services and actually make them tradable, it still requires interaction of inputs from both exporter and importer country, i.e. “jointness in production” (Francois and Hoekman, 2010). Special to services trade, Lennon et al. (2009) proposed a simple model that takes the countries’ co-production of traded services into consideration. The production and/or delivery of services depend on various tasks performed in both source and destination 15

See Green (2011) for the critics and see Plumper and Troeger (2011) for the response. Some examples cited in this paper are Carr`ere (2006), Walsh (2006). 17 See Kepaptsoglou, Karlaftis and Tsamboulas (2010) for the literature reviews and methods used from 2000 to 2010. 18 The recent papers in trade in goods also focus on ‘zeros’ in bilateral trade (Baldwin and Harrigan, 2011; Helpman et al., 2007) using Heckman selection model (Heckman, 1979). It is argued that the absence of trade contains information. Failing to incorporate this information in subsequent regression analysis biases the coefficients in the regression of observed trade flows. We have not found any paper about service trade which uses this approach. It is most probably because of the availability of service trade data. It should be also noted that Silva and Tenreyro (2006) suggest Poisson Pseudo-Maximum Likelihood (PPML) estimation technique to overcome the problems of log-normal gravity equation, which are the estimation difficulties related to zero trade, the bias estimates due to logarithmic transformation and the failure of the homoscedasticity assumption. They estimate gravity model in its original multiplicative form using PPML method. In our case, there are 171 countries that trade in goods with the U.S., but we have information for 31 countries that trade in OPS. Selection of 31 countries out of 171 causes zero inflated data. Furthermore, even if we apply selection model, this selection is not time variant but time invariant. In the literature, panel data sample selection models are still under discussion, especially using panel fixed effects in the first stage. There is no study regarding time invariant selection (there is no change of status in the first stage) as far as we know. Therefore, both Heckman selection and PPML are not applicable to our data. Zero trade and log-normality of gravity model are beyond the scope of this paper. 16

15

country. So, the break in this chain of tasks can result in unsuccessful transactions. They add this property of services by incorporating the performance quality of countries to the model besides the input costs.19 They also consider service expenditure as an economic activity variable instead of GDP. Making analogy to their model, we modify Equation 1 as: Vij = Servi ∗ Servj ∗ wi ∗ wj ∗ qi ∗ qj ∗ τij

(5)

Remember that for both export and import side, the pair country is the U.S. Therefore, for simplicity, we drop ‘j’ and replace with ‘US’ when necessary. Vi Servi(U S) wi(U S) qi(U S) τi

: : : : :

export/import in OPS between country i and the U.S. services expenditure the average cost in the tradable services sector, GDP per capita as a proxy performance quality transaction cost

The performance quality is modeled as:  qi = Tech-Similarityi = 1 − where, Techi(U S)

Techi Techi + TechU S

2

 −

TechU S Techi + TechU S

2

: the simple average of internet users and telephone lines (per 100 people)

This index is bounded between 0 (absolute divergence in technological similarity) and 0.5 (equal technological similarity). Distance is a proxy for the transaction cost of trading (τi ). But distance is not the only determinant of trade costs. Properties such as common language, bilateral trade agreement are also effective. So following Carr`ere’s (2006) transaction cost20 : τi = Dγi 1 eγ2 langi +γ3 bilaterali +γ4 borderi where, Di langi bilaterali borderi

: : : :

distance equal to equal to equal to

between country i and the U.S. 1 if English is native language, otherwise 0 1 if a country and the U.S. have a bilateral agreement, otherwise 0 1 if a county has a border with the U.S., otherwise 0

Taking the logs of Equation 5 and adding the time component, one can get the linear function to be estimated:

lnVit = θ0 + θ1 lnSit + θ2 lnSU S−t + θ3 Rel-GDPpercapitait + θ4 ln(Tech-Similarityit )+ θ5 lnFDIit + γ1 lnDi + γ2 langi + γ3 bilaterali + γ4 borderi + it 19

(6)

See Appendix 7.2 for a brief derivation of their model. We added other variables such as human capital (average school years as proxy) but these variables are highly correlated with other variables and do not vary much as there are only 31 countries. 20

16

Notice that the model is augmented with two variables. GDP per capita enters the equation in terms of relative wage. The other one is FDI of the U.S. in country i, FDIit . FDI has expanded rapidly and plays an important role in the trading activities between countries. The presence of the U.S. firms might affect not only the demand for services by a country but also supply to the U.S. We use fixed effect model as our benchmark regression. Individual fixed effects control the possible omitted variable bias and eliminate the resistance term we mentioned before. In order to be able to estimate time invariant variables, we also run random effect model although there are problems about estimation without fixed effects as discussed earlier. Because time invariant variables - distance, language, bilateral trade agreement, border - are important determinants of trade. Both fixed and random effect models do not deal with temporally and spatially correlated errors. When there is autocorrelation in the model, it is necessary to deal with it because autocorrelated residuals causes seriously inefficient estimates. Several methods are applied to test for residual autocorrelation. These are Wooldridge’s test for serial correlation from the regression of the firstdifferences variables, Durbin-Watson statistic and Baltagi and Wu’s (1999) locally best invariant test statistic. We reject the null hypothesis of non-autocorrelated errors. This result is expected because the only pair country is the U.S. and the shocks to the economy and the trade flows can be persistent.

lnVit = θ0 + θ1 lnSit + θ2 lnSU S−t + θ3 Rel-GDPpercapitait + θ4 ln(Tech-Similarityit )+ θ5 lnFDIit + γ1 lnDi + γ2 langi + γ3 bilaterali + γ4 borderi + αt + αi + uit

(7)

In Equation 7 the error structure becomes a 2-way model as we have both time and country fixed effects with AR1 error structure. 2-way model

: it = αt + αi + uit , with uit = ρui(t−1) + vit where v ∼ N (0, σ 2 )

Baier and Bergstrand (2007) discusses using first-differenced data as opposed to fixed effects. When the number of periods exceed two and when there is serially uncorrelated errors, fixedeffects estimator is more efficient. But, if the error term follows a random walk, first-differencing method is better. Our data has AR1 in some specification. While AR1 structure controls for autocorrelation, we also need to solve heteroskedasticity and contemporaneous correlation in the error term. Therefore, we follow Beck and Katz’s (1995) panel corrected standard errors instead of first-differenced. Prais-Winsten regressions with common AR1 are also estimated. Regarding common AR1 and panel-specific AR1, Beck and Katz (1995) argue that it is skeptical to expect common parameters for all independent variables but different adjustments parameter for the prior shocks. Small sample properties also lead us to focus more on common AR1 process instead of panel-specific AR1. Equation 7 is estimated for total trade, affiliated trade (U.S. parents and their affiliates) and unaffiliated trade separately. They are estimated using fixed effect and random effect with AR1 error structure. Additionally, Prais-Winsten regressions are employed in order to solve the heteroskedasticity and contemporaneous correlation in the error term.

17

5

Results

Our specification for trade in OPS incorporates the capacity to export services, relative cost, information-technology connectedness and the U.S. corporate presence as well as other time invariant variables, which are distance, language, bilateral trade agreement, border. Table 3 shows the results for total export and total import. Table 4 and Table 5 report the results of regression for affiliated (U.S. parents and their affiliates) and unaffiliated trade for export and import, respectively. In each table, first and sixth columns show the fixed effect regressions with AR1 errors. According to the p-values of Wooldridge’s test for serial correlation from the regression of the firstdifferences variables, we reject the null that there is no autocorrelated errors. The value of 1 for Durbin-Watson statistic as well as Baltagi and Wu’s (1999) locally best invariant test statistic21 show that there is a positive serial correlation. Therefore, AR1 correction is applied in each case. The columns after the fixed effect model show the results of random effect as well as panel corrected standard errors22 (which corrects for heteroskedasticity) with and without country dummies. At the bottom of the table, the estimated common ρ is also reported for each model. • Service capacity: ‘Services expenditure’ measures the capacity of country’s services. We hypothesize that as the domestic service sector grows, a country will likely export services because there is experience with how to sell services, more people engaged in the services sector, as well as possibly economies of scale and therefore lower prices. They also import more as their demand for service increase. – In Table 3, the significance of results is mixed based on the econometric model used but when it is significant, it is positive. The elasticities are higher for import side than for export side on average. As the service sector grows abroad, current account surplus of the U.S. in services might shrink. It should be noted that both Mann (2004) and Marquez (2005) find that the export elasticities are greater than import elasticities. The difference can be related to time period as well as countries in the data. While Mann (2004) has region dummies, Marquez (2005) looks at the elasticities in other service categories as well. They also use the income and relative price framework. – Table 4 and Table 5 show if the affiliated trade or unaffiliated trade is more affected by the service capacity. Again, the point estimates are not consistent but affiliated trade is affected more than unaffiliated trade on average. It seems that the higher import elasticities in fact is driven by affiliated parties in Table 3. • U.S.-Supply and U.S.-Demand: Due to the time dummies, we lose the U.S. time series variables if they are not country-specific. This country specific demand and supply variables are designed to consider how the intensity of trade with the U.S. in the goods market might influence the importance of services expenditure. We hypothesize that an increase in demand for services in the U.S. leads to the demand for services imports from countries with which the U.S. already has strong trade ties. The rationale is similar to the case where there is complementarity between goods trade and services trade. But simply due to familiarity, the infrastructure of marketing and branding and so on might support a stronger relationship 21 22

Values under 1.5 are generally considered biased by positive serial correlation STATA’s xtpcse is used.

18

in services trade with countries with deep relationships in goods trade. Similarly for export side, as the U.S. increases its service capacity, it is likely to export more to the countries with which the U.S. has strong export ties in the goods market. – U.S. Supply - capacity of service- is strongly significant for the export side. It is especially important for the unaffiliated trade that the driver of service export is not the affiliated parties. Considering that the U.S has a comparative advantage in services, it reveals itself in unaffiliated service trade. – U.S. Demand - demand for service- is not significant in any of our specification. As the U.S. has a comparative advantage in service sectors, the demand for service trade is not captured by the overall increase in the service expenditure of the U.S. or close ties in goods trade with another country. It is also possible that there is not enough variation in the service expenditure of the U.S. • Relative GDP per capita: According to the theoretical model, we use the relative GDP per capita to measure the relative cost. We hypothesize that as it goes up, the U.S. should export more and import less. On the other hand, it is positive and significant in most of the specification for both export and import. We suspect that it actually measures the relative factor endowments, similarity in tastes, or relative human capital. As the similarity in tastes and human capital converged, the trade among countries would go up. The better proxy for the cost would be the average wages in service sectors in each country. • Information technology connectedness: We hypothesize that if both the origin and the destination market have better IT-connectedness, that this supports more trade in services-both exports and imports. Surprisingly, we find a robust negative relationship between our measure of IT and OPS trade (Table 3).23 – One reason for the negative sign is that just because a country uses the internet does not mean that it uses the internet for business services. There is much literature on how countries with high connectivity nevertheless do not use the internet for commerce, but rather for game playing, just information, or chat. To the extent that this is the case, the negative sign might make sense. – On the other hand, positive sign for affiliated export in Table 4 shows that it is actually important for multinational firms and intra-firm trade. As hypothesized, “jointness in production” requires technological similarity. The delivery of services depends on the tasks performed in both countries. Though there is no positive or negative effect on the import side. – It is possible that the negative sign in unaffiliated for both export and import indicates the technological advancement and the ability of a country to produce its own services. • U.S. corporate presence: We suspect that U.S. FDI in the destination market and services trade is complementary. Previous research that examined FDI-goods and trade-goods found this complementarity, so we want to know if this carries over to FDI-total and trade-OPS. Second, we might suppose that FDI total and trade in services is complementary because trade in some goods is associated with trade in services, such as a sale of a machine includes 23

We tried different measure of IT but the result did not change.

19

a services contract. Finally, we are particularly interested if this variable is differentiated between affiliate and unaffiliated trade. – Indeed, we do observe a complementarity between FDI and services trade, both on export and import side, a bit higher responsiveness of OPS-imports and U.S. FDI in the source market. – Decomposing affiliate and unaffiliated trade reveals important differences in Table 4 and Table 5. FDI and affiliated trade is not significant for all econometric specifications. However, on the unaffiliated side, they are significant. The presence of FDI opens the destination market for more OPS export trade in general, rather than just within the MNC. This may come from agglomeration effects, or ‘experience’ effects (e.g. ‘get to know what the U.S. services are like’). This may also be the consequence of limitations on FDI establishment in destination markets, which the U.S. service provider overcomes through direct unaffiliated exports. That U.S. FDI in the import source market is significant suggests that U.S. FDI opens the U.S. market for competitors to the MNC’s own affiliates’ imports into the U.S. These results are consistent with workers who leave an affiliate, start their own firm, and take over some of the import trade. Or, that the MNC might have spun-off its affiliate, which then becomes unaffiliated U.S. OPS imports. • Distance, Language, Bilateral trade agreement and Common Border: Distance is not significant in any trade type but unaffiliated import. The importance of distance as a barrier to trade is fading away with the technology, especially for the service trade. On the other hand, language plays an important role as the transaction in services needs common communication tool. The results are highly significant for import side. Comparatively, the coefficients are much higher for affiliated than unaffiliated import. It supports the vertical integration theories and the stories about moving some back office operations to other countries such as India. Intra-firm trade takes place more when the host country has certain qualities, one of which is certainly a common language as it reduces the cost of communication. Bilateral trade agreement is significant for overall import but loses its significance when we look at affiliated and unaffiliated trade. Border is positive and significant for only unaffiliated export, which must be the trade with Canada and Mexico.

6

Conclusion

About 30 percent of OPS trade is between a US multinational parent and its affiliates abroad (intrafirm trade), about 60 percent is ‘arms-length’ trade. It is important to determine the driving forces of service trade in the economy as it becomes more tradable day by day with the advancement in technology. For the U.S. economy, service trade has been in surplus in current account as opposed to goods which has been in deficit. BEA provides OPS trade data for 31 countries, which is analyzed based on gravity model in this paper. We first review the challenges of data and possible estimation techniques and end up using fixed effect estimation with AR1 error structure as well panel corrected standard errors. We find that the depth of services sector affects export to the U.S.; total FDI and services trade are

20

complementary; technology is important for affiliated export. Language plays a key role for the import of services, especially for the affiliated import.

21

22 0.70

0.70 0.00 1.00 0.85

558 0.95 Yes No H 0.74

0.19∗∗∗ [0.04] −0.38∗∗∗ [0.08] 0.16∗∗∗ [0.03] −0.02 [0.05] 0.07 [0.04] −0.13∗∗ [0.07] 0.28∗ [0.15] −5.17∗∗∗ [0.96]

0.34∗∗∗ [0.04] 0.37∗∗∗ [0.05]

(4) ln(X)

0.82 0.00 1.00 0.56

527 0.32 Yes Yes

4.04∗∗∗ [1.53]

−0.08 [0.08] 1.16∗∗∗ [0.38] −0.02 [0.17] 0.12∗∗∗ [0.04]

0.22 [0.31]

(5) ln(M) 0.79∗∗∗ [0.27]

(7) ln(M) 0.72∗∗∗ [0.07]

(8) ln(M)

0.82

558 0.87 Yes No

558 0.93 Yes Yes H 0.70

558 0.87 Yes No H 0.87

0.00 −0.05 0.06 [0.06] [0.07] [0.05] 0.24∗∗∗ 0.75∗∗ 0.24∗∗∗ [0.09] [0.31] [0.08] −0.18∗ −0.23∗∗ −0.27∗∗ [0.10] [0.09] [0.12] 0.17∗∗∗ 0.15∗∗∗ 0.22∗∗∗ [0.04] [0.04] [0.04] −0.29 −0.29∗∗∗ [0.21] [0.08] 0.61∗∗∗ 0.61∗∗∗ [0.17] [0.08] 0.46∗∗ 0.37∗∗∗ [0.22] [0.12] −0.22 −0.30 [0.58] [0.22] −14.67∗∗∗ −13.49∗ −14.22∗∗∗ [2.70] [7.45] [1.38]

0.81∗∗∗ [0.09]

(6) ln(M)

Standard errors in brackets AR1(-P)(-H): AR1 autocorrelation(-panel specific)(-heteroskedastic and contemporaneously correlated) ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

558 0.97 Yes Yes H 0.55

558 0.92 Yes No

527 0.81 Yes Yes

Constant

Border

Bilateral

Language

ln(Distance)

ln(FDI-US)

ln(Tech-Sim)

Observations R-square Time Dummies Country Fixed Error-AR1 ρ Wooldridge(p-val) Durbin-Watson Baltagi-Wu LBI

0.13 [0.27] 0.20∗∗∗ [0.06]

0.04 0.18∗∗∗ 0.47 [0.30] [0.05] [0.31] −0.24∗∗ −0.30∗∗∗ −0.28∗∗∗ [0.10] [0.06] [0.08] 0.09∗∗∗ 0.11∗∗∗ 0.09∗∗∗ [0.03] [0.03] [0.03] −0.05 [0.11] 0.09 [0.09] −0.05 [0.12] 0.46 [0.32] −7.90∗∗∗ −6.45∗∗∗ 2.42 [1.88] [1.60] [7.10]

0.44∗∗∗ [0.05] 0.25∗∗∗ [0.05]

0.55∗∗ [0.23] 0.12∗∗ [0.06]

(3) ln(X)

ln(Rel GDPpercapita)

US-Demand

US-Supply

ln(Service Exp)

(2) ln(X)

(1) ln(X)

Table 3: Total OPS Export and Import

23

∗∗∗

(3) ∗

(4) ∗∗∗

(5) ∗∗

(6) ∗∗∗

(7)

ln(X-Unaff) (8)

0.71

0.71 0.00 1.00 0.80

558 0.83 Yes Yes H 0.66

558 0.67 Yes No H 0.79

0.69 0.00 1.00 0.85

527 0.81 Yes Yes

0.69

558 0.92 Yes No

558 0.95 Yes Yes H 0.54

558 0.93 Yes No H 0.73

Standard errors in brackets AR1(-P)(-H): AR1 autocorrelation(-panel specific)(-heteroskedastic and contemporaneously correlated) ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

558 0.78 Yes No

527 0.29 Yes Yes

Observations R-square Time Dummies Country Fixed Error-AR1 ρ Wooldridge(p-val) Durbin-Watson Baltagi-Wu LBI

0.51 0.96 0.36 0.51 0.44 0.13 0.39∗∗∗ [0.12] [0.54] [0.09] [0.25] [0.06] [0.28] [0.04] 0.21∗∗ 0.15 0.27∗∗∗ 0.12∗ 0.24∗∗∗ 0.19∗∗∗ 0.32∗∗∗ [0.10] [0.15] [0.09] [0.07] [0.05] [0.07] [0.05] 0.08 −0.52 0.08 0.16 0.11∗∗ 0.58∗ 0.08∗∗ [0.11] [0.64] [0.09] [0.33] [0.05] [0.30] [0.04] 0.35∗∗ 0.45∗∗∗ 0.21 −0.48∗∗∗ −0.39∗∗∗ −0.44∗∗∗ −0.35∗∗∗ [0.15] [0.13] [0.15] [0.11] [0.07] [0.08] [0.08] 0.24∗∗∗ 0.07 0.43∗∗∗ 0.10∗∗∗ 0.10∗∗∗ 0.11∗∗∗ 0.10∗∗∗ [0.06] [0.07] [0.09] [0.04] [0.03] [0.04] [0.03] −0.27 −0.16 0.01 0.01 [0.24] [0.12] [0.11] [0.05] 0.09 0.07 0.14 0.13∗∗∗ [0.20] [0.11] [0.09] [0.05] 0.46∗ 0.32 −0.09 −0.13∗∗ [0.26] [0.20] [0.13] [0.07] −0.43 −0.44 0.73∗∗ 0.58∗∗∗ [0.69] [0.37] [0.32] [0.15] −8.70∗∗ −20.07 −7.70∗∗∗ −7.28∗∗∗ −7.32∗∗∗ 1.90 −6.34∗∗∗ [3.44] [14.41] [2.48] [2.18] [1.62] [7.35] [1.02]

(2)

−29.78∗∗∗ [4.15]

1.34 [0.53] 0.05 [0.14] −0.91 [0.70] 0.72∗∗∗ [0.24] 0.04 [0.07]

∗∗

Constant

Border

Bilateral

Language

ln(Distance)

ln(FDI-US)

ln(Tech-Sim)

ln(Rel GDPpercapita)

US-Supply

ln(Service Exp)

(1)

ln(X-US Aff)

Table 4: OPS Export: Intra-firm (US parents and affiliates) vs Arm’s-length

24

∗∗∗

(3) ∗∗∗

(4) ∗∗∗

(5) ∗

(6) ∗∗∗

(7)

ln(M-Unaff) (8)

0.76

0.76 0.00 1.00 0.71

558 0.78 Yes Yes H 0.60

558 0.52 Yes No H 0.80

0.79 0.00 1.00 0.59

527 0.67 Yes Yes

0.79

558 0.84 Yes No

558 0.90 Yes Yes H 0.62

558 0.83 Yes No H 0.81

Standard errors in brackets AR1(-P)(-H): AR1 autocorrelation(-panel specific)(-heteroskedastic and contemporaneously correlated) ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

558 0.78 Yes No

527 0.53 Yes Yes

Observations R-square Time Dummies Country Fixed Error-AR1 ρ Wooldridge(p-val) Durbin-Watson Baltagi-Wu LBI

1.00 2.98 0.90 0.66 0.69 0.31 0.64∗∗∗ [0.17] [0.59] [0.15] [0.38] [0.09] [0.28] [0.06] 0.13 0.23 0.13 0.00 0.03 0.01 0.05 [0.14] [0.20] [0.13] [0.09] [0.07] [0.08] [0.05] 0.13 −0.64 0.10 0.38 0.21∗∗ 0.72∗∗ 0.22∗∗∗ [0.17] [0.83] [0.20] [0.47] [0.09] [0.34] [0.07] 0.50∗∗ 0.18 0.46 −0.54∗∗∗ −0.39∗∗∗ −0.39∗∗∗ −0.48∗∗∗ [0.23] [0.33] [0.35] [0.20] [0.11] [0.10] [0.11] 0.26∗∗∗ 0.22∗ 0.34∗∗∗ 0.13∗∗ 0.17∗∗∗ 0.14∗∗∗ 0.22∗∗∗ [0.09] [0.13] [0.09] [0.05] [0.04] [0.05] [0.04] −0.19 −0.18 −0.29 −0.26∗∗ [0.35] [0.15] [0.20] [0.11] 1.12∗∗∗ 1.11∗∗∗ 0.56∗∗∗ 0.55∗∗∗ [0.28] [0.20] [0.16] [0.09] 0.58 0.45∗ 0.31 0.26∗∗ [0.39] [0.25] [0.22] [0.10] −1.06 −0.99∗ 0.12 0.12 [0.99] [0.53] [0.57] [0.22] −26.20∗∗∗ −82.50∗∗∗ −24.38∗∗∗ −12.01∗∗∗ −12.85∗∗∗ −2.84 −12.78∗∗∗ [4.74] [16.61] [2.52] [2.23] [2.70] [7.78] [1.81]

(2)

−45.77∗∗∗ [5.46]

1.94 [0.80] 0.04 [0.20] 0.15 [0.99] 0.19 [0.39] 0.05 [0.11]

∗∗

Constant

Border

Bilateral

Language

ln(Distance)

ln(FDI-US)

ln(Tech-Sim)

ln(Rel GDPpercapita)

US-Demand

ln(Service Exp)

(1)

ln(M-US Aff)

Table 5: OPS Import: Intra-firm (US parents and affiliates) vs Arm’s-length

References Adler, J. H. (1945). United States Import Demand During the Interwar Period, The American Economic Review 35(3): 418–30. Adler, J. H. (1946). The Postwar Demand for United States Exports, The Review of Economic Statistics 28(1): 23–33. Amiti, M. and Wei, S.-J. (2004). Fear of Service Outsourcing: Is It Justified?, Working Paper WP/04/186, IMF. Anderson, J. E. (1979). A Theoretical Foundation of the Gravity Equation, The American Economic Review 69(1): 106–116. Anderson, J. E. and van Wincoop, E. (2003). Gravity with Gravitas: A Solution to the Border Puzzle, The American Economic Review 93(1): 170–192. Baier, S. L. and Bergstrand, J. H. (2002). On the Endogeneity of International Trade Flows and Free Trade Agreements, Technical report, American Economic Association Annual Meeting. Baier, S. L. and Bergstrand, J. H. (2007). Do Free Trade Agreements Actually Increase Members’ International Trade?, Journal of International Economics 71(1): 72–95. Baldwin, R. and Harrigan, J. (2011). Zeros, Quality and Space: Trade Theory and Trade Evidence, American Economic Journal: Microeconomics 3(2): 60–88. Baltagi, B. H. and Wu, P. X. (1999). Unequally Spaced Panel Data Regressions with AR(1) Disturbances, Econometric Theory 15(06): 814–823. Beck, N. and Katz, J. N. (1995). What to do (and not to do) with Time-Series Cross-Section Data, The American Political Science Review 89(03): 634–647. Borga, M. and Zeile, W. J. (2004). International Fragmentation of Production and the Intrafirm Trade of U.S. Multinational Companies, BEA Working Paper WP2004(02). Brainard, L. S. (1997). An Empirical Assessment of the Proximity-Concentration Trade-off between Multinational Sales and Trade, The American Economic Review 87(4): 520–544. Carr`ere, C. (2006). Revisiting the Effects of Regional Trade Agreements on Trade Flows with Proper Specification of the Gravity Model, European Economic Review 50(2): 223–247. Chang, T. (1945-46). International Comparison of Demand for Imports, The Review of Economic Studies 13(2): 53–67. Chang, T. C. (1948). A Statistical Note on World Demand for Exports, The Review of Economics and Statistics 30(2): 106–116. Feenstra, R. C., Markusen, J. R. and Rose, A. K. (1998). Understanding the Home Market Effect and the Gravity Equation: The Role of Differentiating Goods, NBER Working Paper 6804, The National Bureau of Economic Research.

25

Filippini, C. and Molini, V. (2003). The Determinants of East Asian Trade Flows: A Gravity Equation Approach, Journal of Asian Economics 14(5): 695–711. Francois, J. and Hoekman, B. (2010). Services Trade and Policy, Journal of Economic Literature 48(3): 642–92. Freund, C. and Weinhold, D. (2002). The Internet and International Trade in Services, American Economic Review 92(2): 236–240. Green, W. (2011). Fixed Effects Vector Decomposition: A Magical Solution to the Problem of Time-Invariant Variables in Fixed Effects Models?, Political Analysis 19(2): 135–146. Gr¨ unfeld, L. A. and Moxnes, A. (2003). The Intangible Globalization: Explaining the Patterns of International Trade in Services, NUPI Working Paper 657, the Norwegian Institute of International Affairs. Hanson, G. H., Mataloni, R. J. and Slaughter, M. J. (2001). Expansion Strategies of U.S. Multinational Firms, NBER Working Paper 8433, The National Bureau of Economic Research. Hanson, G. H., Mataloni, R. J. and Slaughter, M. J. (2005). Vertical Production Networks in Multinational Firms, The Review of Economics and Statistics 87(4): 664–678. Hausman, J. A. and Taylor, W. E. (1981). Panel Data and Unobservable Individual Effects, Econometrica 49(6): 1377–1398. Head, K., Mayer, T. and Reis, J. (2009). How Remote is the Offshoring Threat?, European Economic Review 53(4): 429–444. Heckman, J. (1979). Sample Selection Bias as a Specification Error, Econometrica 47(1): 153–161. Helpman, E., Melitz, M. J. and Rubinstein, Y. (2007). Estimating Trade Flows: Trading Partners and Trading Volumes, NBER Working Paper W12927, The National Bureau of Economic Research. Hooper, P., Johnson, K. and Marquez, J. (2000). Trade Elasticities for the G-7 Countries, Princeton studies in international economics, Princeton University. Kepaptsoglou, K., Karlaftis, M. G. and Tsamboulas, D. (2010). The Gravity Model Specification for Modeling International Trade Flows and Free Trade Agreement Effects: A 10-year Review of Empirical Studies, The Open Economics Journal 3: 1–13. Kimura, F. and Lee, H.-H. (2006). The Gravity Equation in International Trade in Services, Review of World Economics 142(1): 92–121. Lennon, C. (2009). Trade in Services: Cross-Border Trade vs. Commercial Presence. Evidence of Complementarity, wiiw Working Papers 59, The Vienna Institute for International Economic Studies. Lennon, C., Mirza, D. and Nicoletti, G. (2009). Complementarity of Inputs across Countries in Services Trade, Annals of Economics and Statistics 93/94: 183–206.

26

Mann, C. L. (2004). The US Current Account, New Economy Services, and Implications for Sustainability, Review of International Economics 12(2): 262–76. Marquez, J. (2005). Estimating Elasticities for U.S. Trade in Services, International Finance Discussion Paper No. 836 . Board of Governors of the Federal Reserve System - International Financial Transactions Section. Martinez-Zarzoso, I. and Nowak-Lehmann, F. (2003). Augmented Gravity Model: An Empirical Application to Mercosur-European Union Trade Flows, Journal of Applied Economics VI(2): 291–316. Mirza, D. and Nicoletti, G. (2004). What is So Special about Trade in Services?, Research Paper 2004/02, University of Nottingham. Pain, N. and van Welsum, D. (2004). International Production Relocation and Exports of Services, OECD Economic Studies 38(1). Plumper, T. and Troeger, V. E. (2007). Efficient Estimation of Time-Invariant and Rarely Changing Variables in Finite Sample Panel Analyses with Unit Fixed Effects, Political Analysis 15(2): 124–139. Plumper, T. and Troeger, V. E. (2011). Fixed-Effects Vector Decomposition: Properties, Reliability, and Instruments, Political Analysis 19(2): 147–164. P¨oyh¨onen, P. (1963). A Tentative Model for the Volume of Trade between Countries, Weltwirtschaftliches Archiv 90(1): 93–100. Silva, J. S. and Tenreyro, S. (2006). The Log of Gravity, The Review of Economics and Statistics 88(4): 641–658. Tinbergen, J. (1962). Shaping the World Economy: Suggestions for an International Economic Policy. UN, European Commission, IMF, OECD, UNCTAD and WTO (2002). The Manual and Statistics of International Trade in Services. van Welsum, D. (2004). In Search of ‘Offshoring’: Evidence from U.S. Imports of Services, Working Papers in Economics & Finance BWPEF 0402, Birkbeck, University of London. Walsh, K. (2006). Trade in Services: Does Gravity Hold? A Gravity Model Approach to Estimating Barriers to Services Trade, The Institute for International Integration Studies Discussion Paper Series iiisdp183, IIIS. Yeaple, S. R. (2003). The Role of Skill Endowments in the Structure of U.S. Outward Foreign Direct Investment, Review of Economics and Statistics 85(3): 726–734.

7 7.1

Appendix Appendix: Figures 27

Figure 10: Trade Balance by Service Category

Figure 11: Services Supplied to Foreign Markets

2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996 1995 1994 1993 1992

−500000

0

500000

Millions Service Supplied by US affiliates Export affiliated

Export Export US affiliated

Notes: Beginning in 2004, Service Supplied by US affiliates series has a major revision so there is discontinuity in the time series data comparison Source: BEA

28

Figure 12: Services Supplied to the U.S. Market

2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996 1995 1994 1993 1992

−500000

0

500000

Millions Service Supplied by foreign affiliates Import affiliated

Import Import US affiliated

Notes: Beginning in 2004, Service Supplied by foreign affiliates series has a major revision, so there is discontinuity in the time series data comparison Source: BEA

29

7.2 7.2.1

Appendix: Theoretical Background Demand side

The representative consumer of country f , f ∈ 1, . . . , F , maximizes the Spence-Dixit-Stiglitz subutility function subject to his/her budget constraint.

Uf =

X Nd D X

σ−1 σ



σ σ−1

αdf cvdf

(8)

d=1 v=1

where; cvdf : demand of representative consumer in country f for variety v from country d Nd : number of variety in country d αdf : proxy for home bias or brand image of the good of country d σ : elasticity of substitution between different available varieties, (σ > 1) To simplify, let’s assume that Pvdf = Pdf ∀v, so cvdf = cdf ∀v

max Uf = c

wrt

X D

σ−1 σ



σ σ−1

Nd αdf cvdf

d=1 D X

Nd Pdf cdf = Ef

d=1

where Ef : total expenditure devoted to consuming services. The first-order conditions give the demand equations of country f from each country d.   Pdf αkf −σ cdf = ckf Pkf αdf Multiplying each side by Nd Pdf and summing over the number of countries D lead to the bilateral export expression of country d to country f: σ xdf = Ef Nd αdf

Pdf 1−σ Pf

where,

Pf =

X D

1−σ σ Pdf Nd αdf

d=1

30



1 1−σ

(9)

7.2.2

Supply side

O-ring production function (Kremer, 1993) is considered. Assume a production of a variety of a traded service that consists of different tasks, which interact with each other. For simplicity, all varieties are treated equally which allows to remove the subscript v from the following relations. A group of tasks are undertaken in the domestic country d, and another group in the foreign country f. Each task k in a country h (h ∈ d, f ) employs a number Lk,d and Lk,f of domestic and foreign inputs respectively. The performance q of these inputs ranges from 0 to 1. If the task is perfectly performed (q=1) then the quality of the task and the contribution to the production process are maximal. If the task is not performed (q=0), then the chain of production stops and the output is 0. Within any assigned task, inputs are perfect substitutes but remain complementary between tasks, because tasks themselves are complementary. The supply of a service that is traded abroad can then be expressed by:

Ydf =

Kd Y

K Y

qk,d Lk,d

k=1

qk,f Lk,f

(10)

k=Kd +1

Every task k is associated with a quality qk,h , ∀h ∈ d, f of the performance that might differ from all other tasks within and across countries. If foreign inputs do not participate in the production process, then qk,f = 0, and the transaction between the two countries is not realised (Y = 0). Each firm minimizes total cost under the constraint of this production function and given labour cost and quality.

min T Cdf = L

Kd X

wk,d Lk,d +

K X

wk,f Lk,f

(11)

k=Kd +1

k=1

wrt (3) Thus we can represent total cost function in terms of quantity of one task level (say i) labor demand and wage: T Cdf = KLi,d wi,d (12) We can also write an equation for Li,d from production function (3). By plugging it into total cost fuction, a dual cost expression of the service production function can be derived.

T Cdf =

1/K KYdf

   Kd  K Y qk,d 1/K Y qk,f 1/K wk,d wk,f

k=1

(13)

k=Kd +1

At the monopolistic competition equilibrium, price is equal to marginal cost with mark-up of the firm realized on the foreign market (µdf ) and transaction cost from trading (τdf ). 1−K

Pd,f = µdf Ydf K

   Kd  K Y qk,d 1/K Y qk,f 1/K τdf wk,d wk,f

k=1

k=Kd +1

31

(14)

Q 1/K Plugging Pd,f into (2) and denoting by y¯h = k yk,h the geometric mean of any variable x relative to a country d or f, the expression of trade in services becomes: K−1 (σ−1) K

xd,f = Ef Nd Ydf

  q¯d 1−σ q¯f 1−σ 1 1−σ 1−σ σ 1−σ µdf αdf τdf w ¯d w ¯f Pf

32

(15)

Appendix: Country List Table 6: Country List Region America

Country Argentina Brazil Canada Chile Mexico Venezuela Europe Belgium-Luxembourg France Germany Italy Netherland Norway Spain Sweden Switzerland United Kingdom Asia China Hong Kong India Indonesia Japan Korea, Republic of Malaysia Philippines Singapore Thailand Middle East Israel Saudi Arabia Africa Pacific South Africa Australia New Zealand Notes: Bermuda and Taiwan are not on the list due to missing data in other variables.

33