7104 2018

June 2018

The Hardships of Long Dis-tance Relationships: Time Zone Proximity and Knowledge Transmission within Multinational Firms Dany Bahar

Impressum:  

CESifo Working Papers ISSN 2364‐1428 (electronic version) Publisher and distributor: Munich Society for the Promotion of Economic Research ‐ CESifo GmbH The international platform of Ludwigs‐Maximilians University’s Center for Economic Studies and the ifo Institute Poschingerstr. 5, 81679 Munich, Germany Telephone +49 (0)89 2180‐2740, Telefax +49 (0)89 2180‐17845, email office@cesifo.de Editors: Clemens Fuest, Oliver Falck, Jasmin Gröschl www.cesifo‐group.org/wp   An electronic version of the paper may be downloaded  ∙ from the SSRN website:           www.SSRN.com ∙ from the RePEc website:          www.RePEc.org ∙ from the CESifo website:         www.CESifo‐group.org/wp    

 

CESifo Working Paper No. 7104 Category 8: Trade Policy

The Hardships of Long Distance Relationships: Time Zone Proximity and Knowledge Transmission

Within Multinational Firms

Abstract Using a unique dataset on worldwide multinational corporations with precise location of headquarters and affiliates, I present evidence of a trade-off between distance to the headquarters and the knowledge intensity of the foreign subsidiary’s economic activity, emerging from dynamics related to the proximity-concentration hypothesis. This trade-off is strongly diminished the higher the overlap in working hours between the headquarters and its foreign subsidiary. In order to rule out biases arising from confounding factors, I implement a regression discontinuity framework to show that the economic activity of a foreign subsidiary located just across the time zone line that increases the overlap in working hours with its headquarters is, on average, about one percent higher in the knowledge intensity scale. I find no evidence of the knowledge intensity and distance trade-off weakening when a non-stop flight exists between the headquarters and the foreign subsidiary. The findings suggest that lower barriers to real-time communication within the multinational corporation play an important role in the location strategies of multinational corporations.

JEL-Codes: F230, L220, L250.

Keywords: multinational firms, multinational corporations, knowledge, location, proximity concentration hypothesis, FDI.

Dany Bahar

The Brookings Institution Harvard Center for International Development Harvard University / Cambridge / MA / USA

db21@post.harvard.edu http://www.danybahar.com

May 30, 2018 Previously circulated under the title “Heavier than air? Knowledge Diffusion within the Multinational Firm”. This paper benefited from helpful comments at various stages from Martin Abel, Laura Alfaro, Pol Antras, Sam Asher, Sebastian Bustos, Marcus Casey, Matias Cattaneo, Michele Coscia, Ricardo Hausmann, Elhanan Helpman, Bill Kerr, Marc Melitz, Paul Novosad, Nathan Nunn, Lant Pritchett, Krista Rasmussen, Brina Seidel, Ran Shorrer, Ernesto Stein, Stephen Yeaple, Muhammed Yildirim, Richard Zeckhauser, Oren Ziv and the participants of the XIX Dynamics in Economic Growth and International Trade conference, as well as attendants in different seminars at Harvard University, the NBER, the Inter-American Development Bank, Georgetown University and LACEA. Jean-Paul Leidenz Font and Samuel Rosenow provided excellent research assistance. I acknowledge financial support from Harvard Economics Department. All errors are my own. Download newest version: https://www.dropbox.com/s/06cdlvoh8bi2kv0/BAHAR_KnowledgeMNC.pdf?dl=0

1 IntroductionAbout fifty percent of cross-country income variation is explained by differencesin productivity.1 This begs the question: if productivity-inducing knowledgeis available in some places, why isn’t it easily transferable to others? Arrow(1969) suggests that the transmission of knowledge is difficult and costly. Thesedifficulties arise because effective knowledge transmission involves human inter-action, which cannot be fully replaced with written words, neither be embeddedin goods that can be shipped at low costs.2 A firm, as any other economicagent, faces difficulties when transferring knowledge among different divisionsand affiliates, as has been extensively explored in the literature (e.g., Oldenski,2012; Keller and Yeaple, 2013; Giroud, 2012). The ability of a multinationalfirm to transfer knowledge to its foreign affiliates, however, should depend onseveral dimensions that characterize the locations of the firm’s headquarters andits subsidiaries. This paper explores the role of time zones in reducing the costsof knowledge transmission within multinational corporations (MNCs).

This paper’s contribution to the literature is threefold. First, using a highlydetailed establishment-level worldwide dataset on MNCs I complement the ex-isting empirical evidence suggesting that costs of knowledge transmission facedby MNCs play a role in the proximity-concentration hypothesis, in particularby creating a trade-off between the level of knowledge intensity of a foreign sub-sidiary’s economic activity and the geographic distance between such subsidiaryand its global headquarters. Second, I show that this “knowledge and distancetrade-off” significantly weakens, and even disappears, the more overlap in work-ing hours there is between a foreign subsidiary and its global headquarters.Third, to rule out the role of other unobservables explaining my results, I usea regression discontinuity designed to exploit discrete spatial variation in timezones and show that foreign subsidiaries located at roughly the same distancefrom their headquarters are active in economic activities that significantly differin knowledge intensity depending on whether they are located in a time zonecloser to the headquarters or not.

The trade-off between distance and knowledge intensity cannot be explainedusing earlier frameworks that looked at the proximity-concentration hypothesisand the fragmentation of MNCs which, implicitly or explicitly, assumed zeromarginal cost or costs orthogonal to distance of transferring knowledge betweenheadquarters and subsidiaries (i.e., Helpman, 1984; Markusen, 1984; Brainard,1993; Markusen et al., 1996; Markusen, 1997; Carr et al., 2001; Helpman et al.,2004; Markusen and Maskus, 2002).3 This trade-off, however, emerges in anymodel that incorporates the idea of marginal costs of transferring knowledgeincreasing in geographic distance. For example, Ramondo and Rodriguez-Clare

1e.g., Hall and Jones (1999); Caselli (2005)2Knowledge that resides in human minds is usually referred to as tacit (Polanyi, 1966).

Tacit knowledge is information that cannot be easily explained, embedded or written down.3A number of empirical studies have tested the validity of these models’ predictions, but

there has been little or no emphasis on testing the assumption that knowledge transmissionis costless.

2

(2013) study the gains from openness in a comprehensive model that includesboth multinational production and trade, and assume that all multinational pro-duction activity entails iceberg type efficiency losses that vary across countrypairs. This marginal cost, they assume, is an increasing function of geographicaland cultural distance. Ramondo (2014) models multinational production whereforeign subsidiaries face a combination of a variable and a fixed cost when relyingin technology or knowledge from their headquarters, capturing the idea of lim-ited span of control for the headquarters, thus generating a loss of productivityat the affiliate level. Arkolakis et al. (2013) model an economy in which coun-tries specialize either in innovation or in production. When a country specializesin innovation it implies that its firms open production subsidiaries abroad witha marginal cost that affects their productivity. These marginal costs are meantto capture various impediments that multinationals face when operating in adifferent economic, legal or social environment, as well as the various costs oftechnology transfer incurred by multinationals in different production locations.Tintelnot (2014) presents a model of multinational production in which foreignplants engage in variable costs that are determined, in part, by distance be-tween the headquarters and the location of the foreign subsidiary. Keller andYeaple (2013) provide an explanation to why the marginal cost of knowledgetransmission increases with distance by modeling such cost as shipping costs forintermediate goods embedding headquarters services.4 In their model, shippingof intermediate goods is more prevalent for subsidiaries active in knowledge in-tensive activities. My study builds on the research by Keller and Yeaple (2013),in further exploring the “distance and knowledge intensity” trade-off emergingfrom their setting studying the proximity-concentration hypothesis. In partic-ular, a contribution of this study is the finding that this “trade-off” cannot befully explained by trade costs of intermediate goods within the MNC, but alsoby a better ability of firms to communicate in real time, as measured by a head-quarters and its foreign subsidiary having more overlap in working hours (basedin their geographic location).

An interpretation of this finding is that not all knowledge can be fully em-bedded in intermediate goods. Examples of the type of knowledge that getstransferred from the headquarters to its foreign subsidiary are in the form ofmanagement, monitoring, coordination, troubleshooting, etc. which is a formof tacit knowledge: it lives on people brains and cannot be fully written downor embedded in a machine. This type of knowledge requires human interactionof some sort for its transmission, as suggested by Arrow (1969). Thus, the lossin efficiency that distant foreign affiliates face as evidenced in the literature islikely also related to the difficulties in transferring tacit knowledge (Polanyi,1962). In fact, the consensus in the existing literature on the economics ofknowledge is that the transmission of knowledge is not immediate, and thatknowledge diffusion strongly decays with distance. For instance, the paper byJaffe et al. (1993) was among the first to make this claim, showing that patent

4See Irarrazabal et al. (2013) for a similar setting which does not include the knowledgedimension.

3

citations are more frequent within the same geographic area. Bottazzi and Peri(2003) followed up using European data. Along the same lines, Keller (2002)showed that knowledge spillovers decrease with distance by looking at produc-tivity changes as explained by foreign R&D investment. He documents that thehalf-life of such spillovers is 1200Km. More recently, Bahar et al. (2014) showthat a country is 65% more likely to add a new product to its export basketwhenever a geographic neighbor is a successful exporter of the same good, afinding that is attributed to the local character of knowledge diffusion. Ba-har and Rapoport (2018), consistently with the idea that the transmission oftacit knowledge requires human interaction,show that migrants are an impor-tant driver of knowledge diffusion across nations.5 In the context of MNCs, ithas also been shown that more complicated tasks require more time and effortfor coordination and monitoring, and this becomes much more difficult at longerdistances (e.g., Gumpert, 2015).

The empirical exercise in this paper is based on a sample of about 55000domestic and 25000 foreign horizontal subsidiaries belonging to over 2000 MNCsfrom the Worldbase dataset by Dun & Bradstreet.6 For the most part, I focusmy attention on MNCs active in the manufacturing sector that have horizontallyexpanded into foreign countries. For each one of the foreign subsidiaries in thesample I have information on their physical location and primary economicactivity, as defined by the 1987 Standard Industry Classification (SIC).

After computing precise distances between each foreign affiliate and its MNCglobal headquarters following a geocoding process using Google Maps, I docu-ment a negative partial correlation between knowledge intensity and the distancebetween a headquarters and its foreign subsidiaries (after including a number ofcontrols that would account for other possible explanations). For example, ev-erything else equals, the estimation in the paper implies that an American MNCheadquartered in Houston in Texas would locate a meat packing subsidiary –aneconomic activity with low knowledge intensity levels– in Kabul in Afghanistan(approximately 12000Km away from the headquarters) and a semiconductorplant –an economic activity with high levels of knowledge intensity– in Ireland(at approximately 7000Km of distance from the headquarters). A novel sourceof variation in this empirical exercise is a industry-level knowledge intensitymeasure that I construct based worker-level characteristics in each industry, asopposed traditional measures that rely on balance-sheet data. The measuresaim to capture the tacit knowledge intensity of an economic activity by averag-ing the accumulated experience and training of the workforce of such industry,using occupational characteristics defined in the O*NET project dataset.7

5See Keller (2004) for a review of this literature.6The dataset was privately acquired from D&B and is not publicly accessible. It has been

previously used in the literature by Lipsey (1978), and more recently by Black and Strahan(2002); Harrison et al. (2004); Acemoglu et al. (2009); Alfaro and Charlton (2009); Alfaro andChen (2012) and Alfaro et al. (2015).

7Previous studies have also used the O*NET database to construct industry level mea-sures. For example, Oldenski (2012) measures the importance of communication with theheadquarters and importance of the communication with the customer, for each industry.Costinot et al. (2011) uses O*NET to create an industry level measure of task routine-ness

4

I then use a number of measures that ease the transmission of knowledge byenhancing communication and monitoring between a MNC’s headquarters andits foreign subsidiaries. First, the existence of a non-stop flight between airportslocated within 100Km of both the headquarters and the foreign subsidiary. Sec-ond, the number of overlapping working hours in a business day between thelocations of both the headquarters and the foreign subsidiary. I then show bothdescriptively and using a regression discontinuity design that the “knowledge-intensity and distance” trade-off is weakened when there is a larger overlap inworking hours between the foreign subsidiary and the headquarters. Thus, thecost of shipping intermediate goods, which would be just as relevant withinthe same time zone (because north-south shipping is equally as expensive aseast-west shipping), is not enough to explain the fact that MNCs tend to locatetheir foreign subsidiaries active in knowledge intensive geographically nearby,as suggested by Keller and Yeaple (2013).

Overall, the results in this paper suggests that tacit knowledge plays isan important determinant of the concentration-proximity hypothesis, and inparticular, the knowledge and distance trade-off that emerges from it. Thesefindings have larger implications for a number of yet-unanswered questions ineconomics. For instance, high barriers to knowledge transmission may explainpersistent differences in productivity levels between countries and the divergenceof their incomes over time (e.g., Pritchett 1997, Hall and Jones 1999), becauseproductivity-inducing knowledge does not diffuse easily.

The rest of the paper is divided as follows. Section 2 describes the datasetand the construction of relevant variables. Section 3 presents descriptive evi-dence documenting, among other results, the distance and knowledge intensitytrade-off that arises from the proximity-concentration hypothesis. Section 4explores how the variables measuring the ease of communication between a for-eign subsidiary and its headquarters (i.e., existence of a non-stop flight and theoverlap in working hours) affects the aforementioned trade-off. Section 4 alsoimplements a regression discontinuity design to estimate whether knowledgeintensity significantly differs with changes in time zones that facilitate commu-nication with the headquarters. Section 5 concludes and addresses areas forfuture research regarding the role of tacit knowledge in economic activity.

2 Data and Definitions of Variables

2.1 Worldbase dataset by Dun & BradstreetI use the Worldbase dataset by Dun & Bradstreet (acquired in May 2012) asthe main data source for the empirical exercise. The dataset has informationon more than one hundred million establishments worldwide with data fromyear 2012. Each establishment is uniquely identified and linked to its global

for 77 sectors. Keller and Yeaple (2013) also present results making use of knowledge inten-sity variables constructed with O*NET in their Appendix. Autor et al. (2003) use O*NETpredecessor, DOT, to construct measures for routine and non-routine tasks.

5

headquarters (referred to as the “global ultimate”). For this study I focus onforeign plants engaged in manufacturing industries (SIC codes 2000 to 3999)owned by MNCs. As suggested by Caves (1971), an MNC is “an enterprise thatcontrols and manages production establishments – plants – located in at leasttwo countries.”8

The sample obtained from the dataset includes about 55 thousand domesticand 25 thousand foreign horizontal subsidiaries of MNCs scattered active in themanufacturing sector in over 100 countries, which report to over 2000 MNCs.9For the analysis, I will use the reported main SIC code as the only indicatorof a plant’s economic activity. There are about 450 unique SIC 4-digit codes(in manufacturing) reported by subsidiaries as their main economic activity inthe dataset. The sample I use is significantly smaller than the overall Dun &Bradstreet dataset (which originally has over 124 million establishments) forseveral reasons. First, I only include subsidiaries of multinational corporations:that in itself reduces the sample to be about 9 percent of the original dataset.Second, I only use manufacturing subsidiaries, which is a much smaller share theuniverse of subsidiaries. I limit the sample to subsidiaries in the manufacturingsector, which significantly reduces the sample further, also to about 9 percentof the subsidiaries that are part of a MNC (as known, most of the MNCs are inthe service sector, as shown too by Alfaro and Charlton (2009)). Further, thesample has all the domestic subsidiaries but the foreign ones are limited to onlythose that can be classified as a foreign horizontal expansion (as I explain innext subsection). Also, from Alfaro and Charlton (2009) we know that this is asmall share of all subsidiaries. Hence, we end up with a relatively small samplegiven the initial size of the dataset.

In order to obtain the precise location of each plant I geocode the datasetusing Google Maps Geocoding API to find the exact latitude and longitude of itsheadquarters and each one of its foreign subsidiaries. With this I computed theexact distance between each headquarters and its foreign subsidiaries. Figure1 maps the unique locations of all foreign subsidiaries (dots) and headquarters(triangles) in the sample.

[Figure 1 about here.]

For instance, Figure 2 shows the headquarters and subsidiaries of an Amer-ican car manufacturing multinational firm. The firm, headquartered in the US,has a number of foreign subsidiaries on different continents. The lines originat-ing from the headquarters represent the geographic distance to each subsidiary.

[Figure 2 about here.]

Online Appendix Section B discusses this dataset more in detail.8I exclude MNCs for which 99% of their subsidiaries or employees are in the home country,

besides them having plants in two or more countries. This drops a small number of ChineseMNCs with one or two subsidiaries in Hong Kong and the rest in China.

9I performed on the dataset an algorithm that would group multinationals not only basedon their assigned number, but also on their names when the differences are small (e.g., SonyCorporation vs. Sony Corp).

6

2.2 Main Variables Definitions2.2.1 Horizontal Foreign Subsidiary

For the plant-level dataset I define a foreign subsidiary as a horizontal expansionbased on its SIC code vis-à-vis all the SIC codes reported by the firm, in allof its domestic subsidiaries in the home country. This resolves the data issuesthat arise when the economic activity of the headquarters does not necessar-ily represent the main business of the firm. For instance, in the dataset, theheadquarters of a well known worldwide multinational in the cosmetic world isdefined under SIC code 6719 (“holding company”). However, many of its do-mestic subsidiaries are classified under SIC code 2844 (perfumes, cosmetics, andother toiletries), which would be a more natural classification for the firm as awhole. Hence, by limiting the definitions to the global ultimate’s SIC categoryonly, horizontal relationships would be underestimated.

In the sample, 32% of subsidiaries are foreign horizontal. Similarly to Alfaroand Charlton (2009), I find that some horizontal foreign affiliates also classifyas vertical (both downstream and upstream).10 I keep these observations in thesample, and allow in every empirical specification for a different constant forthese types of subsidiaries by adding the proper dummy variables.

2.2.2 Knowledge Intensity Measures

In order to estimate the knowledge intensity of industries I create a new mea-sure that aims to capture the tacit knowledge intensity for each industry. Themeasures use data from the Occupational Employment Statistics (OES) fromthe Bureau of Labor Statistics,11 and occupational profiles compiled by theOccupational Information Network (O*NET) project.12 OES breaks down thecomposition of occupations for each industry code,13 based on a list of about 800occupations. These occupations can be linked to occupational profiles generatedby O*NET, which includes results from a large number of survey questions onthe characteristics of each occupation.

The relevant questions in the survey that capture the learning component ofthe workers, as mentioned above, are the ones related to experience and training.The exact form of the questions from O*NET are:

10I follow Alfaro and Charlton (2009) methodology of using the US input-output tables todefine vertical relations. Online Appendix Section B.3 expands on this discussion.

11Data from 2011, downloadable from ftp://ftp.bls.gov/pub/special.requests/oes/oesm11in4.zip12O*NET is the successor of the US Department of Labor’s Dictionary of Occu-

pational Titles (DOT). I use the O*NET database version 17, downloadable fromhttp://www.onetcenter.org/download/database?d=db_17_0.zip. Costinot et al. (2011) alsouse O*NET to create an industry level measure of task routineness for 77 sectors. Keller andYeaple (2013) also present results making use of knowledge intensity variables constructedwith O*NET in the Appendix.

13I used Pierce and Schott (2012) concordance tables to convert industrycodes from NAICS to 1987 SIC. The concordance table is downloadable fromhttp://faculty.som.yale.edu/peterschott/files/research/ data/appendix_files_20111004.zip.

7

• How much related experience (in months) would be required to be hiredto perform this job?

• How much “on-site” or “in-plant” training (in months) would be requiredto be hired to perform this job?

• How much “on-the-job” training (in months) would be required to be hiredto perform this job?

Using these questions I generate the main knowledge intensity measure that Iwill be using in the empirical analysis section. The measure, which I refer to itas “Experience plus training” throughout the paper, is constructed by measuringthe (wage-weighted) average months of experience plus on-site and on-the-jobtraining required to work in each industry. In particular, for each sector sknowledge intensity is defined as:

KIi =∑o

ωo,icumexpo

Where cumexpo is the sum of experience and training associated to occupa-tion o, ωi,o is the weight of occupation o in industry i, measured by either shareof employment or wage. Naturally, we have that

∑o ωo,i = 1 for every i. While

I use within-sector wage share to define ωo,i, the results are robust to using thewithin-sector employment share instead.

Using this measures, manufacturing industries ranking highly are computerrelated (SIC 3573, 3571 and 3572), communications equipment (SIC 3669, 3663and 3661) and electronics and semiconductors (SIC 3672, 3674 and 3676). Sincethese measures are based on US data only, I will assume the US ranking in theknowledge intensity of industries proxies that of the rest of the world.

Online Appendix Section C provides a discussion on these measures andcompares them to other commonly used measures in the literature that proxyfor knowledge intensity at the firm level, such as R&D share of expenditures.The O*NET based measures overcome important shortcomings associated withother prevalent measures used in the literature. Their distribution is smoother,they do not rely on a sampling of firms, and they use the same standardizedinputs for all industries. Hence, I use these indicators as the main proxies forknowledge intensity throughout the paper.

2.2.3 Unit shipping costs

Unit shipping costs for SIC manufacturing industries are computed using datafrom Bernard et al. (2006).14 This industry-level measure aims to proxy theunit shipping cost variable (referred to as t in the theoretical framework inOnline Appendix Section A), which accounts for how costly it is to transportone unit of that good irrespective of industry. For instance, goods with the

14Downloadable from http://faculty.som.yale.edu/peterschott/files/research/data/xm_sic87_72_105_20120424.zip

8

highest unit shipping costs in the dataset include ready-mixed concrete and ice,which require special forms of transportation.

The variable measures the amount of US dollars required to transport 1$worth of a good per every 100Km. It is computed by averaging the same measureper industry across all countries exporting to the US in year 2005. To deal withlong tails, this variable will be used in a logarithmic scale in all the differentempirical specifications.

Unit shipping cost figures correlate negatively with the knowledge intensitymeasures, with a correlation coefficient of about -0.6.

2.2.4 Country levels controls

In some specifications I use country level controls such as income per capitafrom the World Bank’s World Development Indicators. In addition, I also ofteninclude data on conventionally measured factors of production (stock of physicalcapital, human capital and land) from UNCTAD (Shirotori et al., 2010).

2.2.5 Ease of communication proxies

In order to proxy for the ease of communication between a subsidiary and itsheadquarters, I use two variables: non-stop flights and working hours overlap.

The first variable is used because the existence of non-stop flights wouldproxy for the ease of managers and workers to do more frequent business trips,given the convenience of a direct flight. Business trips, by allowing face-to-face interaction, would facilitate the transmission of tacit knowledge. However,it is important to note that business trips, even if convenient, happen muchless often than phone calls due to the elevated costs associated with them. Inorder to compute the existence of a non-stop air route between a headquartersand its subsidiary, I matched all the existing airports within a 100Km radius(conditional on being in the same country), using the geocoded latitude andlongitude. The data for airports (with their respective coordinates) and activeair routes come from the OAG Flights15 Through this matching I create adummy variable which takes the value of 1 if there is a non-stop flight betweenthe headquarters and its subsidiary.

The second variable, overlap in working hours, aims to capture the “real-time” communication ability between managers and workers in the two plants.Being in the same time zone allows workers to use phone or videoconferencecommunication more frequently (substituting partially for means such as fax oremail). This allows for better transmission of tacit knowledge, which is valuablefor troubleshooting or crisis solving. In order to compute the overlap in workinghours I use the geocoded longitude of each subsidiary to find its time zone, andcompare it to that of its headquarters. Assuming that working hours run from8:00am to 6:00pm (10 hours in total), the variable measures, for a single day, thenumber of hours that overlap in the working schedule of both the headquartersand its subsidiary.

15The dataset was privately acquired in January 2015.

9

3 Descriptive evidenceIn this section I use the global character of the dataset and the "tacit" knowledgeintensity measures to estimate two results already established (either empiricallyor theoretically) in the literature based on the existence of marginal costs totransfer knowledge within the MNC (e.g., costs of monitoring, communicating,traveling, etc.). First, that a MNC is less likely to horizontally expand into aforeign destination the more knowledge intensive is the economic activity underconsideration (as shown by Keller and Yeaple (2013) using American MNCs).Second, that conditional on horizontal expansion, there is a trade-off betweenthe knowledge intensity level of the foreign subsidiaries’ economic activity andtheir geographical distance to the headquarters, consistently with what recentframeworks show (e.g., Ramondo and Rodriguez-Clare, 2013; Arkolakis et al.,2013; Keller and Yeaple, 2013; Ramondo, 2014; Tintelnot, 2014).

Table 1 provides some descriptive details about the used sample, in terms ofthe distribution of foreign affiliates across regions of the world and developingvs. developed countries. This sample includes domestic subsidiaries and foreignsubsidiaries that replicate production abroad (i.e. an horizontal expansion). Inparticular, a foreign subsidiary is included when it reports being active in anindustry (a SIC 4 digit code) for which the MNC has one domestic subsidiaryalso active in.16 In total, the sample includes 84,881 subsidiaries that are owedby 2030 MNCs. About 32% of these subsidiaries are foreign. The table alsoincludes the knowledge intensity variable measured in standard deviations fromthe mean (denoted by KI), averaged over domestic and over foreign subsidiaries.The last column presents the difference, with stars denoting the correspondentp-value level.

[Table 1 about here.]

On average, industries of the foreign subsidiaries are less knowledge-intensivethan the industries manufactured by their domestic counterparts. The samepattern holds for all presented cuts of the data, besides for few firms based onnon-OECD countries, though the difference is not statistically significant. InWestern Europe, however, it seems like the pattern is inverted. Beyond thepurely descriptive statistics shown in 1 I use plant-level data to estimate thefollowing specification, in an effort to rule out possible confounding factors:

Foreignh,s = βk · log(ks) + βt · log(ts) + controlss + ϕh + eh,s (1)

Where s indexes a subsidiary and h its headquarters. The independent vari-able is a dummy which takes the value 1 if the subsidiary is a foreign horizontalaffiliate of the firm and 0 if it is a domestic one. ks is a measure of knowledgeintensity of the economic activity (i.e., the manufactured good or product) ofthe foreign subsidiary. ts is the unit shipping cost for the good manufacturedin the foreign subsidiary. controlss is a vector of variables that control for the

16See Appendix Section B.3 for further explanation on this definition

10

size of the market and factor endowments of the country hosting the foreignsubsidiary which aims to controls for aggregate demand and cost of producing(understanding the limitation of this approach makes this exercise suggestive,rather than definitive).17 I also include terms to control for whether such foreignhorizontal subsidiary also classifies as vertical, either through a downstream orupstream linkage to one of the industries produced domestically by the MNC,allowing for a different constant term for those particular cases.18 eh,s is the er-ror term. In particular, the idea of this test is to find out whether MNC are lesslikely to expand internationally their knowledge intensive activities, as a proxyto for worse performance relative to exports (with the proper caveats discussedabove).

Using plant-level data provides an additional degree of identification. Thatis, by adding ϕh, which represents MNC fixed effects, we can control, for ex-ample, for the overall productivity level of the MNC. Thus, the results arewithin-MNC, implying that the partial correlation estimated in βk is indepen-dent of the aggregate productivity level of the MNCs, but only on the economicactivity the affiliate is engaged in. It also implies that the comparison is betweendifferent subsidiaries active in more or less knowledge intensity activities, withinthe same MNC. At this point it is worth mentioning that subsidiaries withina single MNC might differ in their economic activity, thus allowing for within-firm variation in the right hand side variables of the empirical specification (seeFigure B4 in the Appendix).

Some specifications will also include the term ηs which represents a hostcountry fixed effect (i.e. the country where the affiliate is located);19 and inseparate specifications, the term ηs× θP which represents a fixed effect for hostcountry and 2-digit SIC code (denoted by P ) combination. The ηs × θP set offixed effects will be included to control for issues such as intellectual propertyregulation, aggregate costs and aggregate demand to the extent they vary atthe 2-digit industry level. When both ϕh and ηs are used in the estimation, Iexclude the controls defined at the level of the countries of both the headquartersand foreign subsidiary, given multicollinearity.

Before presenting the results it is important to discuss the relevance andinterpretation of this particular estimation. In a sense, this estimation aims toestimate whether the level of knowledge intensity of an economic activity playsa role in the proximity-concentration hypothesis. In other words, if knowledgeintensive activities are less likely to be expanded horizontally, then this is sug-gestive evidence that costs associated with the transmission of knowledge are an

17These variables include measures of factor endowments between in the subsidiary countriesas well as income per capita.

18As explained in Section 2.2.1, about 32% of those subsidiaries classified as horizontalwould also clarify as vertical according to the Alfaro and Charlton (2009) "5-cents" rulein the input-output relationship. See Appendix Section B.3 for further explanation on thisdefinition. By adding these variables as dummies, we allow for a different constant for theseparticular subsidiaries.

19I slightly abuse from notation when using ηs, which refers to a fixed effect for the hostcountry of the subsidiary, and not for the subsidiary s. I believe this is less confusing thanusing a new index for the country of the subsidiary.

11

element driving such trade-off, consistently with others studies –some of themmentioned above– which find a similar result in other contexts. Yet, we alsomust acknowledge the limitations of this estimation: first and foremost, the ab-sence of exports by each MNC-product combination to particular destinations.Without this, it is very difficult to say something precise about the proximity-concentration trade-off, because it is impossible to know whether the lack ofhorizontal expansion is happening in a place where the same product is beingexported to. Thus, given the limitations, I defer to this descriptive exercisewhich aims to shed light on whether there is a differential pattern in terms ofhorizontal foreign expansion with respect to knowledge intensity.20 The resultsare presented in Table 2. The table uses the experience plus training measurediscussed above as a proxy for k.

[Table 2 about here.]

Column 1 presents the preferred specification, which controls for trade unitcosts and all the country pair variables. The results suggest that, everything elseequal, industries for which their knowledge intensity is 10% higher are about 1.6percentage points less likely to be replicated abroad. For instance, according tothis estimation, semiconductors (SIC 3674), which is characterized by havingworkers with an average of over 80 months of required experience plus training,is about 12 percentage points -or 21%- less likely to be replicated abroad than ameat packing plant (SIC 2011), which its workers have, on average, 37 monthsof experience plus training.

The endowment controls suggest that international expansion is more likelyin countries that are richer but smaller. It also suggests that it is less likely incountries that are capital abundant and that have high levels of human capital,two variables that typically correlate with higher wages. The estimates for up-stream and downstream linkage industries are positive and significant, thoughbesides allowing for a different constant, their interpretation is not straightfor-ward. Finally, it is important to mention that the point estimate for the tradecost variable is negative, though not statistically significant. However, we wouldhave expected a positive estimate according to the proximity-concentration hy-pothesis. Yet, it is important to notice that this unexpected result is probablydue to the inclusion of multinational fixed effects, given that most of the vari-ation of the trade cost variable is across MNCs and not within. Thus, theinclusion of MNCs fixed effects comes at the price of losing the ability to ex-ploit the variation of this variable. Alternative standardizations of the tradecost variable often provide positive point estimates, though still statisticallyinsignificant. I discuss this particular result in detail later on.

Consistent with the idea that the transmission of knowledge plays a role inthe proximity-concentration hypothesis, the estimator for βk is negative and ro-bust across all specifications. The inclusion of host country fixed effects denoted

20An alternative approach would be to “fill” with zeros for each country-product combinationwhere there is no horizontal expansion, but this implies assuming that there is (equal) demandeverywhere, which is a very strong assumption.

12

by ηs in Column 3 rules out other potential channels that could be driving theresults. For instance, poor intellectual property regulation in different coun-tries.21 Column 4 includes a more strict control: parent industry (2-digit),denoted by P , interacted with host country fixed effects, ηs × θP , to allow fordifferential policies at the country level for different types of industries. Notethat in columns 2 and 3 we exclude the country level controls due to high mul-ticollinearity. Overall, the estimate of βk is robust to the inclusion of theseadditional controls in terms of its magnitude, negative sign and its statisticalsignificance.22

The second implication of the existence of marginal costs of transferringknowledge within the MNC are that foreign affiliates locate in closer geographicproximity to the headquarters the more intensive in knowledge their main eco-nomic activity is (assuming these costs are increasing in distance). To answerthis question I restrict the dataset to only existing foreign subsidiaries, leavingabout 25 thousand subsidiaries in the sample (i.e. I exclude observations whichcorrespond to a domestic subsidiary of the MNC). Table 3 presents summarystatistics for this foreign subsidiaries, including the geographic distance in be-tween the headquarters and the subsidiary itself, the knowledge intensive andunit shipping cost values as well as other control variables. For instance, the av-erage distance between a headquarters and its foreign subsidiary in the sampleis about 8.2 in logs (in levels, it corresponds to about 5600 Km). The table alsopresents the average knowledge intensity for economic activity of the foreignsubsidiary, as well as the corresponding unit shipping cost (which distributesbetween 0 and 1, and thus its logarithmic transformation is always negative).As explained above, some of the foreign subsidiaries that classify as horizontal,also classify as vertical (both downstream and/or upstream), and thus in allthe specifications I allow for a different constant term by adding dummies forthese foreign subsidiaries. In particular 29% (15.3%) of the horizontal foreignsubsidiaries are in an economic activity that could be upstream (downstream)to the economic activities of the domestic subsidiaries. There is an overlap be-tween the two, so in total, about 30% of the foreign subsidiaries would classifyas horizontal and vertical (either upstream or downstream). The lower panelinclude summary statistics for variables that measure the ease of communica-tion between the headquarters and the foreign subsidiary, which will be used inthe next subsection. For instance, about 33% of the headquarters and foreignsubsidiary pairs have a non-stop route between their nearby airports.23 Thetable also shows that, on average, there are almost 7 working overlapping hoursbetween a headquarters and its subsidiary.

21Results are robust to excluding China from the sample, alleviatng possible concerns givenbiases its inclusion might generate in the estimations due lack of enforcement with respect tointellectual property rights.

22This exercise is similar to the one by Keller and Yeaple (2013), though not quite, as itdoes not control for local demand in the foreign location. In Online Appendix Section D,however, I estimate the same relationship as in Keller and Yeaple (2013) using aggregateddata on exports and sales of foreign affiliates of US companies by industry, making use of theknowledge intensity measures I constructed.

23Using 2012 data, the same year the Dun and Bradstreet data is representative of.

13

[Table 3 about here.]

Figure 3 show graphic results suggesting the existence of a trade-off betweendistance to the headquarters and knowledge intensity of the economic activ-ity of the foreign subsidiary. It plots the estimation of βk using the followingspecification:

log(dh,s) = βk · log(ks) + βt · log(ts) + ϕh + eh,s (2)

Where s indexes a subsidiary and h its headquarters. The left hand sidevariable, log(dh,s) is the logarithmic transformation of the geographic distancebetween the location of the headquarters and that of the foreign subsidiary.The right hand side includes the log of knowledge intensity associated to theeconomic activity of the subsidiary (denoted by ks), the log unit shipping cost(ts), as well MNC fixed effects. We would expect βk < 0: foreign subsidiarieswill tend to be closer the more knowledge intensive they are (given that thetransmission of knowledge is increasing in distance). Note that the inclusion ofMNC fixed effects, imply that the exploited variation within firm.

The results are indicative of the trade-off between distance (d) and knowl-edge intensity (k). In fact, according to the estimation, for every 10% increase inthe knowledge intensity scale, a foreign subsidiary is about 5% closer in distanceto the headquarters.24 In practice, the results suggest that, roughly, an Ameri-can MNC headquartered in the Houston, Texas would locate its low knowledgeintensive activity, say a meat packing subsidiary, in Kabul, Afghanistan (ap-proximately, 12000Km), while a high knowledge intensive activity, for example,a semiconductor plant, would be located in a place that is about 40% closer, so,for instance, in Ireland (approximately 7000km distance), ceteris paribus.

[Figure 3 about here.]

To avoid any misunderstanding coming from using the term “knowledge anddistance trade-off”, an explanation is in place. The term refers to the findingthat –conditional on foreign expansion– MNCs tend to locate their foreign sub-sidiaries active in knowledge intensive activities closer to the headquarters thanthose active in less knowledge intensive activities. This result is an extension ofthe traditional proximity-concentration hypothesis in the presence of marginalcosts of transferring knowledge within the MNC that increase in distance. Thus,in a way, the result is could be interpreted as part of the proximity-concentrationhypothesis.

24Allowing for a more flexible fit, such as a quadratic one, suggests that the estimatedquadratic relationship does not seem to be monotonically decreasing for the lower values ofknowledge intensity (although a flat or even negative slope in that area cannot be rejectedin the data either). However, and perhaps more importantly, for higher levels of knowledgeintensity there is a clear negative relationship with distance. This result is qualitatively im-portant, given that it would be consistent with the idea that distance appears to matter muchmore for higher levels of knowledge intensity. Intuitively, this means that after certain levelof knowledge intensity, the more sophisticated products are the closer the foreign subsidiarieswill be located to the headquarters. Results of this exercise are available upon request.

14

4 The distance-knowledge trade-off and the easeof communication

With this evidence in hand, it might be difficult to attribute the documentedtrade-off between k and d to the somewhat abstract idea of marginal costs oftransferring knowledge. The empirical results might be driven by factors otherthan knowledge not accounted for, in the presence of omitted variable bias.A conventional explanation in the literature would be that knowledge intensivesectors are associated with higher intra-firm trade of intermediate goods, makingit less profitable to locate those plants in far away locations (Keller and Yeaple,2013; Irarrazabal et al., 2013). Keller and Yeaple (2013), in particular, assumein their model that knowledge can be fully embedded in intermediate goods,that are in turn shipped to remote locations.

However, this assumption might not be proper for all type of knowledge.Tacit knowledge cannot be, by definition, embedded in intermediate goods. Forinstance, transmission of tacit knowledge within the firm could involve costs ofmonitoring by the management in the headquarters or also real-time problemsolving efforts by the chief engineering team dealing with the foreign team incertain manufacturing processes. Distance would play an important role in thismatter: business travel, for instance, or the ability to work in teams in realtime (through phone, video conferences, etc.) would be critical for industrieswhere monitoring and work across borders is prevalent. Thus, it could well bethat it is the cost of transmitting this type of knowledge which partly drives thedocumented relationship, and not the shipping of intermediate goods.

I perform a test that aims to disentangle between both explanations. If thecost of transferring knowledge is indeed an increasing function of distance -asargued- and thus, a determinant in the location decisions of firms, then easiercommunication between headquarters and subsidiaries would work as a cost-reducing mechanism for the purpose of transmitting knowledge. This would behard to explain with the intra-firm trade mechanism, given that arguably theease of communication is orthogonal to the transportation costs of intermediategoods.

I test for this hypothesis by re-estimating specification (2), this time includ-ing variables that proxy for the ease of communication within the firm, ofteninteracting them with the knowledge intensity covariate. These variables, allmeasured for each subsidiary and its headquarters, are (1) the existence of acommercial non-stop air route (between airports close to the headquarters andthe foreign subsidiary, measured as being within 100Km and within the countryborders), and (2) the number of overlapping working hours in a business day(see Section 2.2.5 above for more details on the construction of these variables).

The purpose of utilizing these variables is to proxy for forms of commu-nication that allow for the transmission of knowledge, though they are quitedifferent between themselves. As explained above, business travel provides theopportunity to work face-to-face, though it occurs with less frequency, given thehigh costs of traveling. In this context, Giroud (2012) has shown that the exis-

15

tence of commercial air routes between subsidiaries and headquarters positivelyaffects the profitability of the former. Being in the same time zone allows forconvenient real-time, day-to-day, communication, significantly reducing wait-ing time between the two ends for problem solving or consulting about specifictasks. Stein and Daude (2007), for instance, find that time zone is an importantdeterminant of aggregate FDI flows, which they attribute to better monitoring.

The results are presented in Table 4. All columns use the experience plustraining indicator to proxy for k. Columns 1 and 2 present result using the ex-istence of a non-stop flight between the headquarters and the foreign subsidiary,while Columns 3 and 4 present results using the overlap in working hours be-tween the foreign subsidiary and its global headquarters (the value goes from 0to 10).

[Table 4 about here.]

First, across all specifications, the estimates for βk are negative and statis-tically significant, signaling that the trade-off between distance and knowledgeintensity remains there, on average.25 Column 1 shows that neither the dummyvariable measuring the existence of a non-stop flight in between the headquar-ters and its foreign subsidiary, nor its interaction with the knowledge intensitymeasure are statistically different from zero. In fact, the estimate of βk is notvery different than the one estimated and plotted in Figure 3. Column 2 repeatsthe exercise including a fixed effect for the country of the foreign subsidiary, andthus the variables controlling for factor endowments in those countries are ex-cluded. The results, however, are robust to this inclusion, yet they are harderto interpret because of the variation left on the dependent variable after bothfixed effects being added simultaneously.26

Columns 3 and 4 show a different result when focusing on overlap in workinghours as a measure of ease in communication between the headquarters andthe foreign subsidiary. In fact, there the estimate for βk becomes significantlylarger maintaining its negative sign, implying that when there are no overlapin working hours, the distance and knowledge intensity trade-off becomes muchmore pronounced. In addition, and perhaps more importantly, the interaction

25All specifications control for the trade unit costs, and the estimator for βt is not signifi-cantly different from zero. While we would have expected a positive point estimate, the resultsare not surprising as we only exploit the within MNC variation. The other controls implythat distance from headquarters to foreign affiliate correlates positively with both populationand human capital levels and negatively with capital per worker levels of the foreign affiliate’shost country. It is not straightforward to interpret these numbers beyond their role as con-trol, but one possible interpretation in this context is that higher level of human capital in theaffiliate’s country correlates with better local management, and thus the cost of transferringknowledge is reduced, allowing for more distant subsidiaries. Larger population and lowerlevels of capital per workers is correlated with lower wages, too, which might play a role inlocating subsidiaries in more distant locations. The estimate for upstream and downstreamlinkages dummy variables do not statistically differ from zero, and the results are robust totheir exclusion.

26In part this is the reason I don’t report results that include the ηs × θP fixed effects as inprevious tables. The results, however, are robust to the inclusion of this additional control.

16

term appears to be positive and statistically significant, implying that the trade-off weakens the more overlap in working hours. It is also worth noting thatacross all specifications the ease of communication coefficients are estimated tobe negative, and this is simply because the longer the distance between thesetwo countries, these variables tend to be smaller (i.e. less working hours overlap,less non-stop flights).

These findings are insightful on their own. The results suggest that being inthe same time zone seems to facilitate the transmission of knowledge. The abil-ity of managers in the headquarters to communicate with colleagues in foreignlocations, for troubleshooting or consulting on an open-ended range of issues,is more efficient when communication happens in real time, without long wait-ing times. This might be even more relevant for transmitting tacit knowledge,given that complicated problems would require real-time interaction, and notjust explanations being sent through fax or email. Furthermore, this logic couldalso serve as an example for arguing that the barriers of transmitting knowl-edge is increasing with distance: managers and workers in the headquartersmight require working extra hours to communicate with their peers in foreignsubsidiaries, incurring additional compensation and operational costs.

Figure 4 presents a graphical representation of the distance and knowledgeintensity trade-off for different the different possible values that define the over-lap in working hours between the foreign subsidiary and its headquarters. Theleft panel plots the estimated slope of such trade-off as a function of overlapin working hours. When the overlap is 0, the slope of the trade-off is highlynegative. As the overlap becomes larger, the trade-off weakens and even dis-appears. The right panel, plots the trade-off for different values of overlap inworking hours (normalizing the log(k) to be zero at its smallest value of the dis-tribution, for comparison purposes), showing that the trade-off is even slightlypositive when there are 10 hours of overlap between the two locations. Sim-ply put, this result implies that the cost of shipping intermediate goods, whichwould be just as relevant within the same time zone (because north-south ship-ping is equally as expensive as east-west shipping) is not enough to explain thefact that MNCs tend to locate their foreign subsidiaries active in knowledgeintensive geographically nearby.

[Figure 4 about here.]

A regression discontinuity design to study the role of timezonesIn order to rule out biases arising from confounding factors, I implement aregression discontinuity framework exploiting the fact that time zones vary dis-cretely with distance. By doing so, I effectively compare the knowledge intensityof foreign subsidiaries that belong to the same MNC but that are on differentsides of a time zone line.27 To the best of my knowledge, using a regression

27I refrain from presenting results using a fuzzy regression discontinuity design using theexistence of a non-stop flight based on the work by Campante and Yanagizawa-Drott (2016),

17

discontinuity design exploiting time zone differences is novel in the context ofinternational economics, and in particular, when studying patterns of behaviorof multinational corporations.

As a first step it is important to define the running variable centered aroundeach time zone line: it corresponds to the distance (in kilometers) from everyforeign subsidiary to the nearest time zone line. A MNC will often have foreignsubsidiaries in located in time zones that are both eastwards and westwardsrelative to the headquarters location. Our "treatment" in this setting is beingin a time zone that allows for a higher overlap in working hours between theheadquarters and the subsidiary. Figure 5 clarifies how the running variable isdefined and normalized using an hypothetical example. For instance, foreignsubsidiaries 1 and 2 are located east of their headquarters’ location, while sub-sidiaries 3 and 4 are located west of it. The running variable is based on thedistance from each one of these subsidiaries to the nearest time zone line. Inthat sense, because foreign subsidiary 1 is located on the side of the line clos-est to the headquarters (e.g., higher overlap in working hours), its distance ismarked as positive. On the other hand, since foreign subsidiary 2 is located onother side of the line, then the running variable in that area is negative. Sameconcept applies to the running variable for foreign subsidiaries 3 and 4.

[Figure 5 about here.]

Each observation in the sample corresponds to a foreign subsidiary, and thusthe "treatment" can be defined as being in a closer time zone to its headquar-ters. Each headquarters, of course, has several foreign subsidiaries that are nearseveral distinct time zone lines. Thus, in this setting it is very important tomake sure that the regression discontinuity design it is indeed comparing for-eign subsidiaries that belong to the same MNC and at the same time, that arenear the same time zone line to one side or the other. Thus, I estimate thefollowing specification:

log(kh,s) = βTZcloserTZh,s,tz + distTZlineh,s,tz+ (3)closerTZh,s,tz × distTZlineh,s,tz + ϕh + τtz + eh,s

Where h is a subscript for each MNC, s represents a subsidiary and tzrepresents a time zone line which is the nearest one to subsidiary s. log(kh,s) isthe knowledge intensity of the economic activity of the foreign subsidiary. Thevariable is a dummy which is defined as closerTZh,s,tz = 1{distTZlineh,s,tz >

who show that the likelihood of having a non-stop flight is experiences a jump at the 6000miles (or 9600 kilometers) of distance. The reason not to present those results are twofold.First, the exercise shown in Table 4 finds no results when including a non-stop flight as an easeof communication variable. Second, while the reduced form (using the 6000 mile threshold)shows results consistent with what expected (e.g., observations just below the threshold aremore knowledge intensive than those just above it), the results are not robust when using thenon-stop flight in a fuzzy regression discontinuity setting. This is probably because the sampledoes not include all the existing flights, but rather only those flying to and from airports neara headquarters or foreign subsidiaries. Results of this, however, are available upon request.

18

0}. distTZlineh,s,tz is the running variable, defined as distance from subsidiarys to its closest time zone line tz following the guidelines explained before andrepresented in Figure 5. The interaction closerTZh,s,tz×distTZlineh,s,tz allowsmore flexibility when estimating the slopes before and after the cutoff (e.g.,the time zone line). ϕh and τtz are MNC and time zone line fixed effects,respectively. By adding these fixed effects I make sure that we compare foreignsubsidiaries belonging to the same MNC and that are across the same time zoneline.

Figure 6 graphically represents the estimation of Specification (3) using thepackage rdplot (Calonico et al., 2014, 2017). It shows that, indeed, that foreignsubsidiaries located just across the time zone line that puts them in a closer timezone to their headquarters (above zero in the horizontal axis which measuresvalues of the running variable) are, on average, are active in economic activitieshigher in the knowledge intensity scale (measured in the vertical axis), as com-pared to those who are on the other side of the time zone (below zero in thehorizontal axis). This graph uses observations for which their distance to thetime zone line corresponds to the optimal running variable bandwidth whichis 299.157 kilometers, using the methodology by Cattaneo et al. (2018) whobuilds on the work by Imbens and Kalyanaraman (2012). Note that the verticalaxis has values below and above zero because for this plot I use the residual oflog(kh,s) after controlling for ϕh and τtz.

[Figure 6 about here.]

The actual estimation of Specification (3) using data on foreign subsidiariesare presented in Table 5. Each column in the table presents results using adifferent bandwidth definition which range from 150 to 350 kilometers as wellas the optimal bandwidth in the last column. The estimation uses a triangularweight scheme, giving higher weight to observations closer to the cutoff point.

[Table 5 about here.]

The results suggest that a foreign subsidiary just across (any) time zone linecloser to the headquarters –thus increasing the overlap in working hours– is in aneconomic activity that is, on average, 0.6 to 0.84 percent higher in the knowledgeintensity scale, depending on the bandwidth used. Note that the estimator forβTZ is statistically significant (the standard errors are clustered at the MNCand time zone line level). At first this might look like a very small number. Yet,keep in mind that the treatment itself is also quite small: this exercise comparesforeign subsidiaries very close to each other (certainly measured in terms relativeof distance to their headquarters), some of them (the treated) have an extrahour of overlap with the work day of their colleagues at the headquarters, orsometimes even less than that (some time zones change in increments of halfan hour). Also, this is an average treatment effects for all possible differenttreatments (e.g., increasing the overlap in working hours from 2 to 3, or from6 to 7, for example) and as such is a weighted average of different treatmenteffects which could be larger or smaller, depending on the relative improvement.

19

Yet, at most we will always be limited by a “small” treatment. Thus, the resultsare qualitatively important taking into consideration that the only thing whichdiffers between subsidiaries located at either side of the time zone line is havingone more hour of overlap with their colleagues working at the headquarterslocation, even if their external validity cannot be proven beyond the estimationof presented in Table 4, with all the limitations such approach has.

However, a question remains: is it overlap in working hours the only thingthat varies substantially when comparing subsidiaries at either time of a timezone line? The evidence suggest that, in the context of what is relevant for thisexercise, it is. To show that I re-estimated Specification (3) replacing the de-pendent variable for other observable measures considered determinants of thelocation decision of foreign subsidiaries (used as controls in several specificationsabove). These are the trade unit cost and measures of factor endowments in thecountry where the subsidiary is located (income per capita, capital per workerand human capital). Since some of the horizontal foreign subsidiaries in thesample could also be classified as being upstream or downstream to the portfo-lio of economic activities produced at home by the MNC, I also test for those.The results can be found in Table 6 where each column uses a different depen-dent variable in the context of the regression discontinuity (using the optimalbandwidth and a triangular weighting scheme). As can be seen, the treatmenteffect is statistically not different from zero across all of those variables, reduc-ing remaining concerns of other confounding factors: time zone changes cannotexplain differential patterns in terms of trade costs (Column 1), nor whetherthe horizontal foreign subsidiary is being misclassified (Columns 2 and 3), andneither in the factor endowments of the country where the subsidiary is located(Columns 4, 5 and 6).

[Table 6 about here.]

5 Concluding RemarksIt is straightforward that, in order to operate smoothly, MNCs make investmentsso that their workers located both in the headquarters and in foreign subsidiariescan work together to maintain a smooth business operation. "Working together"could imply many things, such as managing, coordinating, monitoring, trou-bleshooting, etc. One broad way to denominate all these activities is under theumbrella of knowledge transmission. Knowledge transmission can certainly becostly and more intense for certain industries. These costs, of course, are animportant aspect determining location decisions of MNCs. This reasoning is,of course, the main explanation behind the existing trade-off between distanceto the headquarters and the knowledge intensity level of a foreign subsidiary, aresult consistent with the work of many others (e.g., Ramondo and Rodriguez-Clare, 2013; Ramondo, 2014; Arkolakis et al., 2013; Tintelnot, 2014; Keller andYeaple, 2013).

The main contribution of this paper, however, is to show that this trade-offsignificantly weakens when time zones are taking into consideration: knowledge

20

intensity is, on average, higher for roughly equidistant foreign subsidiaries thatare closer to the headquarters in terms of in time zones. This result is at oddswith suggesting that larger intra-firm trade costs (between the headquartersand its knowledge intensive subsidiaries) can fully explain the aforementionedtrade-off (e.g., Keller and Yeaple, 2013). Some other distance-varying costs mustbe playing a role in explaining this trade-off. In particular costs that can bereduced by engaging in real time communication between different locations. Apossible interpretation is the intra firm transmission of tacit knowledge.

This paper also contributes to the literature by adapting a regression discon-tinuity design that exploits multiple spatial differences across time zone lines inthe context of researching the behavior of MNCs. To the best of my knowledge,this is the first time this methodology has been used in this context.

Nonetheless, this paper has left open some other specific questions that willshed light on our general understanding of how the transmission of knowledgeplays a role in the location decisions and the performance of MNC. For instance,what other tools and means are at a firm’s disposal to enhance the processthrough which it transfers knowledge to its subsidiaries and workers?28 Howdoes the knowledge intensity of the good relate to the existence of regionalhubs, as opposed to different plants serving every foreign market? Is the costof knowledge transmission a relevant determinant for service provider firms, asit is for manufacturing firms? Given the difference in the nature of services vs.manufacturing industries in terms of their tradability, we can expect differentpatterns in the data.

Naturally, this research agenda also contains questions that have relevantpolicy implications. While governments intend to develop their private sectorsby attracting foreign investment, designing an effective policy should answerquestions such as: is there enough infrastructure in place to allow effectivecommunication for foreign firms? Should the focus be on specific types of firmsand specific industries for which knowledge transmission will be easier? Doall types of products have the potential to generate productivity spillovers todomestic firms, or only those for which the cost of knowledge transmission islow?

All in all, despite the fact that productivity outweighs factor accumulationin growth accounting exercises (e.g., Hall and Jones, 1999; Caselli, 2005), theprocess through which knowledge is accumulated by economic agents is stillan under-researched area. However, a better understanding of this process iscritical to answering open questions in economics. The difficulties associatedwith transferring and acquiring knowledge, which translates into productivityshifts, are not unique to MNCs. They can also relate to domestic firms (e.g.,Bloom et al., 2012; Kalnins and Lafontaine, 2013), investors (e.g., Coval andMoskowitz, 2001), innovation (e.g., Kerr, 2008) and even countries’ export bas-kets diversification (e.g., Bahar et al., 2014; Bahar and Rapoport, 2018). At alarger scale, the documented evidence reinforces the importance of knowledge

28Using German data Gumpert (2015) shows that MNCs active in knowledge-intensiveactivities "distribute the knowledge" by hiring more capable workers to foreign subsidiaries.

21

transmission in overall economic activity. Thus, understanding the ways knowl-edge affects economic activity lies at the core of important and unansweredquestions on convergence, development and growth. Knowledge and its diffu-sion, after all, are significant phenomena that can alter global economic patternsin as-of-yet unexplored ways.

ReferencesAcemoglu, Daron, Simon Johnson, and Todd Mitton. “Determinants of VerticalIntegration : Financial Development and Contracting Costs.” The Journal ofFinance LXIV, 3: (2009) 1251–1290.

Alfaro, Laura, Pol Antras, Davin Chor, and Paola Conconi. “InternalizingGlobal Value Chains: A Firm-level Analysis.” NBER Working paper 41,December: (2015) 1–5.

Alfaro, Laura, and Andrew Charlton. “Intra-Industry Foreign Direct Invest-ment.” American Economic Review 99, 5: (2009) 2096–2119.

Alfaro, Laura, and Maggie Xiaoyang Chen. “Surviving the Global FinancialCrisis: Foreign Ownership and Establishment Performance.” American Eco-nomic Journal: Economic Policy 4, 3: (2012) 30–55.

Arkolakis, Costas, Natalia Ramondo, Andrés Rodríguez-Clare, and StephenYeaple. “Innovation and Production in the Global Economy.” Technical Re-port 18792, National Bureau of Economic Research, Cambridge, MA, 2013.

Arrow, Kenneth J. “Classificatory Notes on the Production and Transmissionof Technologcal Knowledge.” The American Economic Review 59, 2: (1969)29–35.

Autor, David H, Frank Levy, and Richard J Murnane. “The Skill Content ofRecent Technological Change: An Empirical Exploration.” The QuarterlyJournal of Economics 118, 4: (2003) 1279–1333.

Bahar, Dany. “Measuring knowledge intensity in manufacturing industries: anew approach.” Applied Economics Letters .

Bahar, Dany, Ricardo Hausmann, and Cesar A. Hidalgo. “Neighbors and theevolution of the comparative advantage of nations: Evidence of internationalknowledge diffusion?” Journal of International Economics 92, 1: (2014)111–123.

Bahar, Dany, and H. Rapoport. “Migration, Knowledge Diffusion and the Com-parative Advantage of Nations.” The Economic Journal .

Becker, Randy A, Wayne B Gray, and Jordan Marvakov. “NBER-CES Manu-facturing Industry Database: Technical Notes.” Technical Report February,2013.

22

Bernard, Andrew B., J. Bradford Jensen, and Peter K. Schott. “Trade costs,firms and productivity.” Journal of Monetary Economics 53, 5: (2006) 917–937.

Black, SE, and PE Strahan. “Entrepreneurship and bank credit availability.”The Journal of Finance LVII, 6: (2002) 2807–2833.

Bloom, Nicholas, Benn Eifert, Aprajit Mahajan, David McKenzie, and JohnRoberts. “Does Management Matter? Evidence from India.” The QuarterlyJournal of Economics 128, 1: (2012) 1–51.

Bottazzi, Laura, and Giovanni Peri. “Innovation and spillovers in regions :Evidence from European patent data.” European Economic Review 47, 4:(2003) 687–710.

Brainard, SL. “A simple theory of multinational corporations and trade withtrade-off between proximity and concentration.” NBER Working Paper Series, 4269.

Calonico, Sebastian, Matias D. Cattaneo, Max H. Farrell, and Rocío Titiunik.“Rdrobust: Software for regression-discontinuity designs.” Stata Journal 17,2: (2017) 372–404.

Calonico, Sebastian, Matias D. Cattaneo, and Rocío Titiunik. “Robust data-driven inference in the regression-discontinuity design.” Stata Journal 14, 4:(2014) 909–946.

Campante, Filipe, and David Yanagizawa-Drott. “Long-Range Growth: Eco-nomic Development in the Global Network of Air Links.” NBER WorkingPaper No. 22653: (2016) 50.

Carr, DL, JR Markusen, and KE Maskus. “Estimating the knowledge-capitalmodel of the multinational enterprise.” The American Economic Review 91,3: (2001) 693–708.

Caselli, Francesco. “Accounting for cross-country income differences.” Handbookof Economic Growth 1, 05: (2005) 679–741.

Cattaneo, Matias D, Nicolas Idrobo, and Rocío Titiunik. “A Practical Introduc-tion to Regression Discontinuity Designs.” Cambridge Elements: Quantitativeand Computational Methods for Social Science - Cambridge University PressI.

Caves, RE. “International Corporations: The Industrial Economics of ForeignInvestment.” Economica 38, 149: (1971) 1–27.

Costinot, Arnaud, Lindsay Oldenski, and James Rauch. “Adaptation and theboundary of multinational firms.” The Review of Economics and Statistics93, February: (2011) 298–308.

23

Coval, Joshua D., and Tobias J. Moskowitz. “The Geography of Investment:Informed Trading and Asset Prices.” Journal of Political Economy 109, 4:(2001) 811–841.

Fan, JPH, and LHP Lang. “The Measurement of Relatedness : An Applicationto Corporate Diversification.” The Journal of Business 73, 4: (2000) 629–660.

Giroud, Xavier. “Proximity and Investment: Evidence from Plant-Level Data.”The Quarterly Journal of Economics 128, 2: (2012) 861–915.

Gumpert, Anna. “The Organization of Knowledge in Multinational Firms.”http://papers.ssrn.com/abstract=2622415.

Hall, R. E., and C. I. Jones. “Why do Some Countries Produce So Much MoreOutput Per Worker than Others?” The Quarterly Journal of Economics 114,1: (1999) 83–116.

Harrison, Ann E., Inessa Love, and Margaret S. McMillan. “Global capital flowsand financing constraints.” Journal of Development Economics 75, 1: (2004)269–301.

Helpman, E. “A Simple Theory of International Trade with Multinational Cor-porations.” The Journal of Political Economy 92, 3: (1984) 451–471.

Helpman, Elhanan, Marc J Melitz, and Stephen R Yeaple. “Export VersusFDI with Heterogeneous Firms.” American Economic Review 94, 1: (2004)300–316.

Imbens, Gguido, and Karthik Kalyanaraman. “Optimal bandwidth choice forthe regression discontinuity estimator.” Review of Economic Studies 79, 3:(2012) 933–959.

Irarrazabal, Alfonso, Andreas Moxnes, and LD Opromolla. “The Margins ofMultinational Production and the Role of Intrafirm Trade.” Journal of Po-litical Economy 121, 1: (2013) 74–126.

Jaffe, A.B., M. Trajtenberg, and R. Henderson. “Geographic Localization ofKnowledge Spillovers as Evidenced by Patent Citations.” The Quarterly Jour-nal of Economics 108, 3: (1993) 577.

Kalnins, Arturs, and Francine Lafontaine. “Too Far Away? The Effect of Dis-tance to Headquarters on Business Establishment Performance.” AmericanEconomic Journal: Microeconomics 5, 3: (2013) 157–179.

Keller, Wolfgang. “Geographic localization of international technology diffu-sion.” American Economic Review 92, 1: (2002) 120–142.

. “International Technology Diffusion.” Journal of Economic LiteratureXLII, September: (2004) 752–782.

24

Keller, Wolfgang, and Stephen Ross Yeaple. “The Gravity of Knowledge.” Amer-ican Economic Review 103, 4: (2013) 1414–1444.

Kerr, W R. “Ethnic Scientific Communities and International Technology Dif-fusion.” Review of Economics and Statistics 90, 3: (2008) 518–537.

Lipsey, RE. “The Creation of Microdata Sets for Enterprises and Establish-ments.” Annales de l’INSEE , 30: (1978) 395–422.

Markusen, James R. “Multinationals, multi-plant economies, and the gains fromtrade.” Journal of international economics 16, 1984: (1984) 205–226.

Markusen, James R., and Keith E. Maskus. “Discriminating among alternativetheories of the multinational enterprise.” Review of International Economics10, 4: (2002) 694–707.

Markusen, JR. “Trade versus Investment Liberalization.” NBER Working PaperSeries , 6231.

Markusen, JR, AJ Venables, DE Konan, and KH Zhang. “A unified treatmentof horizontal direct investment, vertical direct investment, and the pattern oftrade in goods and services.” NBER Working Paper Series , 5696.

Nunn, Nathan, and D Trefler. “The boundaries of the multinational firm: anempirical analysis.” In The Organization of Firms in a Global Economy,edited by E Helpman, D Marin, and T Verdier, Cambridge, MA: HarvardUniversity Press, 2008, 55–83.

Oldenski, Lindsay. “Export Versus FDI and the Communication of ComplexInformation.” Journal of International Economics 87, 2: (2012) 312–322.

Pierce, JR, and PK Schott. “A concordance between ten-digit US HarmonizedSystem Codes and SIC/NAICS product classes and industries.” Journal ofEconomic and Social Measurement , 301.

Polanyi, M. Personal knowledge: Towards a post-critical philosophy. London,UK: Routledge, 1962.

. The Tacit Dimension. Chicago; London: University of Chicago Press,1966, 2009 edition.

Pritchett, Lant. “Divergence, big time.” The Journal of Economic Perspectives11, 3: (1997) 3–17.

Ramondo, Natalia. “A quantitative approach to multinational production.”Journal of International Economics 93, 1: (2014) 108–122. http://dx.doi.org/10.1016/j.jinteco.2014.01.004.

Ramondo, Natalia, and Andrés Rodriguez-Clare. “Trade, Multinational Pro-duction, and the Gains from Openness.” Journal of Political Economy 121,2: (2013) 273–322.

25

Shirotori, M, B Tumurchudur, and O Cadot. “Revealed Factor Intensity Indicesat the Product Level.” Policy Issues in International Trade and Commodities2010, 44.

Stein, Ernesto, and Christian Daude. “Longitude matters: Time zones and thelocation of foreign direct investment.” Journal of International Economics71, 1: (2007) 96–112.

Tintelnot, Felix. “Global Production with Export Platforms.”, 2014.

26

Figure 1: Unique locations of headquarters and subsidiaries

The figure shows a World map with the geocoded location of all the headquarters (triangles) andforeign subsidiaries (dots) in the sample.

27

Figure 2: Headquarters and foreign subsidiaries of an American MNC

The figure is an example of the resolution of the data. It shows a World map with the geocodedlocation of the headquarters of an American car manufacturing firm and all of its subsidiaries.

28

Figure 3: Estimated relationship of k and d

−.3

−.2

−.1

0.1

.2lo

g(d

)

−.4 −.2 0 .2 .4log(k)

95% CI Fitted values

log(d)= 3.74e−07 − .486*log(k)

The figure presents the empirical fit for the relationship between log(d) and log(k) (the latterproxied by the experience plus training measure), after controlling for unit trade costs and MNCfixed effects. 95% confidence intervals marked in grey, based on robust standard errors clustered atthe MNC and industry level.

29

Figure 4: KI and distance trade-off by working hours overlap

−3

−2

−1

01

Tra

de

off

Estim

ato

r

0 2 4 6 8 10Working hours overlap

−1

.5−

1−

.50

Dis

tan

ce

Km

(lo

g)

0 .2 .4 .6 .8 1KI (log, min=0)

0 2 4

6 8 10+

The figure presents the empirical fit for the relationship between log(d) and log(k) (the latter proxiedby the experience plus training measure), after controls specified in Specification (\eqref{eq:dd-dk}), and adding different controls for the ease of communication between the headquarters andthe foreign affiliate.

30

Figure 5: Running variable definition

The figure presents a graphical representation to understand how the running variable (distance tothe nearest time zone line) is centered around the time zone line.

31

Figure 6: Regression discontinuity graphical representation

−.0

4−

.02

0.0

2.0

4K

I M

ea

su

re

−400 −200 0 200 400Distance from TZ Line

The figure presents a graphical representation of the regression discontinuity design in Specification3.

32

Table 1: Descriptive Statistics (Domestic Vs. Foreign Subsidiaries)MNC # Subs Foreign (%) KIForeign KIDomestic ∆

All Observations 2030 84881 .32 .27 .34 -.079∗∗∗

Non OECD 73 2445 .13 .39 .37 .027OECD 1957 82436 .32 .26 .34 -.079∗∗∗

East Asia & Pacific 396 20579 .13 .45 .44 .0023Latin America & Caribbean 30 2247 .56 -.36 -.34 -.017North America 714 33679 .24 .36 .38 -.016∗South Asia 43 1557 .13 .38 .45 -.067∗Western Europe 847 26819 .55 .23 .17 .064∗∗∗

The table presents descriptive statistics from the sample. It presents for different cuts of the sample,based on the home country of the MNC, the total number of MNC firms, the number of subsidiaries,the proportion of those subsidiaries that are foreign (horizontal) subsidiaries, the average knowledgeintensity of the foreign subsidiaries, the average knowledge intensity for the domestic subsidiaries,and the difference between these averages, denoted by ∆. Stars represent statistical significance ofthe difference: ∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01

33

Table 2: Determinants of Foreign Replication of ProductionDependent Variable: Horizontal Foreign Subsidiary Binary Variable

(1) (2) (3)log(k) -0.1610 -0.1600 -0.3427

(0.096)* (0.079)** (0.090)***log(t) -0.0158 -0.0156 -0.0089

(0.028) (0.025) (0.020)GDpc (affiliate) 0.3271

(0.124)***Pop (affiliate) -0.0663

(0.018)***Cap per worker (subsidiary) -0.2772

(0.058)***HumCap per worker (subsidiary) -0.5447

(0.198)***Downstream Linkage 0.1905 0.1951 0.2366

(0.074)** (0.078)** (0.084)***Upstream Linkage 0.5607 0.5159 0.4744

(0.055)*** (0.055)*** (0.067)***

N 80542 80632 80487Adj. R2 0.59 0.62 0.67ϕh Y Y Yηs N Y -ηs × θP N - Y

The left hand side variable is a binary variable that takes the value 1 if the subsidiaryis foreign. The variables in the right hand side include the unit shipping cost associatedwith the industry, knowledge intensity measures and other controls. All specificationsinclude MNC fixed effects. Robust standard errors clustered at the industry and MNClevel are presented in parentheses.∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01

34

Table 3: Summary Statistics for Foreign SubsidiariesVariable N Mean sd Min MaxDistance, km (log) 25,166 8.201 1.15 1.1 9.9Knowledge Intensity (log) 25,166 4.112 0.13 3.6 4.4Unit Shipping Cost (log) 25,166 -1.894 0.76 -3.6 -0.1Ratio GDP per capita (log) 25,151 0.125 0.71 -2.7 3.8Ratio Population (log) 25,151 -0.423 1.71 -6.5 5.6Ratio Capital per Worker (log) 25,151 0.350 0.98 -2.9 4.9Ratio Human Capital (log) 25,151 0.118 0.33 -1.1 2.4Upstream Linkage 25,166 0.290 0.45 0.0 1.0Downstream Linkage 25,166 0.153 0.36 0.0 1.0Ease of Communication VariablesNon Stop Air Route 25,166 0.330 0.47 0.0 1.0Working Hours Overlap 25,166 6.960 2.47 0.0 10.0

The table presents descriptive statistics from the sample of foreign subsidiaries, for distance inbetween the headquarters and the subsidiary, the knowledge intensity of the foreign subsidiary andthe unit trade cost. The lower panel presents headquarters-subsidiary variables that measure theease of communication in between them.

35

Table 4: Distance, Knowledge Intensity and Ease of CommunicationDependent Variable: Log Distance to Headquarters

(1) (2) (3) (4)log(k) -0.5292 -0.3960 -1.4060 -1.2865

(0.268)** (0.230)* (0.515)*** (0.477)***log(k)XdirectFlightOAG -0.0955 0.0414

(0.200) (0.200)log(k)Xwhours_overlap 0.1520 0.1600

(0.071)** (0.071)**directFlightOAG -0.1020 -0.5741

(0.829) (0.835)whours_overlap -0.9359 -0.9742

(0.304)*** (0.305)***log(t) -0.1091 -0.0381 -0.0632 -0.0276

(0.079) (0.054) (0.052) (0.035)GDpc (affiliate) -0.1282 -0.1277

(0.267) (0.183)Pop (affiliate) 0.3237 0.1628

(0.043)*** (0.029)***Cap per worker (subsidiary) -0.4883 -0.6073

(0.110)*** (0.095)***HumCap per worker (subsidiary) 1.7045 1.4679

(0.503)*** (0.281)***Downstream Linkage 0.0489 0.0721 -0.0714 -0.0081

(0.099) (0.069) (0.076) (0.037)Upstream Linkage 0.0898 0.0740 0.0013 -0.0349

(0.117) (0.076) (0.076) (0.033)

N 25166 25237 25166 25237Adj. R2 0.60 0.67 0.81 0.87ϕh Y Y Y Yηs N Y N Y

The table presents results for the estimation of Specification (2) using a sample of foreign subsidiariesof MNCs, including controls and interactions with between log(k) and ease of communication proxies.The left hand side variable is the distance from the foreign subsidiary to its global headquarters inlogs. The variables on the right hand side include knowledge intensity of the affiliate’s economicactivity (in logs), the unit shipping cost (in logs) and other controls. The right hand side also includesvariables measuring the ease of communication between a headquarters and its subsidiaries, both ontheir own and interacted with log(k). All specifications include MNC fixed effects. Robust standarderrors clustered at the industry and MNC level are presented in parentheses.∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01

36

Table 5: Regression discontinuity estimationDependent Variable: Foreign subsidiary’s knowledge intensity (log)

150 250 350 OptimalcloserTZ 0.0084 0.0060 0.0075 0.0066

(0.003)** (0.003)* (0.002)*** (0.002)***distTZzero -0.0000 -0.0000 -0.0000 -0.0000

(0.000)*** (0.000)*** (0.000)*** (0.000)***closerTZ × distTZzero -0.0000 0.0000 -0.0000 0.0000

(0.000) (0.000) (0.000) (0.000)

N 5232 9702 14179 12367Adj. R2 0.87 0.87 0.86 0.87

The table presents results for the estimation of Specification (3) using a sample of foreignsubsidiaries of MNCs estimated using several bandwidths for the running variable specified ineach column. The last column uses the optimal bandwidth computed using the methodologydescribed in Cattaneo et al. (2018) who build on the work by Imbens and Kalyanaraman(2012). The estimation uses a triangular weight scheme, giving higher weight to observationscloser to the cutoff point. All specifications include MNC fixed effects and time zone line fixedeffects. Robust standard errors clustered at the MNC and time zone line level are presentedin parentheses.∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01

37

Table 6: Regression discontinuity estimation for control variablesDependent Variable: Control variables

log(ts) USs DSs log(GDPpcs) log(Kpws) log(Hpws)closerTZ -0.0240 0.0246 0.0186 0.0028 -0.0359 -0.0114

(0.018) (0.017) (0.012) (0.028) (0.052) (0.022)distTZzero 0.0001 -0.0001 -0.0001 -0.0003 -0.0001 -0.0001

(0.000)* (0.000)** (0.000)* (0.000)** (0.000) (0.000)**closerTZ × distTZzero -0.0001 0.0002 -0.0000 0.0002 -0.0003 -0.0000

(0.000)* (0.000)** (0.000) (0.000) (0.000) (0.000)

N 12006 12367 12367 12309 12323 12323Adj. R2 0.92 0.58 0.74 0.91 0.93 0.92

The table presents results for the estimation of Specification (2) using a sample of foreign subsidiaries of MNCs, includingcontrols and interactions with between log(k) and ease of communication proxies. The left hand side variable is the distancefrom the foreign subsidiary to its global headquarters in logs. The variables on the right hand side include knowledgeintensity of the affiliate’s economic activity (in logs), the unit shipping cost (in logs) and other controls. The right handside also includes variables measuring the ease of communication between a headquarters and its subsidiaries, both ontheir own and interacted with log(k). All specifications include MNC fixed effects. Robust standard errors clustered atthe industry and MNC level are presented in parentheses.∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01

38

Appendix forThe hardships of long distance

relationships: time zone proximity andknowledge transmission within

multinational firmsMay 30, 2018

A Conceptual FrameworkIn this section I augment the model by Helpman et al. (2004) – referred to asHMY hereafter – by including a new parameter capturing the intra-firm costof transmitting knowledge between headquarters and foreign subsidiaries. Thisextension allows us to understand how the cost of knowledge transmission facedby firms affects their decision to serve foreign markets. First the common set-upis described and then the proper adaptation is incorporated.

As in HMY, there are N countries producing H+1 sectors with labor as theonly input of production. H sectors (indexed 1, 2,...,H) produce a differentiatedgood, while the other sector (indexed 0) produces an homogenous good (whichserves as the numeraire). In any given country, individuals spend a share βh > 0of their income on sector h, such that

∑0≤h≤H

βh = 1. Country i is endowed with

Li units of labor and the wage rate in this country is denoted by wi.Consider now a particular differentiated sector, h. For simplicity of notation,

the index h is dropped in the next few paragraphs, but it is implicit that allsector specific variables may vary across sectors.29 In order to enter the industryin country i a firm bears a fixed and sunk cost fE denominated in units of labor.After bearing this cost, the potential entrant learns its labor-per-unit cost, a,drawn from a common and known distribution G (a). Upon observing this cost,the firm may choose not to enter, and thus bear no additional costs and receiveno revenues. If it chooses to produce, however, an additional cost of fD units oflabor is incurred. There are no other fixed costs if the firm chooses to produceand sell in the local market only.

The firm can choose to serve a foreign market either by exporting or creatinga foreign subsidiary. If the firm chooses to export, it bears an additional costof fX (per country it exports to). If it chooses to create a foreign affiliate, itincurs an additional cost of fI for every foreign market it chooses to serve thisway. Similar to HMY, fX can be interpreted as the cost of forming a localdistribution and service network in the foreign market, and fI includes all ofthese costs, as well as the cost of forming a subsidiary in the foreign countryand the overhead production costs embodied in fD. Hence, fI > fX > fD.

29Some sector-specific variables are explicitly kept in the notation, such as t and k, sincethese variables will be relevant in the empirical analysis.

1

The homogenous good is freely traded at no cost.30 Differentiated goodsthat are exported from country i to country j are subject to a “melting-iceberg”transport cost τ(t, dij) which is an increasing function of the per unit shippingcost of the good (denoted by t, and proxies for weight or other good specificcharacteristics) and the distance between countries i and j (denoted by dij). Itis assumed that that τ(t, dij) > 1. That is, a firm in country i has to ship τunits of a good for 1 unit to arrive in country j.

Analogously, serving a foreign market through an affiliate is subject to amarginal cost κ(k, dij) related to the transfer of knowledge. κ(k, dij) is as-sumed to be an increasing function of both the knowledge intensity of thegood (represented by k) and distance (dij). The cost of transferring knowl-edge includes resources and time used for communicating with foreign affiliatesto transmit proper knowledge required for efficient production. It is assumedthat τ(t, dij) > κ(k, dij) > 1 for all goods. The last inequality implies that for amultinational corporation, the cost of selling 1 unit of a good through a foreignaffiliate is κ(k, dij).

The cost of knowledge transmission in knowledge intensive sectors beinghigher is justified given that these sectors require higher interaction and com-munication among their workers. Thus, firms pay for business travel and com-munication services that occur more often within these sectors. In addition,and perhaps more importantly, knowledge intensive activities usually encom-pass tasks with higher probability of failure and thus requiring trained andexperienced workers. This too raises operational costs. Assuming that knowl-edge transmission costs are increasing in distance is consistent with empiricalevidence (e.g., Jaffe et al., 1993; Bottazzi and Peri, 2003; Keller, 2002, 2004;Bahar et al., 2014), and a standard way to model these costs in the literaturestudying MNCs (e.g., Ramondo and Rodriguez-Clare, 2013; Ramondo, 2014;Arkolakis et al., 2013; Tintelnot, 2014; Keller and Yeaple, 2013; Gumpert, 2015).

All the producers which serve a market engage in monopolistic competition.Consumer preferences across varieties of a differentiated product h have thestandard CES form, with an elasticity of substitution ε = 1

1−α > 1. It iswell known that these preferences generate a demand function Aip−ε for eachproduct in the industry in country i, where Ai = β´

0

nipi(s)1−εds

Ei, ni is the

measure of firms active in the industry in country i, and pi(s) is the consumerprice for a product indexed s.

In this setting, an active producer with labor requirement of a optimally setsa price of wia

α . Consequently, the price of a locally produced good is wiaα , the

price of a good which is exported to country j is τ(t,dij)wjaα , and the price of a

good that is sold by a foreign affiliate in country j is κ(k,dij)wjaα , where a is the

labor required for the producer to manufacture one unit of the product.In what follows, it is shown that the balance of forces ruling the trade-off of

serving a foreign market through exports or FDI is influenced by the knowledge30Thus, as long as the numeraire good is produced in all countries the wage rate is equalized.

2

intensity of the product.The assumption that the numeraire good is produced in each country simpli-

fies the analysis, as it implies that the wage rate is equalized across all countriesand is equal to 1. Hence, the operating profit for a firm in country i witha labor coefficient of a from serving the domestic market maybe expressed asπiD = a1−εBi − fD, where Bi = (1− α) Ai

α1−ε . The additional profits from ex-porting to country j are πiX = (τ(t, dij) · a)

1−εBj − fX and those from selling

in country j through a foreign affiliate are πiI = (κ(k, dij) · a)1−ε

Bj − fI . Birepresents demand parameters for country i and are considered exogenous toeach individual firm.

Hence, in this setting, the productivity parameter a will be critical for afirm’s decision of whether to serve the local market only or to serve foreignmarkets, either through exports or FDI. The sorting pattern is similar to theone in HMY and is based in the following equations:

(aD)1−ε ·Bi = fD, ∀i (A1)

(τ(t, dij) · aX)1−ε ·Bj = fX , ∀i, ∀j 6= i (A2)[

κ(k, dij)1−ε − τ(t, dij)

1−ε] · a1−εI ·Bj = fI − fX , ∀i, ∀j 6= i (A3)

Similar to HMY, the first two equations define the productivity thresholdsafter which firms will sell domestically or export, respectively. The minimumproductivity threshold after which firms will engage in FDI is derived fromEquation (A3).31 This threshold is defined as:

a1−εI =

fI − fX[(κ(k, dij))

1−ε − (τ(t, dij))1−ε]Bj

,∀i, ∀j 6= i (A4)

Predictions derived from this model will serve as the basis for the empiricalanalysis. The implications of the original HMY model are straightforward. Anincrease in τ(t, dij), either through an increase in either t or dij , will result inlower πE making it more likely to substitute exports with FDI. This is partof the mechanism of the concentration-proximity trade-off. However, with theinclusion of κ(k, d) into the model, some new predictions arise, assuming fullsymmetry in fixed costs and demand variables for all sectors and countries. Thepropositions are presented in terms of φ(aI) = a1−ε

I .

Proposition 1. For a higher level of k fewer firms will substitute exports to-wards FDI.

Proposition (1) is a direct consequence of adding κ into the model. Allelse equal, for higher levels of k, the threshold a1−ε

I for which above it FDIis more profitable than exports is larger, implying FDI will be less likely for

31Condition (A3) will have a positive solution if we assume κ(k, dij)ε−1fI >

τ(t, dij)ε−1fX > fD, which is homologous to condition (1) in HMY (with equal wages across

countries), but including κ.

3

sectors with higher k. The graphical representation of the model in Figure A1shows the case for two sectors that differ in their knowledge intensity, k andk (where k > k). Notice that the profit functions for both sectors originatein the same fixed cost value fI , but the function is flatter for the sector k(dashed line). Hence, the productivity threshold required for a firm to substituteexports with FDI becomes higher for sectors with higher levels of k. That is,(aij,kI )1−ε > (a

ij,kI )1−ε.

[Figure A1 about here.]

Proposition 2. Conditional on expansion, firms face a trade-off between k andd

Proposition (2) follows from the assumptions that ∂κ/∂k > 0 ∂κ/∂d > 0.Figure A2 exemplifies this proposition.32 The figure represents the isoprofitcurves of a firm for both exports and FDI activities as a function of k and d. Afirm will choose to expand horizontally (over exporting) in every combinationof k and d where πD +πI > πD +πE , consistently with the equations describedabove. Notice that conditional on FDI, for a given level of profits πD + πI , themodel predicts a trade-off between k and d.

[Figure A2 about here.]

The predictions coming out of the model following the inclusion of an intra-firm cost of transferring knowledge (κ) have testable implications in the data.First, ceteris paribus, industries with higher levels of knowledge intensity will beless likely to expand horizontally to foreign destinations. Second, conditionalon horizontal expansion, firms will face a trade-off between distance to theheadquarters and knowledge intensity of the activity.

32The exact shape of the curves will depend on assumptions of κ(k, d). I assume a generalform for now.

4

Figure A1: Increase in k (knowledge intensity)

Graphical representation of the model, for a case considering two sectors with different levels of k,where k > k. The result suggests that the threshold aI is an increasing function of k. Thus, FDIwill be less likely for sectors with higher k.

5

Figure A2: Profit curves in k and d dimension

The figure is a graphical representation of a firm’s profit as a function of k and d. The curve πD+πE

represents total profits for an exporting firm, while the curve πD + πI represent total profits for afirm engaging in FDI instead.

6

B On the Dun & Bradstreet Dataset

B.1 ReliabilityThe Worldbase dataset collected by Dun & Bradstreet is sourced from a numberof reliable organizations all over the world, including public registries. Accord-ing to Dun & Bradstreet’s website, "the data undergoes a thorough qualityassurance process to ensure that our customers receive the most up-to-date andcomprehensive data available".33 However, it is important to acknowledge that,given the lack of access to public registries for every country, it is not possible toasses with full accuracy the representativeness of the data. Alfaro and Charlton(2009), however, compare the dataset with the US multinational firms sampleby the US Bureau of Economic Analysis, and find consistencies between the twodatasets.

Some basic relationships drawn from the sample behave as expected. Forinstance, the number of countries in which an MNC has foreign affiliates isrelated to the overall size of the MNC, as can be seen in Figure B1. In particular,the figure shows the relationship between the size of MNC firms (in number ofestablishments in the left panel, and in total number of employees in the rightpanel34) against the number of foreign countries in which their subsidiaries arelocated (on the vertical axis). Each observation in the scatterplot is an MNClabeled with its headquarters’ country ISO3 code. The figure shows smallerMNCs are present in fewer countries, while larger MNCs tend to be more spreadout in terms of the number of countries they have a presence in.

[Figure B1 about here.]

Focusing the analysis on the within-MNC dimension, by adding MNC fixedeffects to all specifications, significantly diminishes the sampling concerns fur-ther. This is because, while methods for gathering information may not besymmetric across countries, they would not systematically differ by firm. Itis reasonable to assume that the per-country likelihood of missing data wouldbe the same for all firms, controlling for the location of the headquarters of theMNC. Thus, concerns regarding biases caused by possible sampling asymmetriesare not particularly large for the purpose of this empirical exercise.

B.2 IndustriesWhile the dataset has information on up to six industries per plant (a main oneplus five other) the number of establishments that report more than one activityvaries dramatically per country. The left panel of Figure B2 shows the averagenumber of reported industries across all subsidiaries per country, while the rightpanel shows, per country, the percentage of firms reporting one, two, three, four,five or six industries. In most countries, the average number of reported firms

33http://dnb.com.au/Credit_Reporting/The_quality_of_DandBs_data/index.aspx34Including their domestic plants for both.

7

is below two; and the majority of firms in more than half the countries reportonly one SIC code.

[Figure B2 about here.]

I also present results of the distribution of sectors among foreign affiliates,to understand whether in the sample there are some sectors that are morelikely to appear (i.e. be reported) than others. In terms of industries, thedistribution of different sectors in the sample is not homogenous, as can be seenin Figure B3. Some sectors are more prevalent than others in the data. Theindustries that appear the most in the data are Ready-Mixed Concrete (SIC3273), Pharmaceutical Preparations (SIC 2834) and Motor Vehicles Parts (SIC3714). To alleviate concerns on how this distribution could affect the results, allthe standard deviations calculations allow for clustering at the industry level.

[Figure B3 about here.]

In addition, it is worth emphasizing that each foreign subsidiary in the sam-ple manufactures a specific product. Hence, if a MNC has several foreign sub-sidiaries, then each one of those could be manufacturing a different product (inits 4 digit classification). The sample that a single MNC that has more thanone foreign subsidiary could be manufacturing more than one product. FigureB4 shows that larger MNCs (as measured by number of affiliates) tend to makea larger number of different products.

[Figure B4 about here.]

B.3 Using the input-output table to define vertical rela-tionships

In order to filter out from the definition of horizontal those links that couldalso be defined as vertical, either upstream or downstream, I use the US input-output provided by Fan and Lang (2000). I follow the methodology suggestedby Alfaro and Charlton (2009) and Acemoglu et al. (2009) to define verticalrelationships.

More in general, the diagram in Figure B5 is useful to understand how hor-izontal and vertical links are defined in the dataset. Within a single MNC firm,an horizontal link is defined as a foreign subsidiary that is classified under thesame SIC code as any of its domestic subsidiaries. Then I use the US I/O tableby Fan & Lang (2000) to define vertical relationships, both downstream andupstream. A subsidiary is defined as upstream vertical if its main economicactivity is an input of $0.05 or more per each dollar of output of any of the do-mestic subsidiaries of the firm. Similarly, a subsidiary is defined as downstreamvertical if any of the domestic subsidiary provides an input to it of $0.05 or moreper each dollar of output.

After such classification, those subsidiaries that fall into both categories(horizontal and vertical) are filtered out from the horizontal classification. This

8

implies that the sample classifies as horizontal only final goods (which is thematter of study of the theoretical framework presented in Online AppendixSection A).

[Figure B5 about here.]

The use of $0.05 in the main body of the paper follows the precedent set byAlfaro & Charlton (2009), but its choice is not determinant for the results ofthe paper.

A limitation of this methodology is that technologies might vary across coun-tries, and hence, the US I/O table would loss some validity in defining upstreamor downstream relationships. While acknowledging this limitation I assume thatthe US I/O table is a good proxy for measuring vertical links, regardless of thecountry, in line with the previous literature.

9

Figure B1: MNC size vs. number of countries

USA

JPN

JPN

USA

DEU

JPN

USA

JPN

JPN

JPN

USA

JPN

MEX

USA

JPN

USA

IRL

DEU

USA

USA

ESP

JPN

JPN

USA

USAUSA

FRA

USA

USA

KOR

DEU

USA

USA

DEU

DEU

USA

FIN

KOR

JPN

USA

MEX

DEU

USA

USA

JPNJPN

USA

USA

USA

USA

USA

JPN

USA

IRL

USA

USA

USA

JPN

USAJPN

DEU

JPN

JPN

USA

JPNJPN

USA

JPN

JPN

USA

DEUFINUSA

JPN

USA

USA

USA

JPN

USA

JPN

USA

USA

USAUSA

DEU

JPN

USA

USAUSA

USA

JPN

USA

USA

JPN

LUX

USA

USA

USA

JPN

USAUSA

USA

JPN

DEU

JPN

JPN

JPN

JPN

JPN

USA

MEX

IND

USA

USA

USA

JPN

CAN

USA

JPN

JPN

USA

JPN

ITAJPN

FRA

USA

USA

JPN

JPN

USA

USA

USA

JPN

USA

JPN

USA

USA

USA

JPN

FIN

CAN

JPN

JPNUSA

USA

USAFRA

JPN

DNK

DEU

USA

JPN

USA

USA

USA

USA

JPN

USAUSA

JPN

JPN

CAN

JPN

USA

USA

USA

BRA

USA

JPN

FIN

AUS

FIN

USAJPN

DEU

JPN

JPN

DEU

AUS

USA

USA

SWE

USA

USA

USA

JPN

USA

USA

AUS

GBR

USA

USA

USA

USA

JPN

DEU

USA

USA

IRL

JPN

USA

FIN

JPN

BEL

USA

JPN

CAN

USA

IRL

USA

FIN

USACAN

USA

USA

USA

USA

JPNJPN

USA

IND

JPNJPN

FIN

USA

NOR

USA

IRL

JPN

USA

JPN

USA

USA

USA

KOR

JPN

JPNUSA

JPN

DEU

JPN

USA

CAN

JPNJPN

USA

USAUSA

NOR

JPN

JPN

AUSJPN

GBR

GBR

JPN

USA

USA

JPN

JPN

USA

IND

JPN

USA

USA

JPN

USAUSA

DEU

USAESP

USA

JPN

USA

JPN

GBR

JPN

JPN

USA

DEU

USA

USA

JPN

USA

JPN

JPN

USA

JPN

CAN

DEU

USAUSA

JPN

JPN

JPNJPN

USA

NOR

USA

KOR

JPN

USA

JPN

JPN

USA

USA

USA

DEU

JPN

USA

JPN

USA

USAJPN

USA

JPNJPN

USA

USA

CAN

JPNCANJPN

DNK

JPN

USA

JPNJPN

USA

CANUSA

USA

USA

JPNJPNMEXCAN

JPN

AUSJPN

USAUSA

DNK

USA

DEUUSA

LUX

JPN

IND

SAU

USA

JPN

USA

USA

BEL

JPN

ESP

JPN

USA

JPN

USA

JPN

JPN

JPN

JPN

DEU

USA

JPN

USA

USA

USA

USA

USA

USA

USA

AUT

JPN

JPN

USA

JPN

JPNJPN

ESP

CAN

USA

USA

USA

USA

JPN

JPNGBR

DEUDEU

FIN

DEU

JPNCHN

JPN

KOR

USA

USA

FIN

JPNJPNDEU

DEU

DEU

BMUUSA

USA

JPNESP

CAN

USA

JPN

USA

DEU

USA

USA

JPNUSAIND

JPNJPN

GBR

JPN

JPNJPN

DEUBELJPN

USA

JPN

NOR

JPN

USA

USA

JPN

USA

ITAJPN

DEU

JPN

USA

USA

USA

USA

ISR

USA

USA

JPN

USA

USA

USA

USA

DEU

AUSGRC

USA

USA

USA

ITA

JPN

USA

USA

USA

USA

USA

USA

JPN

USA

USA

JPN

ITA

AUT

USA

JPN

USA

KOR

USA

JPNJPN

JPN

AUS

DEU

USA

JPN

JPN

USA

USA

USA

JPN

JPN

USA

USAJPNJPN

USAJPN

USA

JPN

USAJPN

JPN

USASWE

NOR

USA

SWE

ESP

USA

USA

USAUSA

USA

FRA

ITA

USA

USA

USA

JPNUSA

ITA

JPNJPN

USA

USA

USA

JPN

CAN

JPN

USA

USA

JPN

DEU

JPN

USA

USAJPN

DEU

USAMYS

USAITA

USA

JPN

USA

JPNJPNIRLUSAJPN

USA

JPNDNK

JPNJPN

JPN

TWN

JPN

JPNUSA

USA

USA

USA

USAUSA

USA

JPN

DEU

USA

JPN

USA

DEU

USA

JPN

JPN

JPNIND

USA

USA

FIN

DEU

USA

JPN

JPN

USA

GBR

USA

USA

USAUSAAUTUSA

USA

USA

JPN

USA

JPN

JPN

USAJPN

USA

USA

USA

ESP

CAN

BEL

USA

ESP

DEUITAJPNDEU

USA

AUS

USA

JPN

KOR

DEUJPN

JPNITA

DEUUSA

USA

USA

JPN

USA

USAUSAJPNJPN

DNK

USA

JPN

USACANCAN

TWN

GBR

USA

JPNNZL

DEU

USA

JPN

USA

USA

TWN

NOR

NLD

USAUSA

USA

DEU

USA

CHE

JPNJPNDEUJPNUSA

USAIND

GBR

DNK

USA

DEUINDUSA

JPN

IND

USA

USAUSAFINJPN

ITA

USA

AUT

JPN

DEU

USA

JPN

DEU

JPN

JPNDEU

DEU

DEUUSA

USA

DEU

USA

CAN

USA

JPN

USA

JPNKORIND

JPN

DEU

JPNUSA

BEL

JPNDEU

USA

CAN

JPN

USA

USAAUS

USA

JPNUSA

USA

USA

DEU

ESP

USAUSAJPN

JPNCANESP

USAUSAUSA

JPN

JPN

USAJPN

USA

GBR

JPNJPN

CAN

CANJPN

AUS

ESP

JPN

USA

USAKOR

ITA

JPN

USA

JPNUSA

NLD

USA

USA

JPN

DEU

DEU

USA

USA

AUT

JPN

JPN

USA

USACHLJPNITA

CAN

USAUSA

USA

JPN

USA

GBRDEUJPN

JPNGBRUSA

CAN

USA

JPNGBR

JPN

JPNCHE

SWE

NZLUSA

CANJPNIND

JPNJPN

USA

USA

KOR

JPN

CANJPN

GBR

USAUSA

USA

DEU

JPN

JPNUSA

DEUCAN

USAFRA

DEU

JPN

DEU

JPNUSAAUS

KOR

USAJPNCAN

USAUSA

JPN

JPN

JPNAUSCANAUS

JPN

FIN

JPNFIN

USA

ITAUSAFRABRAUSAJPN

DEU

DEU

USA

ITA

JPNESP

JPN

USAJPN

DEU

USA

IND

USACANUSA

CHE

DEU

JPN

USA

HUNDEUUSAUSAJPN

USA

GBRUSA

DEUNLD

FINHKG

CAN

JPN

DEUIRL

DEUJPNNORAUSDEU

USA

DEU

USA

JPNJPN

NOR

CANJPNGBRUSAITAUSA

DEU

MCO

ITA

CANNOR

JPN

DEU

JPNUSA

USA

CANUSA

FRA

USACHLUSALUXDEU

GBR

BEL

USA

NOR

JPN

USA

USA

DEU

JPN

USA

JPN

USAUSACAN

JPN

USA

ISLESP

RUS

ESP

GBR

CANJPNJPN

DEU

ITACHEJPNDEUUSAJPNKORDEUJPNJPNUSA

ITA

USA

DEU

CAN

USA

JPNJPNJPNCAN

CAN

KORITAJPN

USA

USAINDESP

USA

USAJPNIND

CAN

IND

DNKUSA

DEU

ESPITAESPDEU

USA

ESP

DEU

MEX

USAMEXJPN

JPNUSA

SMR

USA

JPNUSA

DEU

USADEUJPNJPNESPJPNKORUSAUSA

USAFRA

DEU

GBRGBR

JPN

JPNUSA

CAN

JPNDEU

JPNJPN

DEU

JPN

USA

USA

JPN

USA

CANJPN

FRA

ITAUSA

JPN

FRAITAESP

JPN

JPN

USA

USA

TWN

USA

CANCHEMEX

ITA

KOR

TWN

CAN

ESPUSA

USA

DEU

SWE

ITA

IND

JPNJPNUSA

USAFRA

JPN

FIN

GBRCAN

USA

DEUUSA

GBR

NORUSAJPN

GRCDEU

CANBEL

USA

JPNJPNPRTJPN

SGPDEU

JPN

TWN

NORINDIND

JPN

USA

CANJPNITAUSA

GBR

GBR

KOR

USAUSAINDJPNFINTWNUSAJPN

DEU

ITAJPNUSA

JPN

JPNESPUSA

FIN

JPNUSAJPNDEUJPN

JPNDEU

DEU

CHE

BELDEUUSAUSAJPN

TWN

JPNUSA

ESPGBR

JPN

AUS

USAUSATWNUSA

DEU

USA

DEU

ITAJPNCANUSA

USA

USAJPNCANUSACANDEUUSACHEUSAUSAJPN

USA

USA

JPN

USAUSA

TWN

DEU

JPNUSAJPNKOR

DEU

INDUSADEUUSAJPNNOR

USALUX

ITADEUJPNFINJPNESPNORCANJPNJPNUSA

JPN

JPNNORFINCAN

DEU

JPNTWNCANUSAUSAUSAJPN

ITA

USAJPNITACAN

NOR

FRACHN

DNK

USAIRLUSADEUUSA

USA

CHNUSACANFRAKORUSAUSAJPN

DEU

FRADEUISR

DEU

IND

ISR

JPNDEUUSACANUSA

SAUDEUDEU

JPNJPN

DNK

JPNUSAIND

TWN

JPNCANJPNFINGBRGBR

NOR

INDDEUDEUJPNKORUSAUSA

DEU

JPNCAN

KOR

JPN

DEU

USAMEXBELJPN

AUS

USAITAKORGBRUSATWNUSAIRLCANCANFINJPNGBRDEU

USAUSA

USAGBR

USA

DEU

USA

JPN

ESP

CANUSACANDEUFRA

DEU

DEUBRADEU

FRA

AUTUSA

ISR

JPNDEUCANJPNAUTITAUSATWNISR

USATWN

CAN

AUS

USADEUJPNCANCANUSACAN

GBR

ISR

DEU

THA

JPNDEUJPNUSAGBRJPN

ITA

ITAJPNUSAJPNUSAESPITAGBRDEUJPNUSAUSA

IRL

ISRUSAGBRITATWNUSADEUISLUSAJPNGBRDEUCAN

DEU

USA

JPN

URYJPNESPMEX

USA

CANUSAFRADEUKORAUT

USA

DEU

USA

USAJPNNOR

FINDEU

ESPKORUSABELDEUUSAITAIND

DEUUSA

DEUFRAUSADEUDEUBEL

CHE

ESPUSA

JPNESPJPNCANITAITAESPGBRJPNESPDEUITANORGRCESPNLDUSA

GRC

ESP

DNK

GBRFINCAN

KOR

CANDEUDEUDEUDEUUSAJPNUSAITAJPNDEUJPNHKGGBRCHEGBRIRLJPNDEU

ESP

INDBEL

DEU

JPNUSAKORUSADEUUSAISRUSAUSAAUTNLDDEUUSAJPNPOLJPNDEUAUTNORCANGRCUSAHKGBELDEUAUSMYSJPNDEUHUNDEUDNKGRCJPN

USA

BELDEUDNKJPNCAN

DEU

DEU

ITA

ITAKORDEUUSAUSAJPNITAUSAITAUSATWNISRGBRESPJPNUSAUSAITA

NOR

CHECANNLDJPNFRACANJPNUSA

NOR

CANUSAPOLDNKCOLINDDEUDEUDEUCAN

ESP

CANUSAINDJPNDEUJPNUSAJPNPOLFRAESPGBR

FIN

DEUDEU

FRA

KORITAUSADEUITACANUSACANGBRAUTUSA

SVN

INDJPNJPNCANITABELUSADEUNORDEU

ESP

DEUESPLUXGBRHUNUSAITACANDEU

CAN

JPNDEUUSAUSA

USA

JPNJPNGBRUSAINDJPNAUTGBR

DEU

AUTITAUSAITAAUSCHEUSADEUDEUBELGBR

ITADEU

USAAUSCAN

ITA

JPNITAESPUSADEUJPNFRAARGJPNUSADEUIRLGBRNZLKORAUTUSAITABELUSA

GBRUSA

TWNTWNGBRFRACANUSADEUBELJPN

DEUFIN

ARGUSADEUDEUDEUUSAJPNJPNCHLUSAFINTWNBELDEUUSAFINCANJPNDEUUSAITACANDNKUSAUSAKORUSAHKGUSAGBRDEUDEUUSASGPMEXCANCANGBRMEX

DEU

HKGMEXKORCHECHNAUSJPNAUS

BEL

JPNUSA

CANDEU

DEUTWNESPUSAKORDNKJPNITACHEUSAUSAUSA

SGP

ISLUSAFRACHNTWNJPNFRACANGBRUSADNKUSACANDEUBRACHEUSAHKGTWNGBRDEUTWNDEUUSAGBRGBRCANGBRBELUSADEUFRAJPNUSAAUTCANUSAUSACHECANCANUSACANAUSUSAKORBHRITADEUDEUSVKUSABELJPNDNKKORNLDNLDCHEDEUDEUUSATWNCHECYMKORTWNFIN

USA

USAMEXUSADEUJPNBELCANDEUDEUKORDEUKORCANINDTWNAUTJPNITAUSACANAUSITADEUESPJPN

ISR

CHEESPGBRCANUSAITAGBRUSAGRCMYSJPNUSADEUCANJPNHKGESPJPNUSAGBRJPNTWNHRVDEUESPUSADEUUSADEUBELITAUSABELCHLUSAUSASWETWNITA

DEU

TWNUSAUSAGBRJPNDEUTWNTWNSGPCANCANDEUDEUUSAGBRCANLUXBELNLDNZLCANDEUUSATWNDEUKORAUTHKGISRTWNGBRUSATWNTWNBELKORCANCHLSWECHECANDEUCHNJPNCANURYKORINDDEUUSAHUNDEUAUSGBRCANKENDEUCOLDNKCHNTWNDEUJPNHKGDEUMUSKORCHNITABELSGPTWNCANKORKORCZECHEDEUUSATURCHNBELKORTWNHKGGBRTWNUSABELBELLUXAUTGBRCHEDEULTUTWNUSALUXCANPOLMYSCHNINDTWNTWNUSAUSAGBRHKGCANISRJPNDEUSWEITAUSAUSACANCANGBRGBRFRAUSAGBRCANCANCANDEUTWNCANGBRDEUESPUSAKORCHNUSAGBRGBRCANGBRJPNNLDUSATWNGBRSWEDNKCHEBELAUTGBRCANGBRUSAURYSWECANGBRDEUUSAAUTCANCANUSATWNHKGGBRSWEDEUGBRUKRGBRISRGBRAUSBELTWNLUXPRTGBRSWEFINDEUAUSPRTBELKORUSAGBRTWNDEULIEDEUSWEGBRDEUUSAINDCHEGBRTHACANCANTWNUSATWNAUTCANDNKDEUCANSWEHKGDEUAUSUSALTUGBRCANHKGSWEJPNCHEUSAUSASWEGBRUSADNKKORDEUCIVPRTDEUUSACHEDNKSGPHKGUSATWNGBRFRADEUFINSWEARETWNTWNSWEUSAIDNGRCFRADNKCANCZEAUTITAGBRTWNCANUSAUSAGBRDEUUSADNKMEXSWEGBRGBRUSACHEDEUHKGGBRDEUGBRTWNTWNUSAAUSUSAHKGTWNIRLCANGBRGBRGBRISLCANKORAUSCHEKORTWNUSAUSADEUTWNLUXARGJPNSWESWEISLUSANLDMEXUSAAUSAUSSWESVNGBRPERTWNFRAPRTAUTAUTPOLTURLTUUSAFRATWNCHEDNKCHEDEUDEUCANSWEDEUGBRNZLHKGSWECHEHKGCHESWEKORJPNGBRESPITATWNUSADNKDEUSWEHUNCANCHEESTFINMCODEUCZEPOLHKGPAKHRVDEUGBRUSASWETWNGBRCHEDNKDEUDEUDEUKORHKGSRBSWETWNCHECZEPOLPOLCZEUSAUSADEUDEUSWEINDUSACANDEUKORTWNISRCANDEUCANSGPHUNKORUSASWEBELGBRCANBELCANUSAUSADEUBELROMUSAHKGITAFRAJPNKORIRLHKGGBRHKGMYSCANGBRHKGESPTWNUSAMEXDEUTURPOLSWEGBRKORITANLDSWEUSAPOLCANTWNCHECHLUSADEUSWECANCHEGBRGBRHKGCHEPOLHKGBELGBRCANNORGRCSWEDEUGBRDEUCANUSAGBREGYPRTLIECANDEUCANLIESWEUSAISRCANJPNGBRMYSTWNDEUKORDEUFRAINDKORDNKCHNCHECANITATWNNLDESTJPNPOLSGPGBRUSASWECANPRTKORGBRCHETWNCHEFRASWEUSAHRVAUSKORUSAUSANLDUSAUSAFINFRAGBRMEXDEUDOMFRATWNGRCCANMYSTWNDEUNLDUSASWEFRAHKGKORSMRUSAJPNISRCANUSAITAUSAFINFRADNKARGDEUUSAUSAHKGFRACANTWNGBRINDCOLAUSKORGBRSWESMRDEUTWNCHETWNTWNGBRSWECANGBRSWESWEMEXGBRTWNTWNGBRPRTCHEUSAKORDEUIRLDEUSAUUSACHEDEUCANMYSGBRFINTWNHKGCHNAUTAUSGBRCANKORISRTWNUSAROMIRLSWEUSADEUAUSGBRBELGBRGBRNLDCANDEUHKGSWEAUSSMRDEUGBRINDNLDHKGESPHKGHKG CHEGBR

CAN

URY VIRDNKJPNHKGGBRHKG

JPN

CHE

USA

CHE

NLD

FRA

GBRBMU CANHKG USACAN

KOR

LIEARG NLDHKGSWESWELIEHKG

SWE

HKGSGP NLDHKGGBR

GBR

GBR ESPGBRGBR

NLD

ITAGBRSGPIRLLIEGBR

DEU

HKG

SWE

USA USAHKGHKGHKG FRALUXGBRUSA JPNLUX

NLD

GBRFRATWNHKG

FRA

GBR SGP

DNK

CHEMYS

NLD

ESP

SWE

DEUDEU

CHE

KWT

HKGHKGHKGNLDFINUSA NLD USAHKG

DEU

HKG

SWE

CHELBNGBRMYSCHE DEULUX CHELUX

IRL

HKGGBR

DEU

NORGBR

CHE

KOR JPNGBR GBR

LUX

USAUSAGBRGBR NLD

CHE

CYM SGP

BEL

LUXHKG DEU

BEL

DEU GBRJPNJPN

DEU

HKGBELHKG

LUX

TWNHKGCANDEU

KOR

HKG GBRPOL

CHE

CHE DEUITAESPUSA GBR

GBR

SGP USAHKG

GBR

CHECHE

BELGBRNLD

GBR

GBR

ITA

USA

DEU

CHN

HKG FRALUX BMUPRT USA

DNK

HKGDNKDEUHKGTWNIRL

NLD

HKGMUS NLD

FRA

NLD USAGBR NLD MEXMLT JPNDNKGBRBELHKGCHEHKGHKGTWN GBRDNK JPNNLDCYP

AUT

GBRHKG

GBR

HKGLUX DEUGBRCHE USABEL

JPN

LUX USAHKGLUXFINNLD DNKHKG AUTGBRAUSHKGHKG

FRA

VIR

CHE

HKG

ITA

JPN

HKG

HKGGBRHKG LUX USACHEAUS

DNK

HKGHKG USAGBR

NLDLUX

ISL

BEL

GBR

HKGGBR

DNK

FRA

LIE

GBRHKG AUT

FRA

HKGVIRHKG DNKCHENLDHKGCHE DNKHKGUSACHEDEU

FRA

HKGVIR

JPN

DNK

HKG HKGFRANLDHKGSWE VIRUSAGBRNORSWESWE HKGCANNLDAUT

CAN

GBRUSAHKG

CHE

DNKGBR NLDCOLAUTDNKHKGDNK

ITA

VIR

DEU

USA

BMU

HKG

GBRUSAUSA

ESP FRAGBRHKG

GBR

USAGBR

FINDEU

HKGFRA BMU

GBR

HKG

FRA

HKG AUSNLDDEU

DEU

LUX

AUTESPCHE NLD

LUX

FRA

AUTHKGGBR

ITA

AUSISRGBRCHE SGPGBRLUX NLDNLD IRLSWE DNK

ESP

GBR

DEU

CYMNLDCHE

GBR

IRLVIR GBRINDNLDPOLIRL

LUX

TURGBR

GBR SGPGBRNLD

USA

HKGCYPGBRGBRGBRAUTHKGHKGGBRGBRGBRCYMHKG CYMSWESWEHKGLUXHKGGBRAUTDNKKORGBRHKG GBR

DNK

GBRHKG

DEU

LUX

DEU NLD

SWE

GBRGBRGRC NLDHKG AUTBEL IRLJPNCHEIRL

DEU

USA

DNK

ITA USAHKG IRLHKGGBRHKG BELBELDEU GBRBMUMEXGBRVIRNLD

NLD

GBRURYHKGHKGGBRGBRGBRHKG

JPN

HKG AUSGBR CANDEUFRAGBR GBRITA AUTHKGESPUKRCHE GBRFRANLD CANDEU NLD

CHE

GBR

HKG

GBRGBR AUSGBR AUTIRL NLDUSASWEGBR USAISRNLDGBRLUX

USA

HKGGBRVIRDEU

LUX

GBR CHEHKG LUXHKG LUXHKGBEL

USA

PRTGBR

USA

GBR

GBR

VIR

SWEDEUBMU HKGDEUTWN PRTUSA

CHE

HKGIRLUSA

VIR

NLDCYMHKGDEU

FRA

DNKNLDGBRHKGCHEHKG

HKG

CYP

CHE

CHEHKG

NLD

GBRHKG GBRUSA ITACYP AUTLUX NLDIRL POLNLD

SWE

GBRCHE

GBR

LUX CHEDEUPRTHKGFRACYMCHECAN

ESP

AUT

LUXGBR NLDDNK ESP

DEU

BMU

JPN

SGPHKG DEUGBR USACHEHKG

NLD

GBR NLDHKG DEU

JPN

GBR SWEUSA

DEU

USA

USA

CHE ITA SGPBGR

NLD

GBR

CHE

CHEHKG CHE

DEU

USASWEHKG BMU

IRL

DEU

GBR IND USALUXDEUCANHKG

FIN

GBR AUTNORFRALIE DEU

SWE

DEUCHEGBRHKG DEUHKGSWE ITAUSAIRL

NLD

CHELUX USA

CHE

CHEDNKGBR

FRA

USA ESP

MLT

AUT

FIN JPNGBRHKG

LUX

FRA

JPNGBR GBR GBRDEULUX LUXRUS

USA

HKG USA

ESP

GBR

LUX

GBRVIR LUXSGPDEUHKGLUX CANHKGGBRGBR CHEFRA

SWE

HKG FRANLDHKG LUXHKGHKG

DEU

NLD USAFRA AUTHKGCHECHE

FRA

LUX

MEX

NLD

HKGLUX

NZL

GBR

NLD

HKGCYM PRTNLDHKGGBRNLD

NLD

CAN

NLD

HKG CHE USACHE

FRA

ISLVIR

USA

GBR LUXHKG

DNK

CAN

PRT

CHE

SWE

FRA

GBR

NLDHKG

ITA

CHEGBR USACANCYMTWN JPN

HKG ITA

SWE SGPUSA

FRA

BELHKG DEUHKGVIR USA

ITA

NLDHKG SGPSGP GBRUSA JPNESPGBRGBRNOR GBRCHE FIN ESP JPN ESPGBR DNKHKG

USA

DEUNLDGBRCHEDEULUXGBRHKG

FRA

NLDVIRGBRHKG BMU

NLD

BRA

DNK DNK

HKGDNK DEU

USA

GBRHKG

CHN

GBRLUXGBRLUXDEUIND CANHKGGBRHKGSWEGBRHKG DEU GBRHKG BEL

AUT

CHE GBRDEUBELGBR

DEU

DEUHKGGBRHKG CYPDEU NZL

ESP

NLDGBRGBRHKGGBRHKG USADEUHRV SMR

USA

HKG USANLDUSA

LUX

CHEDEU USALUXHKG

NLD

HKG HKG LUXHKGJPNDNK

CYM

NLDNLDHKGHKG ESP

USA

HKGLUXHKG GBR CANGBR USACYMLUXHKG NLDDNK USAHKG

FRA

ITADNK CYMHKG DEUGBR BHSHKG

DEU

HKG USAHKG USA CHEGBR SGPHKGHKG BMU LUXHKGGBRDEU

CHE

NLD

LUX

HKG NLDGBR

BEL

NOR

CHE

GBRHKGHKGIND HKGGBRHKG CHEGBRGBRKORGBRGBRGBR

CYM

NLDHKGSGPAUT

CHE

HKG

DEU

CHE

FRAHKG

CHE

HKGGBRGBR SGPFIN

DEU

HKGLUXSWEDNK DEU JPNNLD

CYM

GBRHKGHKGGBR LUXHKG INDHKGSWEHKG USAHKGHKGAUTUSAUSA

LUX

AUT

BEL

NORNLD USAGBRCHE USABRBCYM FRA

MYS

USAFIN

FRA

DEU NLDSWEGBRUSAJPNPAN GBRISRGBRSWEHKG

FRA

DNKUSADNK

ESP

GBR CHEHKGLUX

ITA

LUX JPN

ITA

GBRHKGGBR USAITAHKG DEU

DEU

USAGBR GBR

CHE

USAGBRSWE

CHN

GBR

ITA

HKG ARG

CHE

NLD FINMUSNLD CHE

JPN

VIR GBR

NLD

HKG

DEU

TWNUSAHKG BMUGBRGBR CHE

GBR

CAN LUX

CHE

GBR

CYP

GBRCYMLUXGBRHKG

GBR

DEUGBR

AUT

DEU

GBRHKGGBRCAN

LUX

ITAHKGPRTHKGGBRPAN

BRA

NLD

NLDNORGBRLBNCANGBRFRATWNHKGGBRHKGITAIRL GBRHKG

USA

BHS

DEU

GBR

FRA

GBRHKGGBRFRAUSAGBR

AUTJPN

GBR

DEUFRA

SWE

GBR

LUX

PRT ITADEU CHEIND

DEU

FRAHKGHKGHKG CYPPANIND

FRA

HKGHKGVIRGBR

BMU

ESP

JPNLUXITA

DEULUX

LUX

GBR

HKG GBRESP NLDNLD

GBR

JPNCYMGBRSGP GBRHKG HKGGBRCHE

CHE

DNK

NOR

FRACHEBELGBRNLDHKG

LUX

CYM GBRCYPHKGGBRIND

GBR

HKG CANNLDDEU

ITA

NORITA DNKBELCYP DEU FINLUXHKGCHE HKG FRAHKGCHEHKG

NLD

HKGPOLVIR

FRA

NLDGBRGBR

NLD

FRA NLDGBRDEU CAN

GBR

HKG

FRA

GBRUSA

JPN

BEL LUXCHE CANDEUHKG AUS ITACHEHKGGBRGBRDEU JPNITA IRLSGPGBR

GBR

CHEGBRUSAHKGCAN USAUSAHKGCHE

DEU

HKGGBR DEULUX USA

FRA

FRA

NLD

JPN

HKG

NOR

DEUHKG

DNK

LUX

DEU

USA

FRA

ESP

GBRHKGGBRHKG DNK USATWNGBR HKGCHE JPNCHEDEUHKGIRL

GBR

NLDBRAGBR

NLD

CHE

DEU

HKG IRLHKGHKGDEU USA

NLD

LUXGBR

GBR

USA

ITA

SWE

NLD

VIRGBRESPGBRGBRTWNCHEGBR HKGCAN HKGGBRHKG

LUX

HKGHKG GBRLUXGBR

FRA

DNK

GBR

NLD

GBR

DEU

FRA

GBRFRA ESPLUXNLDHKG HKGHKG SGPGBRGBRGBRHKG VIR

FRA

AUT USAGBR HKGGBR

CYM NLD

MEX

AUSIRL

MEXCHE

NLD

AUT

BEL

CHE

FRA

DEUDEU CAN USAUSABEL

ITA

CYP

DEU

POL LUXUSA DEUHKG USAAUS CANHKGKORUSA NLDFRANLDCHE

GBR

GBR FRAHKGAUS AUTHKGGBR NLD

DEU

USANLDGBRHKGBMUDEUUSAUSAHKGCZECANHKGGBRGBRTWNGBR FRALUXGBR CANHKG CHEGBRFRA FINGBR

DEU

BMUUSAHUNHKG IRLBEL

AUT

FRA CANHRV

JPN

NLDDEU

CHE

BHS

BMU

NLDIRLHKG

USA

LUX

GBRCYMGBR NLDUSALIE

NLD

JPN

GBR

USAUSA HKGPANUSA CHEESPCAN CANCYM

FIN

AUTNLD

GBR

USAGBR FRANLD GBRDEU

LUX

DEU

NZL

GBRCHEHKGAREESP

GBR

DNK

GBR

LUXDEU

GBR CZEUSAHKGMLTFINVIRBHSCHEVIRGBRIRL

CHE

GBRHKG

GBR

GBRNLD GBRMYSHKGMYS JPNNORGBRGBR JPNGBRGBRGBR GBRDEU

NLD

HKG LUX USATWNGBRBMUGBRDNK

ESP

SWE

CHE LUXDEU JPN

GBR

ESPNLD LIEHKGNLDJPNAUT

USA

CHEGBR NLDNLDGBR AUTGBRCHE URYGBRDEU

SWE

GBRDEUNLD

LUX

DEUGBRVIR FRADEUHKGNLDHKG

NLD

HKG NLDLUXGBR ITA USASWE

CHE

JPN

HKG DEU

FRA

ISR CZEHKGHKGGBRHKGBHR ESPLUX DEUNLDLUX

NLD

NLD ITAGBR

ISR

ITA

SWE

HKGGBR NLDGBR SWELUXHKGHKGHKGCHEGBRNLDGBRLUX

GBR

HKGSGPHKGCHE USANLDGBRGBRDEUHKG

SWE

MEX NLDNLDDEUCHEDEUGBR HKGGBRGBRNLD ITALUX

DEU

DEUGBR

CHE

GBRGBR NLDHKG FRACYPHKG BMUGBRVIRHKG SWEIRL AUT

GBRUSADEU CAN

JPNDNK

CAN

AUS

HKG NLDNLD

NLD

AUT ITA

CHE

HKG

DEU

CANHKGSGPGTMSWECHEGBRCHEAUT CHEGBRFINTWN COLHKGKNA LUXHKGHKGSWEGBRHKGCHE DEUITANLDDNKPAN LUX

AUT

HKG DEU

CHE

JPNHKGSWELUXHKG ESPTWN GBRGBRMYSHKGDEUGBRFRA

CHE

HKGLUXGBRSGPHKGCANHKG CHEHKG ZAFHKG VIR

LUX

VIRHKG

FRA

IRLDNK CHELUX

JPN

BEL

AUSUSAVIRESP

DEU

CHE

NLDDEUCAN FRA

LUX

CANDEUCYMGBR NLD NLDLUXSGP

GBR

CHENLDNLD AUSCHEGBR ESPHKGUSAMEX CANESP

CHE

HKGCHEPOL DEUPRTGBRGBRGBR SWEUSA USA JPNHKGHKGURY DNK

USA

USAHKG CANHKGHKG ESPLUXURY NLD

DEU

USACANNLD DEU

LUX

GBR

USA

LUX

HKGGBRLUX

SWE

GBR

USA

NLD DEU

MUS

GBRGBRHKG USACHE CHN CAN

CHE

ITA

GBR

DEU

GBR USA BMUGBR GBR

NLD

NLDHKG

NLD

DEULUXHKG BMU

BEL

ITAGBRGBR

ESP

GBRPANHKGHKGHKGDEU

ITADEULUX

BMU

TWN

USA

HKGGBRHKGHKG

ESP

MEXCYMNLDVIRGBRLUX CHEVIRNLDGBR

CHE

CANCOL

BEL

NLD

LUX

LUXCHEHKGGBRGBRGBRHKG DEU GBR FIN

DEU

DEU

GBR NLDGBRHKG ITACHE GBRGBR CYMDNKHKGHKG

AUT

FINHKGCHENLDHKG

SWE

HKGCANCANVIRESP

FRA

FRA

NOR

BMUHKGMYSHKG

ESP

FINHKG

NLD

DEU USAHKGVIR ITAGBRCHEUSAGBR ESPDEUKORVIRLUXLUXGBR GBR DEUHKGGBR IRLITA USAZAFDNK AUTMLT GBRNLDHKG KOR

ZAF

HKGGBRBMUNLDGBRGBR CYMISL ESPGBRAUT

AUT

NLD GRCVIRHKGCHNGBR

NLD

USAGBR CHEMUSGBR

CHE

ISRGBR ESPHKGGBR

DEU

TWNGBR ITA

NLD

HKG

CHENLD

LUX

GBR

VIRMUSHKGCAN USAGBRUSAHKGESPGBR FINGBRHKG

LUX

IRLDEUGBRHKGUSAGBR NLDGBRTWNAUTITAHKG

DEU

GBR

HKG LUXHKG LUXVIRGBR LUX

NLD

DNKGBRHKG

ESP

SWEGBRHKGLUXJPN

ITA

SWE

FRA

HKG

CHE

GBR

USAFINLUXHKGDNK GBRLUXGBRGBRHKGSWE DEU

GBR

GBRTWN

LUX

HKGHKGGBRGBRGBRHKGDEUGBRGBR DEUVIRGBR DEUCHEGBRSWEGBRGBR GBRTWNGBR FRAGBR JPNHKGHKGHKGFRA CANCHEGBRHKGGBRHKGHKGNLD

CHE

DNKGBRGBR CAN

USA

LUXHKGHKG NLDDEUDEUSWE

CHE

INDDEUHKG HKG SWE

FRA

USA

NLDDEUGBR

PRT

ESPCAN DEUGBR VIRGBRLUXGBR DEUHKGHKGGBR CHE USAGBR HKGGBRHKGUSAUSA

BEL

ITA

ESPCYM

CYM

DEU DNKLUX SWENLDHKGGBR CAN

GBR

GBRGBRHKG JPN

DNK

HKG

ISL

HKGNLD ESPGBRGBRGBRHKGIRLGBR

LUX

PAN BHSHKG LUXHKGHKGGBR PRTHKG AUTHKG

FRA

DEU

HKG

NLD JPN

CAN

CHE

FRA

CAN

NOR FRA

GBRSGPHKG CHEHKG

DEU

FIN

NLD

GBR

BEL

HKG

DEU

ESP

GBRGBRHKGHKGGBR

DEU

HKGGBRCZEHKG NLD

CHE

HKG USASWE GBRHKG CANCHEGBRFRA HKGTWNCANGBRHKGGBRGBRHKGTWNDNK LUXNLDESTDNK LUXESPGBRCYMHKG DEU

LUX

CHE DEUHKGUSABMU QATDEUNLDHKG USA

USA

GBR

BMUHKGSWE FRABEL GBRDNK USASWE SGPKORDEU CHEPANGBRHKGDEUHKG

AUT

FRALUX HKGGBRFRA

DNK

NLDDNKLIEGBRVIRGBR ITAIRL

SWE

USA

GBRGBRVIRHKGHKGGBRUSA

ESP

POLBEL USACHEHKGSWEHKGGBRGBRNLDGBRGBR DEU

GBR

HKGMLTSGP GBRHKGUSAHKG

JPN

GBRVIR NLDGBR NLD CYM USAGBRGBR LUX

NLD

NLD

CHE FIN

LUX

NLD

FRANLDDEU CHNFRA

USA

SWE

BMU

SWE

VIRVIRLUX CANCYM GBRCYMCAN

MEX

ESP

DEUCANTWNFRAHKG

NLD

LUX TURDEUGBRGBR

FINGBR

JPNBMU CANGBRHKGGBR

GBR

CANLUXGBR GBR

DEU

LUXAUS

NLD

AUT

IND DEUDEU

DNK

LUX

CYM

USACAN NLDGBRPRT

GBR

DEU

FRA

HKG USADEUSWEGBRDNK

SWE

HKG ITA

USA

ITASWE DNKNLDAUS

JPN

NLDHKG LUX SGPGBR USALUXGBRGBRHUNHKG NLD IRLPOLBELGBR DNKGBRCHE IRLHKGHKGHKGLUXESPJPN DEUUSAAUT

PRT

IRL USATWNNLD

NLD

LUXSWEHKGGBRFRAHKGSWEGBRHKG NOR

USA

AUT DNK

SWE

HKGNLD NLDMUSHKG

SWE

FRA USAPANSWE

GBR

GBR

SWE

NLD

HKGGBRHUNFIN

FIN

URY

LUX

USASWESWE

DEU

KOR

USANLDDNK BMU

ITA

GBR

DEU

HKGGBRHKG LUXNLDHKGHKG DEU

NLD

FRA

HKGGBRHKG DEU

CHE

CHE

HKGCANCHE FRACHE

USA

ITA

FRA

HKGNLDTWNNLD

GBR

USA

DEUGBR GBR

CHN

GBR LUXGBRLUXHKGDNKGBR CANGBRESPHKGGBRHKGGBR HKG

NLD

NLD FIN CANUSAHKG

NLD

DEU

DEU

HKGGBR LUXGBR HKGNLDKORGBRGBRGBRGBRHKG

USA

VIR NLDZAFNLD

PRT

DEU JPN

DEU

HKGGBRGBRHKG FRALUXBHSDEUCHEGBRHKGGBR LUX

BMU

GBRPANHKGDEU

DEU

NLD

NLD GBRHKGHKGCHE AUS USALUXGBRCHE

BMU

VIRHKG CHEVIRTWN VIR MLT USA

NLD

CAN DEUHKG DEUGBRHKGNLD VIRGBR SWENLDNLDNLDHKGESP

DEU

BMUUSADEU

NLD

GBR USANLD

USA

SWECHE

GBR

BMUGBR DEUNLDJPN USADNKLUXHKGGBR FRAITALUXHKGNLD

LUX

HKGMEXCYMHKG

CHE

HKGNLDHKGHKGVIRHKGSWE

NLD

DEU

CHE

CHE

HKG HKGSGP ITAITA USACHE NLD

GBR

USA

ITA

GBR

SWE

CHN

USA

DEU

CHEHKGUSA LIE

CHE

GBR CAN

FRA

HKG ESPDNK NLDGBR AUT AUSHKGHKGGBRBEL

GBR

ITA

CANNLDHKGGBR

GBR

GBRTWN BELDEULUXHKGGBR KOR SWE

SWE

GBR

BEL

ITA AUTNLDNLDDEU GBRVIRSGP

ITA

ESPCHE NLD

CHE

GBRFRA LUX GBRCAN DEUNLD ESPUSAUSACHE

DEU

CHEBEL USA

DEU

GBRFRAVIRHKG DEUGBR

FRA

GBR GBR

FRA

GBR

USA

GBR

FRA

BEL

USA

SWE

FIN

CYP

LUXHKGHKGCHE USA

FRA

DNKCHE LUXNLDBRA DEUHKG MUS

CHE

NLDLUXSMRCHE SWEGBRGBRDNK

FRA

ITALUXHKG

BEL

HKG NZL

DEU

CHE DEUHKG NZLDEUGBR MEXNLDHKG USA

AUT

GBRNLDGBRSGP

GBR

DEUGBR DEUHKGNLDDEU ESPHKGPOL FINPOLGBRCHEGBR NLDTWN

FRA

GBR

DEU

AUTHKGGBR CANHKG AUTHKG

SWE

USA CYMHKGDNK

NLD

DEU

USANLD

NORCHE

DEU

FIN ESPHKGGBR

FIN

HKG CHEGBRGBR USALUXHKGCOL CHLGBRNLD ITA

SWE

VIRNLD ESPTWN

DEU

GBRHKG USAAUTGBRESP ITA

CHE

LUXGBR LUX USAHKG LUX

USA

MEXGBR

CYM

HKG BELDNK

GBR

USACYPFRATWNFRAGBR CANLUXHKG CANUSAGBR SGP

SWE

CHECHE LUXHKG BELDEUFRAHKGCHE JPNNLDNORGBR LUXHKG HKGCZE HKGHKG ITAHKGGBR ISLITA

DEU

NLD

GBR

SWE

HKGLUXHKGHKG

LUX

CHENLDGBR

GBR

NLDGBRGBRCHEUSA AUTUSAGBRHKGHKGHKGHKGGBRHKG SWEHKG

AUT

LUXHKG JPN

JPN

NLD

IRL

PAN

DEU

HKG

NLD

HKGJPNDEUNLD USAITAHKG

DEU

GBRGBRCANGBR USADEUUSA

DEU

HKGSGP

BMU

HKG JPNLIEGBRDEUDEU JPNSWEHKG LUXHKG

SWE

DEUHKG

USA

LUXGBRGBRCANHKG VIR CYMCYMNLDUSA JPNNLDGBR USA JPNNICHKG

NLD

IRLGBR GBRDEU KORDEUHKG AUTCHEHKGBMU CHEVIR

GBR

USA

DNK

GBRITAUSACHEDEU LUXUSAGBRDEU CANPAN NLDAUTCHE BRAJPN ITA LUXHKG GBRCHE

DEU

HKG LUXHKG

CHE

GBRNLDVIR

CHE

DEU GBRNLDHKGGBR USA

GBR

GBRARECANVIR ITAITAGBRGBRPOL

CHE

GBR

CHE

NLD

HKGGBRHKGHKG NLDNLD CANUSAGBRGBRUSA LUXVIR ITAGBR USADNKFRAGBR

FRAITA

GBR

USA

NLD JPNFRA

FRA

HKGSGPGBR ITACANHKGBELGBR

USA

CHE

DEUFRAGBR CAN

NLD

CHEHKGGBRHKG LUXJPNHKGCHE

SWE

HKG

JPN

HKG FRASGPISR NLDHKGDEUNLDHKG

DEU

GBR

CHE

NLD

FRAJPN

JPNGBR

DEU

USAGBRHKGGBRCAN NLDHKGGBR USA

GBR

USA

IRL

VIRSWEGBR GBR

SGPDEU

GBR

FRA

AUTTWNNOR

BEL

GBRCHE VIRTWNNLDTWN

CHE

GBRVIRDNK CHE

CHNCHE ESP

DEUFIN ESPDEUITA

DEU

CYM

SGPCHEHKG NLD

LIE

USA DEUGBRHKG

ITA

GBRBELSWEDEU

NLD

PANGBR DEUNLDHKGHKG

NLD

HKG HKGDNKGBRAUS DEU

LUX

FRAHKGVIRHKGHKG

PRT

NZLHKG

JPN

LUXHKG

CHE

NLDUSA AUSGBR

DEU

GBR

LUX

LUX

GBRGBRLUXDEUGBRAUT CZELUXGBRHKGHKG CANHKGHKGDEUGBR

USA

DNK USA USAAUT

ITA

HKG

SWE

GBRHKG JPNHKGLBN NLDHKG PAN GBRMUS MEX

GBR

CHN DEUSWEBMU ITACYPUSAHKGDNKBELUSA AUS

GBR

SAU

AUSRUSHKGGBRHKGNLD DNKNLD

LUX

GBR BMUHKGGBRINDHKG

FRA

NLDGBR

GBR

MUS ITAGBRTWN

NLD

MEX

HKGIRL

DNK

DEU

MEXBMUHKGHKGGBRGBR GBRGBRCANLUXSVN

DEU

LUX

SGP CANGBR JPNGBRHKGGBRGBR GRLNLD

SWE

LUXKORGBR

AUT

GBRHKGGBRGBR BMUGBR JPNJPNGBRGBR

NLD

GBRJPNHKG

USA

GBRGBR

GBR

USA

ITA

ITAHKGHKGHKG

NZL

DEU USA

FRA

HKG

CHE

CHE

HKGGBRAUT CHEVIRCHEGBRPRT CHENLD ZAFGBRITA ESP

USA

LUXKOR

MEXISL

GBR DEUNLDDNK AUTLUXUSA HKGCYPPRT DEUGBR

DEU

LUXFRANLD ESP

ESP

DEU DEUSWEFINNLDCHEDEU FRAKORVIR CAN

LUX

JPN

NLD

TUR

FRA

HKGDEUHKG DEUFRA LUX

CHE

NLDLUXHKG AUTCAN HKGSWEGBR CHEJPNLIE JPNFRAGBR

BRA

FRA ISRIRL

FRA

PRTIRL

USA

HKGGBR USA

CHE

CANGBRAUT

GBR

ESPMYS GBRHKGGBR AUTFRAGBRGBR

AUT

HKGGBR CANGBR ISR NLD

LUX

NOR

GBR

ESPDNK

FRA

DEUGBRHKG

ITA

HKGGBR

GBR

GBRHKGGBRPANHKGGBRGBR FIN NLDHKGVIR

GBR

JPN

LIE

USAHKG

USA

SWE

NLD

GBR

DEUSWEDEUHKG

ITA

HKG

HKG

DEUNLDHKGFRA USAHKGHKGHKGDEUPRT

ITA

DEU

CHE

USAHKG NLD USAHKG TWNIRL

GBR

GBR FRO

USA

SGPHKG LUXGBR CANMYSNLD

USA

DEU DEUESPITA LUXNLDTWNGBR SGPGBRLUXDNK

JPN

CYPLUXSWETWN

AUS

USAUSA

DEU

LUXHKG FRASGPAUSNLD

SGP

JPN ITAGBRGBR

NLD

LUXSWECHEGBRGBRVIRVIR JPNGBR LUXNLDUSA

USA

GBR LUXJPNAUSGBR JPNGBRGBRTWNBEL

GBR

GBRFRAHKGFINHKGCANVIR

IND SWE

GBRGBR DEUCHEHKGSWE

FRA

NLDHKGGBRGBRCHEPRT FRA

GBR

ITADEU

USA

GBRNLDCHE MUSHKG

GBR

JPN

JPN

GBR

DNK

GBRHKG

VIR

GBRHKGGBRHKGAUS

ESP

HKG LUXIRL VIR FRAHKGHKGVIRDNK

LUX

JPN

CHE

BEL

DEU

DEU USA

BEL

LUX

FRA

GBRCHEGBR CANNLDLUX USAGBRGBR AUTNLDHKG CHE NLDLUXHKGGBRLUX USA USAUSA

SWE

NLDHKGFRAPRT

DEU

HKG DEULTU LUXHKG

ITA

DEUHKGBHSFINGBRNLDCHEHKGHKGHKG BMUPOL

DEU

CYM

HKGHKG

NLD

GBRLUXHKGHKG

FIN

DEU

ESP

FIN

SWEGBRNLDGBR LUXGBR

DEU

TWN AUT

USA

HKG BEL

FIN

SWE

HKG ESP

GBR

NOR ESPLUX CHEHKG JPNGBR

CHE

HKG

SWE

GBRHKG SMRBEL TURGBR

NLD

JPNVIRVIRLUXHKG

BRA

DEU

GBR

ITAGBR DNKDEU

DNK

GBRGBR NLD

NLD

INDGBRGBRGBR BELHKGGBRHKGGBRHKG

ESP

GBR

USAUSA USACHE BRACHE

CHE

HKG AUT

ESP

GBRFIN

JPN

HKG LUXUSAHKG DEUCOLNLD

NLD

URY CHE

GBR

ESPCYM

NLD

TURUSAGBR URYHKGHKGHKGHKG

DEU

ESP

GBRHKGGBRHKGFIN USAGBRCHEHKG GBRHKGHKGGBRDEU

FRA

SWE

LIEHKG

SWE

HKGTWN

DEU

FRALUXITASWE CHENLDCYMDEUVIRVIRHKG

AUT

JPN

MUSHKGDEUHKG LUXHKGUSALUXGBR ESP CANCANESP

JPN

NORHKG GBRGBRHKGDNKHKG

NLD

FINGBRGBRHKGVIRTWNHKG BRAGBR CHE GBRGBRNLD

DEU

GBRGBRGBRDNK

FRA

BMUMUS CHEGBR

GBR

SWE DEUVIRGBR MUSHKGLUX ISRLIEUSADNK PERHKG

DEU

CANGBR

CHE

GBR CANUSA ITA

AUS

NLD CYMDEUGBR NLD

FIN

GBRCYPHKGIRL

CHE

HKGHKGGBRGBRGBRHKGBELIRLGBRVIR SWESWELUXJPNHKG USALUXHKGCHEITADEU CANHKG

FRA

CYM

CYMGBR DEU

ITA

CHE

GBRGBR SWEGBRPOL

CHE CHE

HKG CHE

DEU

BELHKG USAMYS DEULUX USADEU HKGLUX ITANLDSWEGBRHKGCHELUXROMGBRGBR NLDNLDSWEGBRAUTCYMLTU CHE

FRA

JPN

JPNVIRGBR HKGSWEESTMYSCYMHKG USAJPNCYM

DEU

LUX

SWE

HKG

FRA

TCAHKGHKG

JPN

CHE NLD

NLD

JPN

FIN NLDDEU DEU

LUX

USACHEDEUHKGDEUCHEGBRLUXHKG USACYP

CHE

HKGHKGGBRHKG CAN LIE

NLD

USA

ISL JPNUSAGBRGBRGBR GBRPRT

FRA

HKGHKG

FRA

CHEHKGHKGHKGGBRNLDGBRHKG

PRT

CHE

NLD

GBRHKG DEUUSADEUCHEGBR

JPN

HKGLUX USAGBRGBRCANGBR

ITA

ZAF GBRHKG NLDGBRNLDGBRGBRNLD GBRNLDAUTHKGUSAHKG GBRPANUSA NLDHKGDEUGBRITACANSWE

GBR

LIESGPNLD IRLNLD SWEGBR PRT

CHE

DEUNLD JPN

USA

HKG

ITA

DEU JPNSGP

LUX

CYPESP

GBR

USAPAN USACHEHKG GBR

LUX

AUS

DEU

DEU

CHE

DEUHKGVENGBRGBR HKGGBRGBR ESPGBRJPNDEU NLDLUXHKGCANGBRUSAGBRUSAVIRHKGGBR

CHE

USAUSA

GBR

FRA

DNKNLDLUXCHE

NLD

BEL

MYS

ARE LUX

BRA

DEUVIR

DEU

GBRFINFIN GBRHKG CYPCHEGBRHKG LUX

CYM

HKG

DEU

FRA

USA

JPN

GBR ESPCHE

HKGLUX

FRA

SGPESP CHEIRLHKGJPNGBRHKGHKGHKG

USA

HKGGBRSWE

DNK

SWEHKG GBR FRAESPGBR HKGTHAUSASWE

CHE

NLDISR

NLD

DEU

DEU

CYM

NLD USA BHSGBRGBRGBRGBRSGPKORLUXHKG IRL LUXBELUSA

GBR

USALUXGBR

FIN

CYM SWECHESGP

DEU

SWEDEU PRTGBR DEU

DNK

HKGGBRHKG CHEESPPANGBR

DNK

TWN

DEU

HKG IRLGBRGBRGBREST

JPN

FRA

ZAF

NLD

DNK

DEUDNK ESPCHEGBR CHE DEUGBRGBRHKG GBR

DEU

SWEGBR BMUHKGNLDBLZ NLDUSA

GBR

CHE

GBR

FRA

LUXGBRHKGFRALUXNLDSWE VIRGBR

CHE USA

DNKSWEHKGGBR JPNGBR

NLD

SWE

GBR

NLD

SWE GBRGBRUSA CANLUXSGP

CHE

FRASWEGBR GBRGBRDEUHKGDEUNLD DEUHKGHKG

USA

DEU

ESP

ESPAUSNLD FIN

FRA

HKG CHEDEU GBRUSAHKG SGP

LUX

JPN

CHE

GBR

FRA

GBR

DEU

GBR

DEU

GBR

ESP

GBR JPN

CHE

SWEGBRHKG IRLHKG

NLD

SGP JPNGBR GBRLUX

GBR

ITA

GBRDEUDEUTWNGBRJPN

NLDCHE

IRL CHENORDEUHKG NLDGBRHKGGBRHKGHKGBRBSWE

MYS

HKG DNKCYPGBR

DEU

GBR

CHE

ISR

HKG ITAHKG DNKDNKGBRUSAFINGBR

DEU

USA

LUX

JPN

CHE

GBRNLDHKG GBRGBRHKG

GBR

JPN USADEU

FRA

DNK

LUX

FRA

HKGHKGUSAGBR AUT ITA

USA

FRAHKGCHE ITA

GBR

AUTHKG SGP

BEL

USALUXHKG SWENLD DEU

DEU

GBRHKG

CHE

PANLIEHKG

USA

DEUHKGGBRGBR AUTLUX DNKITAAUTHKGGBRMUS

FRA

CHE

SWE

CAN

GBR

JPN

USAGBR USALUX

ITA

ITADNKHKGCHE ITAHKGFRASWE GBRGBR

DEU

CHNNLDAUSNLDHKGHKGGBR

BMU

DEUGBR USAGBRGBR GBRNLDGBRGBRUSAGBR DEUESPHKG LUXGBRCHE

NLD

FRA

USAGBRGBR

CHE

DEUHKG BMUAUTHKG

NLD

ITAGBRGBR

HKG

HKGESTHKGTWN BELSWE

DEU

LUXSAU HKG FRAGBR BELPOLLUXHKGHKG NLDHKG CANGBRHKGHKGHKG SWEDEU NLDGBRJPN

ZAF

USA CANGBRHKGNLD DNKGBR

DEU

USAKWT

AUT

GBR SGPDEU USA

NLDESP

HKG ESPUSAUSAGBR SMRITAHKGNLDITA

AUT

GBRTWN AUTGBR

NLD LUX

DEUFRALIEHKGDEU

FRA

HKGHKG

FIN

DNK

FRA

LUXGBR GBRCHECHENLD ITA FRAGBRESP GRLHKGHKGCHE DEUGBR

FRA

IRL NLD CANUSATWNDEUIRLCHEGBR

CHE

FIN USATTOHKGSWEGBRHKGGBRHKGHKGGBR AUTHKGGBRLUX GBR NLD

SWE

NLD

DEU

CYMHKG GBRDEUHKGGBRCHEIND CYMPOL

BMULUX

NLDIRLHKG ZAFGBRNLDMLTHKG CYMAUT USAHKG

FRA

DEU DNKVIR CHE

NLD

JPNHKGGBRNLD

LUX

GBR NOR

DNK

DEU

USAGBR

ITA

NLDNLD

LUX

NLDHKGSWEUSA

JPN

GBR ITAPAN BELHKG USALUX USAITAHKG JPNGBRBEL IRLTWN

AUT

BELCHEHKGHKG JPNGBRGBRGBRUSAGBRHKG NLDFRAHKG

NLD

CHECYM

DEU

LTUHKG USAGBR GBR

ITA

BEL VIR BRALUXGBR PANGBRGBRHKG DEUGBRHKG USACHEGBR DEUCHNITA USA AUSHKGHKG CAN

FRA

GBRITA CHENORGBR HKGFINHKGDEUHKG JPNHKG

BEL

NLDNLDHKG FRA

GBR

HKGGBRHKGHKG ESP JPNHKGHKG BEL

GBR

CHEUSAGBRSWE USAGBRNLD

SWE

RUSDEU

BEL

DEUCHE CHE NLD

DEU

HKGGBRGBRGBRHKGGBR GBRDEUHKG GBRFINGBRGBRFIN ITADEUHKG

ESPDEU

NLD

THAUSA

CYM

DEUGBRNLD

NLD

CHE

DNKCAN LUX FIN

DEU

GBRPRTBEL USALUXLUX CHEVIRNLD

LIE

AUTHKG

ITA

DEU ESP

DEU

GBRGBRHKGAREGBR FRABELDEUPOL SWELUX

FRA

HKGCHE

USA

GBRNLDGBRHKGNLD SWE

GBR

IRLGBRCHE

GBR

VIR

DEU

HKG GBRGBRUSALUX LUXISR USALUXITA

FIN

BEL

SWE

CHE

ITA

VIR JPN

CHE

HKGBMUGBR GBRGBRBMU JPNLUXNOR JPNHKG HKG

FRA

HKG CHEHKGDEUISR SWE USACHLTWN

DEU

USA

AUT

USA USANLDCRIDNKLUXLUXVIR

DEU

HKGGBR BEL ESPDNK LUX BHRCHE

GBR

SMR FRAGBRVIR DEUGBR

DNKSWE

GBRDEUDEUCHEHKG NLDFINUSASWE

LUX

HKG ZAFITA HKG NLDGBRHKGHUN LUXDNK

BEL

HKG

USA

HKGDNKGBR

NLD

USAGBRHKGDNKGBRHKG

FRA

DEU

HKG DNK LUX

ITA

BMUCHEHKG PRTHKGGBRPANDEUGBRHKGUSA

CHE

GBR NLD ITANLDCHE

USA

USA

GBR GBRHKGCHE

CHE

HKGGBR FRAHKGGBRSGP BELNLDHKGHKG VIRLUX SWEFRA IRL JPNGBR NLDESPNLD SWECAN

JPN

USA

GBRHKG DEUNLDSGPGBR NLDGBRGBR USA BRANLDCHEUSAGBRGBR PRT

GBR

USA

CHE

HKG LUXHKG LUXGBRHKG BELHKG

SWE

GBR MUSDNK CHEUSAHKGGBR

NLD

GBRUSA JPNHKG CAN

ITA

GBRSGP

GBR USAUSAGBRUSATWNGBR LUX GBRSGPHKG HKG NLD USAGBRPERCHEFRA

ITA

CHE

FRA

BEL

AUT

AUT

LUX

USA

NLDGBRLIE

USA

AUSHKGGBRJPNGBR

GBR

ITAVIR

FRA

DNKBEL BHRHKGGBRTWNUSA

GBR

JPN

BEL

CANCYMTWNBHSGBR GBRGBR

DEU

FRA

ESPBELFRAGBRGBRHKGLUXGBRHKGHKGLUX ITAHKG GBRNLDGBRSGPHKG ITAGBR PANBMUGBRHKGHKG

CHE

JPNNOR

USAGBR

DNK

SWE

USA

USA USAGBR ITA

SWE

ISR GBR

GBR

HKGGBR DEUHKGGBR

FRA

HKG USAGBRDEUMACCANPAKGBRGBR NLDHKG DEU IRL

GBRUSA

ITA USALUXHKGGBR LUXAUTBEL GBRGBRGBRHKGLIE

CHE

VIR LUX

BEL

GBRSGPCANDEU

CHN

CHEAUTDEU USA

DNK

ITA CHEHKGHKGTWNPRT DNKNLDHKGTWN ZAF

DEU

DEUDEU FRAUSA

AUS

MUS

GBR

USA

GBRAUT

FIN

HKG GBRGBRDEUGBR

VIR

HKGGBR

CHE

GBRHKGGBRHKGHKGGBR DNKVIRGBR LUXLUX HKG

DEU

FRA

DEUHKGHKGPRT LUXLIEHKGGBRSWE USA DEU NLDHKGDEUCHEGBRHKGGBRHKGHKGGBR LUXCHE

NLD

HKGGBR DNKDEUHKG USAUSA BRA

FRA

HKG CANNLDGBRGBRFRADNKGBR FINHKGHKGGBRGBRNLDHKGGBR DEUHKGHKGUSAGBRNLDSGP DEUHKG CANISRDEUHKG

NLD

DNK

NLD LUXHKGGBR

FRA

HKGNLD

DEU

GBR

GBRNLD JPN

LUX

GBRHKGHKG

NLD

SWEDEU DEUNLDDNK

LUX

HKG USAAUT NLDGBR DEUHKGGBRHKGGBR

CHE

VIRHKGGBRLUX DEU

DNK

COLUSA

DEU

NLDGBR

SWE

SWE JPNGBR

DEU

HKGHKG

AUT

SGPGBR GBR

CAN

DEUSGPAUT SGPNORHKGROM LUXGBR DEUHKG NLD

DEU

AREVIRUSANLDSWECHESWE CANESP

CHE

VIR USA

LUX

USA

CHEHKGHKGCHE BELLUX

ARE

HKGHKGHRVGBRDEU JPNPANSWEINDDEUGBRHKGUSA

CHE

HKG

SWE

GBR JPNUSAAUTFRA

SWE

DNKGBRHKGDNK NLDHKGUSA

NLD

SWEGBR

AUT

ESP

ITA

GBR

FIN

USADEU CHEHKG CANGBR CYMSGPGBR

ITA

SWEHKG

ESP

CYM DEUDEUGBR BMULUX

CHE

DEUGBRCHEHKGGBR BMUAREUSAHKG IND

DEU

GBRLUX

USA

AUT

VIR

NLD CANKORHKGHKGUSA

LUX

HKG

CHE

BEL BMUSGPHKG ITAGBRCHEFRA FRAHKG

FRA

AUT DEUHKG PRTGBR LUX

JPN

PANDEU JPNCHE

SWE

CHE FRAHKG LUXPRTHKGLUXPRTZAFGBRAUT ITASGP FRADEU DNKHKG BMUHKGNLD USAUSAGBRESPBELBELCAN NLDGBR NLDVIR IRLHKGDNK CANDNK

ITA

DNK

GBR CHE CANSWECHEAUS BMUSGPHKG

DEU

DEU ITAFRAMEX ITAHKG

DEU

CHEHUN CANGBR

DEU

MUSDEU GBRMYSHKGHKG NLDGBR

FIN

LUX USACAN USA USASWEHKGHKGAUSHKGSWE DNK ITAHKGVIRHKGHKGGBRUSA ITAHKG

USA

USA

GBRHKGCHEUSAHKG

USA

AUS

BEL

NLD

CAN

GBR

FIN

BEL DEU

DEU

IND

GBR

LUXGBRHKG

FRA

LUXFRA

NLD

ITAVIRGBRDEUGBRNLD

GBR

ARGHKGHKG DEUHKG

NLD

USA USA

FRA BELESPNLD LUXLUX

GBR

NLD

FRA

HKG VIR

01

02

03

0co

un

trie

s_

su

bs

0 1 2 3 4Log subsidiaries

USA

JPN

JPN

USA

DEU

JPN

USA

JPN

JPN

JPN

USA

JPN

MEX

USA

JPN

USA

IRL

DEU

USA

USA

ESP

JPN

JPN

USA

USAUSA

FRA

USA

USA

KOR

DEU

USA

USA

DEU

DEU

USA

FIN

KOR

JPN

USA

MEX

DEU

USA

USA

JPN JPN

USA

USA

USA

USA

USA

JPN

USA

IRL

USA

USA

USA

JPN

USAJPN

DEU

JPN

JPN

USA

JPN JPN

USA

JPN

JPN

USA

DEUFIN USA

JPN

USA

USA

USA

JPN

USA

JPN

USA

USA

USAUSA

DEU

JPN

USA

USAUSA

USA

JPN

USA

USA

JPN

LUX

USA

USA

USA

JPN

USA USA

USA

JPN

DEU

JPN

JPN

JPN

JPN

JPN

USA

MEX

IND

USA

USA

USA

JPN

CAN

USA

JPN

JPN

USA

JPN

ITA JPN

FRA

USA

USA

JPN

JPN

USA

USA

USA

JPN

USA

JPN

USA

USA

USA

JPN

FIN

CAN

JPN

JPNUSA

USA

USAFRA

JPN

DNK

DEU

USA

JPN

USA

USA

USA

USA

JPN

USAUSA

JPN

JPN

CAN

JPN

USA

USA

USA

BRA

USA

JPN

FIN

AUS

FIN

USAJPN

DEU

JPN

JPN

DEU

AUS

USA

USA

SWE

USA

USA

USA

JPN

USA

USA

AUS

GBR

USA

USA

USA

USA

JPN

DEU

USA

USA

IRL

JPN

USA

FIN

JPN

BEL

USA

JPN

CAN

USA

IRL

USA

FIN

USACAN

USA

USA

USA

USA

JPNJPN

USA

IND

JPNJPN

FIN

USA

NOR

USA

IRL

JPN

USA

JPN

USA

USA

USA

KOR

JPN

JPN USA

JPN

DEU

JPN

USA

CAN

JPN JPN

USA

USAUSA

NOR

JPN

JPN

AUSJPN

GBR

GBR

JPN

USA

USA

JPN

JPN

USA

IND

JPN

USA

USA

JPN

USAUSA

DEU

USAESP

USA

JPN

USA

JPN

GBR

JPN

JPN

USA

DEU

USA

USA

JPN

USA

JPN

JPN

USA

JPN

CAN

DEU

USA USA

JPN

JPN

JPN JPN

USA

NOR

USA

KOR

JPN

USA

JPN

JPN

USA

USA

USA

DEU

JPN

USA

JPN

USA

USAJPN

USA

JPNJPN

USA

USA

CAN

JPN CANJPN

DNK

JPN

USA

JPN JPN

USA

CANUSA

USA

USA

JPNJPN MEXCAN

JPN

AUSJPN

USAUSA

DNK

USA

DEU USA

LUX

JPN

IND

SAU

USA

JPN

USA

USA

BEL

JPN

ESP

JPN

USA

JPN

USA

JPN

JPN

JPN

JPN

DEU

USA

JPN

USA

USA

USA

USA

USA

USA

USA

AUT

JPN

JPN

USA

JPN

JPNJPN

ESP

CAN

USA

USA

USA

USA

JPN

JPN GBR

DEUDEU

FIN

DEU

JPN CHN

JPN

KOR

USA

USA

FIN

JPN JPNDEU

DEU

DEU

BMUUSA

USA

JPN ESP

CAN

USA

JPN

USA

DEU

USA

USA

JPNUSAIND

JPNJPN

GBR

JPN

JPNJPN

DEU BELJPN

USA

JPN

NOR

JPN

USA

USA

JPN

USA

ITA JPN

DEU

JPN

USA

USA

USA

USA

ISR

USA

USA

JPN

USA

USA

USA

USA

DEU

AUSGRC

USA

USA

USA

ITA

JPN

USA

USA

USA

USA

USA

USA

JPN

USA

USA

JPN

ITA

AUT

USA

JPN

USA

KOR

USA

JPN JPN

JPN

AUS

DEU

USA

JPN

JPN

USA

USA

USA

JPN

JPN

USA

USAJPN JPN

USAJPN

USA

JPN

USAJPN

JPN

USASWE

NOR

USA

SWE

ESP

USA

USA

USAUSA

USA

FRA

ITA

USA

USA

USA

JPN USA

ITA

JPNJPN

USA

USA

USA

JPN

CAN

JPN

USA

USA

JPN

DEU

JPN

USA

USAJPN

DEU

USAMYS

USAITA

USA

JPN

USA

JPNJPN IRLUSAJPN

USA

JPN DNK

JPN JPN

JPN

TWN

JPN

JPN USA

USA

USA

USA

USAUSA

USA

JPN

DEU

USA

JPN

USA

DEU

USA

JPN

JPN

JPN IND

USA

USA

FIN

DEU

USA

JPN

JPN

USA

GBR

USA

USA

USA USAAUT USA

USA

USA

JPN

USA

JPN

JPN

USAJPN

USA

USA

USA

ESP

CAN

BEL

USA

ESP

DEUITAJPN DEU

USA

AUS

USA

JPN

KOR

DEUJPN

JPNITA

DEUUSA

USA

USA

JPN

USA

USA USAJPNJPN

DNK

USA

JPN

USACANCAN

TWN

GBR

USA

JPNNZL

DEU

USA

JPN

USA

USA

TWN

NOR

NLD

USAUSA

USA

DEU

USA

CHE

JPN JPNDEU JPN USA

USA IND

GBR

DNK

USA

DEU INDUSA

JPN

IND

USA

USA USAFINJPN

ITA

USA

AUT

JPN

DEU

USA

JPN

DEU

JPN

JPNDEU

DEU

DEU USA

USA

DEU

USA

CAN

USA

JPN

USA

JPN KOR IND

JPN

DEU

JPNUSA

BEL

JPNDEU

USA

CAN

JPN

USA

USAAUS

USA

JPNUSA

USA

USA

DEU

ESP

USAUSAJPN

JPN CANESP

USAUSAUSA

JPN

JPN

USAJPN

USA

GBR

JPN JPN

CAN

CANJPN

AUS

ESP

JPN

USA

USAKOR

ITA

JPN

USA

JPN USA

NLD

USA

USA

JPN

DEU

DEU

USA

USA

AUT

JPN

JPN

USA

USACHL JPNITA

CAN

USAUSA

USA

JPN

USA

GBRDEUJPN

JPNGBR USA

CAN

USA

JPNGBR

JPN

JPNCHE

SWE

NZL USA

CANJPN IND

JPNJPN

USA

USA

KOR

JPN

CANJPN

GBR

USAUSA

USA

DEU

JPN

JPN USA

DEUCAN

USA FRA

DEU

JPN

DEU

JPN USAAUS

KOR

USAJPNCAN

USAUSA

JPN

JPN

JPN AUSCANAUS

JPN

FIN

JPN FIN

USA

ITA USAFRABRA USAJPN

DEU

DEU

USA

ITA

JPNESP

JPN

USAJPN

DEU

USA

IND

USACANUSA

CHE

DEU

JPN

USA

HUN DEU USAUSAJPN

USA

GBR USA

DEU NLD

FINHKG

CAN

JPN

DEUIRL

DEUJPN NORAUS DEU

USA

DEU

USA

JPN JPN

NOR

CANJPNGBR USAITA USA

DEU

MCO

ITA

CANNOR

JPN

DEU

JPN USA

USA

CAN USA

FRA

USACHL USALUX DEU

GBR

BEL

USA

NOR

JPN

USA

USA

DEU

JPN

USA

JPN

USA USACAN

JPN

USA

ISLESP

RUS

ESP

GBR

CANJPN JPN

DEU

ITACHEJPN DEU USAJPN KORDEU JPNJPNUSA

ITA

USA

DEU

CAN

USA

JPNJPN JPN CAN

CAN

KORITAJPN

USA

USAINDESP

USA

USAJPNIND

CAN

IND

DNK USA

DEU

ESP ITAESP DEU

USA

ESP

DEU

MEX

USAMEXJPN

JPN USA

SMR

USA

JPN USA

DEU

USADEU JPNJPNESP JPN KORUSAUSA

USA FRA

DEU

GBRGBR

JPN

JPNUSA

CAN

JPNDEU

JPN JPN

DEU

JPN

USA

USA

JPN

USA

CANJPN

FRA

ITA USA

JPN

FRAITA ESP

JPN

JPN

USA

USA

TWN

USA

CANCHE MEX

ITA

KOR

TWN

CAN

ESP USA

USA

DEU

SWE

ITA

IND

JPN JPNUSA

USAFRA

JPN

FIN

GBR CAN

USA

DEU USA

GBR

NORUSAJPN

GRCDEU

CANBEL

USA

JPNJPNPRT JPN

SGPDEU

JPN

TWN

NORIND IND

JPN

USA

CANJPNITA USA

GBR

GBR

KOR

USA USAINDJPN FINTWN USAJPN

DEU

ITA JPN USA

JPN

JPNESP USA

FIN

JPN USAJPNDEU JPN

JPNDEU

DEU

CHE

BELDEU USAUSAJPN

TWN

JPN USA

ESP GBR

JPN

AUS

USAUSATWNUSA

DEU

USA

DEU

ITA JPN CANUSA

USA

USAJPNCAN USA CANDEU USACHE USAUSAJPN

USA

USA

JPN

USAUSA

TWN

DEU

JPNUSAJPN KOR

DEU

IND USADEU USAJPNNOR

USALUX

ITADEUJPNFINJPN ESP NORCANJPN JPN USA

JPN

JPNNORFIN CAN

DEU

JPN TWNCANUSA USAUSA JPN

ITA

USA JPNITACAN

NOR

FRACHN

DNK

USA IRLUSADEU USA

USA

CHNUSACANFRA KOR USAUSAJPN

DEU

FRADEU ISR

DEU

IND

ISR

JPNDEUUSA CANUSA

SAUDEU DEU

JPNJPN

DNK

JPN USAIND

TWN

JPN CANJPNFINGBR GBR

NOR

INDDEU DEUJPNKORUSA USA

DEU

JPN CAN

KOR

JPN

DEU

USAMEXBEL JPN

AUS

USAITAKOR GBRUSATWN USAIRL CAN CANFIN JPN GBRDEU

USAUSA

USAGBR

USA

DEU

USA

JPN

ESP

CANUSACANDEU FRA

DEU

DEUBRADEU

FRA

AUTUSA

ISR

JPNDEU CANJPNAUTITA USATWNISR

USA TWN

CAN

AUS

USADEU JPN CANCANUSACAN

GBR

ISR

DEU

THA

JPNDEU JPNUSA GBRJPN

ITA

ITA JPN USA JPNUSA ESPITA GBRDEUJPN USAUSA

IRL

ISRUSA GBRITA TWNUSADEUISL USAJPNGBRDEU CAN

DEU

USA

JPN

URY JPNESP MEX

USA

CANUSA FRADEU KORAUT

USA

DEU

USA

USAJPNNOR

FINDEU

ESP KORUSABELDEU USAITA IND

DEUUSA

DEUFRA USADEU DEUBEL

CHE

ESP USA

JPNESP JPNCANITAITAESPGBR JPNESPDEU ITANORGRC ESPNLD USA

GRC

ESP

DNK

GBRFINCAN

KOR

CANDEU DEUDEUDEU USAJPN USAITA JPN DEU JPN HKGGBRCHE GBRIRL JPNDEU

ESP

IND BEL

DEU

JPN USAKORUSADEUUSA ISRUSA USAAUT NLDDEUUSA JPNPOL JPNDEUAUT NOR CANGRCUSAHKG BELDEU AUSMYSJPN DEU HUNDEU DNKGRCJPN

USA

BELDEU DNKJPN CAN

DEU

DEU

ITA

ITA KORDEU USAUSAJPNITAUSAITA USA TWNISR GBRESP JPN USAUSAITA

NOR

CHE CANNLD JPNFRACANJPNUSA

NOR

CANUSAPOL DNKCOL INDDEU DEUDEU CAN

ESP

CAN USAINDJPNDEU JPNUSAJPNPOL FRAESP GBR

FIN

DEUDEU

FRA

KORITAUSADEU ITA CANUSA CANGBRAUT USA

SVN

IND JPNJPN CANITA BELUSADEUNOR DEU

ESP

DEUESP LUXGBR HUNUSAITA CANDEU

CAN

JPNDEU USAUSA

USA

JPNJPNGBR USAINDJPN AUTGBR

DEU

AUTITAUSAITA AUSCHE USADEUDEUBEL GBR

ITADEU

USAAUS CAN

ITA

JPN ITAESP USA DEU JPNFRAARG JPNUSADEUIRL GBRNZL KORAUTUSAITA BEL USA

GBR USA

TWNTWN GBRFRA CAN USADEUBEL JPN

DEU FIN

ARG USADEU DEUDEU USAJPNJPN CHLUSA FIN TWNBELDEUUSA FINCANJPNDEU USA ITACAN DNKUSA USA KOR USA HKGUSAGBRDEUDEU USASGPMEX CANCANGBRMEX

DEU

HKGMEX KORCHE CHNAUS JPNAUS

BEL

JPNUSA

CAN DEU

DEU TWNESPUSAKORDNK JPNITACHEUSAUSAUSA

SGP

ISLUSAFRA CHNTWN JPNFRA CANGBR USADNK USACANDEU BRACHE USAHKGTWN GBRDEUTWNDEUUSAGBR GBRCANGBRBEL USADEUFRAJPNUSAAUTCAN USAUSA CHECAN CANUSACAN AUSUSA KORBHR ITADEUDEUSVK USABEL JPNDNKKORNLD NLDCHE DEU DEUUSA TWNCHE CYM KORTWNFIN

USA

USAMEXUSA DEU JPNBEL CANDEUDEU KORDEU KORCANIND TWNAUT JPNITA USACANAUSITADEU ESP JPN

ISR

CHEESPGBR CANUSAITA GBRUSAGRC MYSJPN USADEU CANJPN HKGESPJPN USAGBR JPN TWNHRV DEU ESPUSA DEU USADEU BELITA USABEL CHLUSAUSA SWETWN ITA

DEU

TWNUSAUSAGBR JPNDEU TWN TWNSGP CANCANDEUDEU USAGBR CANLUXBELNLD NZLCAN DEU USATWNDEU KORAUT HKGISR TWNGBRUSA TWN TWNBELKOR CANCHLSWE CHECAN DEUCHNJPNCANURY KOR INDDEU USAHUNDEU AUSGBR CAN KENDEU COLDNK CHNTWNDEUJPN HKGDEU MUS KORCHNITA BEL SGPTWNCANKOR KOR CZECHEDEUUSATURCHN BELKOR TWN HKGGBRTWN USABELBELLUXAUTGBRCHE DEULTU TWNUSA LUXCANPOL MYSCHN INDTWN TWNUSA USAGBR HKGCANISRJPNDEUSWE ITAUSAUSACANCANGBRGBRFRAUSAGBRCANCAN CANDEUTWNCANGBRDEUESPUSAKOR CHNUSAGBRGBRCANGBRJPNNLDUSATWNGBRSWEDNKCHEBELAUTGBR CANGBRUSAURYSWECANGBRDEU USAAUTCANCANUSATWN HKGGBRSWEDEUGBRUKR GBRISRGBRAUSBELTWNLUXPRTGBRSWEFINDEUAUSPRTBELKORUSAGBRTWNDEULIEDEUSWEGBRDEUUSAINDCHEGBRTHACANCAN TWNUSATWN AUTCANDNKDEU CANSWE HKGDEUAUS USALTUGBRCAN HKGSWE JPNCHE USAUSASWE GBR USADNK KORDEUCIVPRTDEU USACHEDNKSGP HKGUSATWNGBRFRA DEUFINSWEARE TWN TWNSWEUSAIDNGRCFRADNK CAN CZEAUTITAGBRTWN CANUSAUSAGBRDEUUSADNKMEX SWEGBRGBRUSACHEDEU HKGGBR DEUGBR TWNTWNUSAAUS USA HKG TWNIRL CANGBRGBRGBRISL CANKOR AUSCHEKOR TWNUSAUSADEU TWNLUXARG JPNSWESWEISL USANLD MEXUSAAUSAUSSWESVNGBR PER TWNFRAPRT AUTAUTPOL TURLTU USAFRA TWNCHEDNK CHEDEUDEUCANSWEDEUGBRNZL HKGSWECHE HKGCHESWE KORJPNGBRESPITA TWNUSADNK DEUSWEHUNCANCHEESTFINMCODEUCZEPOL HKGPAK HRVDEU GBRUSASWE TWNGBR CHEDNKDEUDEU DEU KORHKGSRBSWE TWNCHE CZEPOLPOL CZEUSAUSADEUDEUSWEIND USACANDEUKOR TWNISRCANDEU CAN SGPHUN KORUSASWE BEL GBRCANBEL CANUSA USADEUBELROMUSA HKGITAFRA JPNKOR IRL HKGGBRHKGMYS CANGBR HKGESPTWN USAMEX DEUTURPOLSWEGBR KORITA NLDSWEUSAPOL CANTWNCHECHLUSADEUSWECAN CHEGBR GBRHKGCHEPOLHKGBEL GBRCAN NORGRCSWE DEUGBRDEUCAN USA GBREGYPRTLIE CANDEU CANLIESWEUSA ISRCAN JPNGBR MYS TWNDEU KORDEUFRAIND KORDNK CHNCHE CANITA TWNNLD EST JPNPOLSGPGBR USASWE CANPRT KORGBRCHE TWNCHEFRASWE USA HRVAUSKORUSAUSA NLDUSAUSAFINFRAGBR MEXDEU DOMFRA TWNGRCCANMYSTWNDEUNLD USASWE FRAHKG KORSMRUSA JPNISR CANUSAITA USAFINFRADNKARGDEU USAUSAHKG FRA CANTWN GBRIND COLAUS KORGBRSWESMR DEUTWNCHE TWNTWNGBRSWE CANGBRSWESWEMEX GBR TWNTWNGBRPRTCHE USAKORDEUIRL DEUSAUUSACHE DEUCAN MYSGBR FIN TWNHKGCHN AUTAUSGBRCAN KORISRTWN USAROMIRLSWE USADEUAUSGBRBEL GBRGBRNLDCANDEU HKGSWEAUSSMRDEUGBRIND NLDHKGESP HKGHKGCHEGBR

CAN

URYVIRDNKJPN HKGGBR HKG

JPN

CHE

USA

CHE

NLD

FRA

GBR BMU CANHKGUSA CAN

KOR

LIEARG NLD HKGSWESWELIEHKG

SWE

HKGSGP NLDHKGGBR

GBR

GBR ESPGBRGBR

NLD

ITAGBRSGPIRL LIEGBR

DEU

HKG

SWE

USA USAHKGHKGHKGFRALUX GBR USAJPNLUX

NLD

GBRFRA TWN HKG

FRA

GBR SGP

DNK

CHE MYS

NLD

ESP

SWE

DEUDEU

CHE

KWT

HKG HKGHKGNLDFINUSANLD USAHKG

DEU

HKG

SWE

CHELBNGBR MYSCHE DEULUXCHELUX

IRL

HKGGBR

DEU

NORGBR

CHE

KORJPNGBR GBR

LUX

USAUSAGBRGBRNLD

CHE

CYM SGP

BEL

LUXHKG DEU

BEL

DEU GBRJPNJPN

DEU

HKGBELHKG

LUX

TWN HKGCANDEU

KOR

HKGGBR POL

CHE

CHEDEUITAESPUSA GBR

GBR

SGP USAHKG

GBR

CHECHE

BELGBRNLD

GBR

GBR

ITA

USA

DEU

CHN

HKG FRALUX BMUPRT USA

DNK

HKGDNKDEU HKGTWNIRL

NLD

HKGMUSNLD

FRA

NLD USAGBR NLD MEXMLT JPNDNKGBR BEL HKGCHEHKG HKGTWN GBRDNK JPNNLDCYP

AUT

GBR HKG

GBR

HKGLUX DEUGBRCHE USABEL

JPN

LUX USAHKGLUXFINNLD DNK HKGAUTGBRAUSHKG HKG

FRA

VIR

CHE

HKG

ITA

JPN

HKG

HKGGBR HKGLUX USACHE AUS

DNK

HKG HKG USAGBR

NLDLUX

ISL

BEL

GBR

HKGGBR

DNK

FRA

LIE

GBRHKGAUT

FRA

HKG VIRHKGDNKCHENLD HKGCHEDNK HKGUSACHE DEU

FRA

HKGVIR

JPN

DNK

HKGHKGFRA NLDHKGSWE VIRUSAGBRNORSWE SWE HKGCANNLDAUT

CAN

GBRUSA HKG

CHE

DNK GBR NLDCOLAUT DNK HKGDNK

ITA

VIR

DEU

USA

BMU

HKG

GBRUSAUSA

ESPFRAGBR HKG

GBR

USAGBR

FINDEU

HKGFRA BMU

GBR

HKG

FRA

HKG AUSNLD DEU

DEU

LUX

AUTESPCHENLD

LUX

FRA

AUT HKGGBR

ITA

AUSISRGBRCHE SGPGBRLUX NLDNLDIRLSWE DNK

ESP

GBR

DEU

CYMNLDCHE

GBR

IRL VIRGBRINDNLDPOL IRL

LUX

TURGBR

GBR SGPGBR NLD

USA

HKGCYPGBR GBRGBRAUTHKG HKGGBRGBRGBR CYMHKG CYMSWESWE HKGLUX HKGGBRAUTDNK KORGBR HKGGBR

DNK

GBRHKG

DEU

LUX

DEUNLD

SWE

GBRGBRGRC NLD HKGAUTBEL IRL JPNCHEIRL

DEU

USA

DNK

ITA USAHKGIRL HKGGBR HKGBELBEL DEU GBRBMUMEXGBR VIR NLD

NLD

GBRURY HKGHKG GBRGBRGBR HKG

JPN

HKGAUSGBR CANDEUFRA GBR GBRITAAUT HKGESPUKRCHE GBRFRANLD CANDEUNLD

CHE

GBR

HKG

GBRGBR AUSGBRAUTIRL NLDUSASWEGBR USAISRNLD GBRLUX

USA

HKGGBR VIRDEU

LUX

GBRCHE HKGLUX HKG LUX HKGBEL

USA

PRTGBR

USA

GBR

GBR

VIR

SWEDEU BMUHKGDEU TWNPRT USA

CHE

HKGIRLUSA

VIR

NLDCYM HKGDEU

FRA

DNKNLDGBRHKGCHE HKG

HKG

CYP

CHE

CHEHKG

NLD

GBR HKGGBR USAITACYP AUTLUX NLDIRLPOLNLD

SWE

GBR CHE

GBR

LUX CHE DEUPRT HKGFRACYMCHECAN

ESP

AUT

LUXGBRNLD DNK ESP

DEU

BMU

JPN

SGP HKGDEUGBR USACHEHKG

NLD

GBRNLD HKGDEU

JPN

GBR SWEUSA

DEU

USA

USA

CHEITASGPBGR

NLD

GBR

CHE

CHE HKGCHE

DEU

USASWE HKG BMU

IRL

DEU

GBRIND USALUXDEU CAN HKG

FIN

GBRAUT NORFRALIEDEU

SWE

DEUCHEGBR HKGDEUHKGSWEITA USAIRL

NLD

CHELUX USA

CHE

CHEDNKGBR

FRA

USAESP

MLT

AUT

FIN JPNGBR HKG

LUX

FRA

JPNGBRGBR GBRDEULUXLUX RUS

USA

HKGUSA

ESP

GBR

LUX

GBRVIRLUXSGP DEU HKGLUX CANHKGGBRGBR CHE FRA

SWE

HKG FRANLD HKGLUXHKGHKG

DEU

NLD USAFRAAUT HKGCHECHE

FRA

LUX

MEX

NLD

HKGLUX

NZL

GBR

NLD

HKG CYMPRT NLDHKGGBRNLD

NLD

CAN

NLD

HKGCHE USACHE

FRA

ISLVIR

USA

GBRLUX HKG

DNK

CAN

PRT

CHE

SWE

FRA

GBR

NLD HKG

ITA

CHEGBR USACANCYMTWNJPN

HKGITA

SWE SGPUSA

FRA

BEL HKG DEUHKGVIRUSA

ITA

NLD HKGSGPSGP GBRUSA JPNESPGBRGBRNORGBRCHE FINESP JPN ESPGBR DNKHKG

USA

DEUNLDGBRCHEDEULUX GBR HKG

FRA

NLD VIRGBR HKGBMU

NLD

BRA

DNK DNK

HKGDNK DEU

USA

GBRHKG

CHN

GBRLUXGBRLUX DEUIND CANHKGGBR HKGSWEGBR HKGDEU GBRHKGBEL

AUT

CHE GBRDEU BELGBR

DEU

DEU HKGGBR HKGCYPDEU NZL

ESP

NLDGBRGBRHKGGBR HKG USADEUHRVSMR

USA

HKG USANLD USA

LUX

CHEDEU USALUX HKG

NLD

HKGHKGLUX HKGJPNDNK

CYM

NLDNLDHKG HKGESP

USA

HKGLUXHKG GBR CANGBR USACYMLUX HKGNLDDNK USAHKG

FRA

ITADNK CYMHKG DEUGBRBHSHKG

DEU

HKG USAHKGUSACHEGBR SGPHKG HKGBMU LUX HKGGBRDEU

CHE

NLD

LUX

HKGNLD GBR

BEL

NOR

CHE

GBR HKGHKGIND HKGGBR HKGCHE GBRGBR KORGBR GBRGBR

CYM

NLD HKGSGPAUT

CHE

HKG

DEU

CHE

FRAHKG

CHE

HKGGBRGBR SGPFIN

DEU

HKGLUX SWEDNK DEUJPNNLD

CYM

GBR HKGHKGGBRLUXHKGINDHKGSWEHKG USAHKGHKGAUT USA USA

LUX

AUT

BEL

NORNLD USAGBRCHE USABRBCYM FRA

MYS

USAFIN

FRA

DEUNLDSWE GBR USA JPNPAN GBRISRGBRSWE HKG

FRA

DNKUSADNK

ESP

GBR CHEHKGLUX

ITA

LUX JPN

ITA

GBR HKGGBR USAITA HKGDEU

DEU

USAGBR GBR

CHE

USAGBRSWE

CHN

GBR

ITA

HKGARG

CHE

NLD FIN MUSNLDCHE

JPN

VIRGBR

NLD

HKG

DEU

TWNUSAHKG BMUGBRGBRCHE

GBR

CAN LUX

CHE

GBR

CYP

GBRCYM LUXGBRHKG

GBR

DEUGBR

AUT

DEU

GBR HKGGBRCAN

LUX

ITA HKGPRTHKGGBR PAN

BRA

NLD

NLDNORGBRLBNCANGBR FRATWNHKGGBR HKGITAIRL GBRHKG

USA

BHS

DEU

GBR

FRA

GBRHKGGBRFRA USAGBR

AUT JPN

GBR

DEUFRA

SWE

GBR

LUX

PRT ITADEU CHE IND

DEU

FRA HKGHKGHKG CYPPANIND

FRA

HKGHKGVIRGBR

BMU

ESP

JPNLUX ITA

DEULUX

LUX

GBR

HKGGBRESPNLDNLD

GBR

JPN CYMGBR SGPGBRHKG HKGGBR CHE

CHE

DNK

NOR

FRACHEBELGBRNLDHKG

LUX

CYM GBRCYP HKGGBRIND

GBR

HKGCANNLDDEU

ITA

NORITA DNKBELCYP DEU FINLUXHKGCHE HKG FRAHKGCHEHKG

NLD

HKGPOL VIR

FRA

NLDGBRGBR

NLD

FRA NLDGBR DEUCAN

GBR

HKG

FRA

GBRUSA

JPN

BELLUXCHE CANDEU HKGAUSITA CHE HKGGBR GBRDEU JPNITA IRLSGPGBR

GBR

CHEGBRUSA HKGCAN USAUSAHKG CHE

DEU

HKGGBRDEULUX USA

FRA

FRA

NLD

JPN

HKG

NOR

DEU HKG

DNK

LUX

DEU

USA

FRA

ESP

GBRHKGGBRHKG DNK USATWNGBR HKGCHE JPNCHEDEU HKGIRL

GBR

NLDBRAGBR

NLD

CHE

DEU

HKGIRLHKGHKGDEU USA

NLD

LUXGBR

GBR

USA

ITA

SWE

NLD

VIRGBRESPGBRGBR TWNCHEGBR HKGCAN HKGGBR HKG

LUX

HKG HKGGBRLUXGBR

FRA

DNK

GBR

NLD

GBR

DEU

FRA

GBRFRA ESPLUX NLDHKG HKG HKGSGPGBRGBRGBRHKGVIR

FRA

AUT USAGBR HKGGBR

CYM NLD

MEX

AUSIRL

MEXCHE

NLD

AUT

BEL

CHE

FRA

DEUDEU CAN USAUSA BEL

ITA

CYP

DEU

POLLUX USA DEUHKGUSAAUS CANHKGKORUSANLDFRANLDCHE

GBR

GBR FRAHKG AUSAUTHKGGBRNLD

DEU

USANLDGBR HKGBMUDEU USA USAHKGCZECAN HKGGBRGBRTWNGBR FRALUX GBR CANHKGCHEGBR FRA FINGBR

DEU

BMUUSAHUN HKGIRLBEL

AUT

FRA CANHRV

JPN

NLD DEU

CHE

BHS

BMU

NLDIRL HKG

USA

LUX

GBRCYM GBR NLDUSALIE

NLD

JPN

GBR

USAUSA HKGPAN USA CHEESP CAN CANCYM

FIN

AUTNLD

GBR

USAGBR FRANLD GBRDEU

LUX

DEU

NZL

GBRCHE HKGARE ESP

GBR

DNK

GBR

LUXDEU

GBR CZEUSA HKGMLTFINVIRBHS CHEVIRGBRIRL

CHE

GBR HKG

GBR

GBRNLD GBRMYSHKGMYS JPN NORGBRGBR JPNGBRGBRGBRGBRDEU

NLD

HKGLUX USATWNGBR BMUGBR DNK

ESP

SWE

CHELUX DEU JPN

GBR

ESPNLD LIE HKGNLD JPNAUT

USA

CHEGBRNLD NLDGBRAUTGBRCHE URYGBRDEU

SWE

GBR DEUNLD

LUX

DEUGBRVIR FRADEU HKGNLDHKG

NLD

HKGNLDLUX GBRITA USASWE

CHE

JPN

HKGDEU

FRA

ISR CZEHKGHKGGBR HKGBHR ESPLUXDEU NLDLUX

NLD

NLDITA GBR

ISR

ITA

SWE

HKGGBR NLDGBR SWELUXHKG HKGHKGCHEGBRNLDGBR LUX

GBR

HKGSGP HKGCHE USANLDGBRGBRDEU HKG

SWE

MEXNLDNLDDEU CHE DEUGBR HKGGBRGBRNLD ITALUX

DEU

DEU GBR

CHE

GBRGBRNLD HKGFRACYP HKGBMUGBR VIR HKGSWEIRL AUT

GBR USADEUCAN

JPNDNK

CAN

AUS

HKG NLDNLD

NLD

AUT ITA

CHE

HKG

DEU

CAN HKGSGP GTMSWE CHEGBR CHEAUTCHEGBRFIN TWNCOLHKGKNALUX HKGHKGSWE GBR HKGCHE DEUITANLDDNK PANLUX

AUT

HKGDEU

CHE

JPN HKGSWE LUXHKG ESPTWN GBRGBRMYS HKGDEUGBRFRA

CHE

HKG LUXGBRSGP HKGCANHKGCHEHKG ZAFHKG VIR

LUX

VIR HKG

FRA

IRLDNKCHE LUX

JPN

BEL

AUSUSA VIRESP

DEU

CHE

NLDDEU CAN FRA

LUX

CANDEUCYMGBR NLDNLDLUXSGP

GBR

CHE NLDNLD AUSCHEGBR ESP HKGUSAMEX CANESP

CHE

HKGCHEPOL DEUPRTGBRGBRGBR SWEUSA USAJPN HKGHKGURY DNK

USA

USA HKG CANHKGHKGESPLUX URYNLD

DEU

USACANNLDDEU

LUX

GBR

USA

LUX

HKGGBRLUX

SWE

GBR

USA

NLDDEU

MUS

GBRGBR HKG USACHE CHN CAN

CHE

ITA

GBR

DEU

GBR USA BMUGBR GBR

NLD

NLDHKG

NLD

DEULUX HKG BMU

BEL

ITAGBRGBR

ESP

GBR PAN HKGHKGHKG DEU

ITA DEULUX

BMU

TWN

USA

HKGGBR HKGHKG

ESP

MEXCYM NLDVIRGBRLUXCHEVIR NLDGBR

CHE

CANCOL

BEL

NLD

LUX

LUXCHE HKGGBRGBRGBR HKGDEUGBR FIN

DEU

DEU

GBR NLDGBRHKG ITACHE GBRGBR CYMDNK HKGHKG

AUT

FINHKGCHE NLD HKG

SWE

HKG CANCAN VIRESP

FRA

FRA

NOR

BMUHKGMYSHKG

ESP

FIN HKG

NLD

DEU USAHKG VIRITAGBRCHEUSA GBR ESPDEU KOR VIRLUXLUX GBRGBR DEU HKGGBR IRLITA USAZAFDNK AUTMLTGBRNLD HKG KOR

ZAF

HKGGBR BMUNLDGBRGBR CYMISLESP GBRAUT

AUT

NLD GRC VIRHKGCHNGBR

NLD

USAGBR CHEMUSGBR

CHE

ISRGBR ESP HKGGBR

DEU

TWNGBR ITA

NLD

HKG

CHENLD

LUX

GBR

VIR MUS HKGCAN USAGBRUSA HKGESPGBR FINGBR HKG

LUX

IRL DEUGBR HKGUSAGBRNLDGBR TWNAUT ITA HKG

DEU

GBR

HKGLUX HKGLUXVIRGBR LUX

NLD

DNKGBR HKG

ESP

SWEGBR HKGLUXJPN

ITA

SWE

FRA

HKG

CHE

GBR

USAFINLUX HKGDNK GBRLUXGBRGBR HKGSWE DEU

GBR

GBRTWN

LUX

HKG HKGGBRGBRGBRHKG DEUGBRGBRDEU VIRGBR DEUCHE GBRSWE GBRGBR GBRTWNGBR FRAGBR JPNHKG HKGHKGFRA CANCHEGBR HKGGBRHKGHKGNLD

CHE

DNK GBRGBR CAN

USA

LUX HKGHKGNLDDEUDEUSWE

CHE

IND DEUHKGHKG SWE

FRA

USA

NLDDEUGBR

PRT

ESP CANDEUGBRVIRGBRLUXGBR DEUHKG HKGGBRCHE USAGBR HKGGBRHKGUSA USA

BEL

ITA

ESPCYM

CYM

DEU DNKLUX SWENLD HKGGBR CAN

GBR

GBRGBR HKG JPN

DNK

HKG

ISL

HKGNLDESPGBRGBRGBR HKGIRLGBR

LUX

PAN BHSHKGLUX HKG HKGGBRPRT HKGAUT HKG

FRA

DEU

HKG

NLD JPN

CAN

CHE

FRA

CAN

NORFRA

GBRSGP HKGCHEHKG

DEU

FIN

NLD

GBR

BEL

HKG

DEU

ESP

GBR GBRHKGHKG GBR

DEU

HKGGBRCZE HKGNLD

CHE

HKGUSASWEGBR HKGCANCHEGBRFRA HKG TWNCANGBRHKGGBR GBR HKGTWNDNK LUXNLDESTDNKLUXESPGBR CYMHKGDEU

LUX

CHE DEU HKGUSABMU QATDEUNLD HKG USA

USA

GBR

BMU HKGSWE FRABEL GBRDNKUSASWE SGPKORDEU CHEPANGBR HKGDEUHKG

AUT

FRALUX HKGGBRFRA

DNK

NLD DNKLIEGBR VIRGBRITA IRL

SWE

USA

GBRGBR VIRHKG HKGGBRUSA

ESP

POL BEL USACHE HKGSWE HKGGBRGBRNLDGBRGBR DEU

GBR

HKGMLTSGP GBRHKGUSAHKG

JPN

GBR VIRNLDGBRNLDCYM USAGBRGBR LUX

NLD

NLD

CHE FIN

LUX

NLD

FRANLD DEU CHNFRA

USA

SWE

BMU

SWE

VIRVIR LUXCANCYMGBR CYM CAN

MEX

ESP

DEUCAN TWNFRAHKG

NLD

LUX TURDEU GBRGBR

FINGBR

JPNBMU CANGBR HKGGBR

GBR

CANLUXGBR GBR

DEU

LUX AUS

NLD

AUT

IND DEUDEU

DNK

LUX

CYM

USACANNLDGBRPRT

GBR

DEU

FRA

HKG USADEUSWEGBRDNK

SWE

HKGITA

USA

ITASWE DNKNLD AUS

JPN

NLD HKGLUX SGPGBR USALUXGBR GBRHUN HKGNLD IRLPOLBELGBR DNKGBRCHE IRLHKG HKGHKGLUX ESP JPNDEU USAAUT

PRT

IRL USATWN NLD

NLD

LUX SWEHKGGBRFRA HKGSWEGBR HKGNOR

USA

AUT DNK

SWE

HKGNLD NLDMUSHKG

SWE

FRA USAPANSWE

GBR

GBR

SWE

NLD

HKGGBR HUNFIN

FIN

URY

LUX

USASWESWE

DEU

KOR

USANLD DNKBMU

ITA

GBR

DEU

HKGGBR HKGLUXNLD HKG HKGDEU

NLD

FRA

HKGGBRHKG DEU

CHE

CHE

HKGCANCHE FRACHE

USA

ITA

FRA

HKGNLDTWNNLD

GBR

USA

DEU GBR GBR

CHN

GBRLUXGBRLUX HKGDNKGBR CANGBRESP HKGGBR HKGGBR HKG

NLD

NLD FINCANUSAHKG

NLD

DEU

DEU

HKGGBRLUXGBR HKGNLD KORGBRGBRGBR GBRHKG

USA

VIR NLDZAFNLD

PRT

DEU JPN

DEU

HKGGBRGBR HKG FRALUXBHSDEUCHEGBRHKGGBRLUX

BMU

GBR PANHKG DEU

DEU

NLD

NLD GBR HKGHKGCHEAUS USALUXGBRCHE

BMU

VIRHKG CHEVIRTWN VIRMLT USA

NLD

CAN DEUHKGDEUGBR HKGNLDVIRGBRSWE NLDNLDNLD HKGESP

DEU

BMUUSADEU

NLD

GBRUSA NLD

USA

SWECHE

GBR

BMUGBR DEUNLDJPN USADNKLUX HKGGBR FRAITALUXHKGNLD

LUX

HKGMEXCYM HKG

CHE

HKGNLD HKGHKG VIRHKGSWE

NLD

DEU

CHE

CHE

HKGHKG SGPITAITA USACHE NLD

GBR

USA

ITA

GBR

SWE

CHN

USA

DEU

CHEHKGUSALIE

CHE

GBR CAN

FRA

HKGESPDNK NLDGBR AUT AUSHKG HKGGBRBEL

GBR

ITA

CANNLDHKGGBR

GBR

GBR TWNBELDEULUX HKGGBR KOR SWE

SWE

GBR

BEL

ITA AUTNLDNLD DEU GBRVIRSGP

ITA

ESPCHE NLD

CHE

GBRFRA LUX GBRCANDEUNLD ESP USAUSACHE

DEU

CHE BEL USA

DEU

GBRFRA VIR HKG DEUGBR

FRA

GBR GBR

FRA

GBR

USA

GBR

FRA

BEL

USA

SWE

FIN

CYP

LUX HKGHKGCHE USA

FRA

DNKCHELUX NLDBRA DEUHKG MUS

CHE

NLDLUXSMRCHE SWEGBRGBRDNK

FRA

ITALUX HKG

BEL

HKG NZL

DEU

CHE DEU HKGNZLDEUGBR MEXNLD HKGUSA

AUT

GBRNLDGBR SGP

GBR

DEUGBRDEU HKGNLDDEU ESPHKGPOL FINPOLGBRCHEGBRNLD TWN

FRA

GBR

DEU

AUT HKGGBR CANHKGAUTHKG

SWE

USA CYM HKGDNK

NLD

DEU

USANLD

NORCHE

DEU

FINESP HKGGBR

FIN

HKG CHEGBRGBR USALUX HKGCOL CHLGBRNLD ITA

SWE

VIRNLD ESP TWN

DEU

GBRHKG USAAUTGBRESPITA

CHE

LUXGBRLUX USA HKGLUX

USA

MEX GBR

CYM

HKGBELDNK

GBR

USACYPFRA TWNFRAGBR CANLUX HKGCANUSAGBR SGP

SWE

CHECHE LUX HKG BELDEUFRA HKGCHE JPNNLDNORGBR LUXHKG HKGCZEHKG HKGITA HKGGBR ISLITA

DEU

NLD

GBR

SWE

HKGLUX HKGHKG

LUX

CHENLDGBR

GBR

NLDGBRGBR CHEUSAAUT USAGBR HKGHKGHKGHKGGBR HKGSWEHKG

AUT

LUX HKGJPN

JPN

NLD

IRL

PAN

DEU

HKG

NLD

HKGJPNDEUNLD USAITA HKG

DEU

GBRGBRCANGBR USADEU USA

DEU

HKGSGP

BMU

HKGJPN LIEGBRDEU DEUJPNSWE HKGLUX HKG

SWE

DEUHKG

USA

LUX GBRGBR CAN HKGVIR CYMCYMNLDUSA JPNNLDGBRUSAJPNNIC HKG

NLD

IRLGBRGBRDEU KORDEU HKGAUT CHE HKG BMUCHE VIR

GBR

USA

DNK

GBR ITAUSACHEDEULUX USAGBRDEU CANPANNLD AUTCHE BRAJPN ITALUX HKGGBRCHE

DEU

HKGLUX HKG

CHE

GBRNLDVIR

CHE

DEU GBR NLD HKGGBR USA

GBR

GBR ARE CAN VIRITAITAGBRGBRPOL

CHE

GBR

CHE

NLD

HKGGBR HKGHKGNLDNLD CANUSAGBRGBR USALUX VIRITA GBR USADNKFRA GBR

FRAITA

GBR

USA

NLD JPNFRA

FRA

HKGSGPGBR ITA CANHKGBEL GBR

USA

CHE

DEUFRA GBR CAN

NLD

CHEHKGGBR HKGLUX JPNHKGCHE

SWE

HKG

JPN

HKGFRA SGP ISRNLD HKGDEU NLDHKG

DEU

GBR

CHE

NLD

FRAJPN

JPNGBR

DEU

USAGBR HKGGBRCAN NLDHKGGBR USA

GBR

USA

IRL

VIR SWEGBR GBR

SGPDEU

GBR

FRA

AUTTWN NOR

BEL

GBRCHE VIRTWN NLDTWN

CHE

GBR VIRDNKCHE

CHNCHE ESP

DEUFIN ESPDEUITA

DEU

CYM

SGPCHE HKG NLD

LIE

USADEUGBR HKG

ITA

GBRBELSWE DEU

NLD

PANGBR DEUNLD HKGHKG

NLD

HKGHKGDNKGBRAUS DEU

LUX

FRAHKG VIR HKGHKG

PRT

NZLHKG

JPN

LUX HKG

CHE

NLDUSA AUSGBR

DEU

GBR

LUX

LUX

GBR GBRLUX DEUGBRAUT CZELUXGBR HKGHKG CANHKGHKGDEUGBR

USA

DNK USA USAAUT

ITA

HKG

SWE

GBR HKG JPNHKGLBN NLDHKGPAN GBRMUS MEX

GBR

CHNDEUSWE BMUITACYPUSAHKG DNKBELUSA AUS

GBR

SAU

AUS RUS HKGGBR HKGNLD DNKNLD

LUX

GBR BMUHKGGBRIND HKG

FRA

NLDGBR

GBR

MUSITAGBR TWN

NLD

MEX

HKGIRL

DNK

DEU

MEXBMUHKGHKGGBRGBR GBRGBR CANLUXSVN

DEU

LUX

SGP CANGBR JPNGBR HKGGBRGBR GRLNLD

SWE

LUXKORGBR

AUT

GBR HKGGBR GBR BMUGBR JPNJPNGBRGBR

NLD

GBR JPNHKG

USA

GBRGBR

GBR

USA

ITA

ITAHKG HKGHKG

NZL

DEU USA

FRA

HKG

CHE

CHE

HKGGBRAUT CHEVIR CHEGBRPRT CHENLD ZAFGBRITAESP

USA

LUXKOR

MEXISL

GBR DEUNLDDNK AUTLUXUSA HKGCYPPRT DEUGBR

DEU

LUXFRANLDESP

ESP

DEU DEUSWE FINNLD CHEDEUFRA KORVIR CAN

LUX

JPN

NLD

TUR

FRA

HKGDEU HKGDEUFRALUX

CHE

NLDLUX HKGAUT CAN HKGSWEGBR CHEJPNLIE JPNFRAGBR

BRA

FRA ISRIRL

FRA

PRTIRL

USA

HKGGBR USA

CHE

CANGBRAUT

GBR

ESP MYSGBR HKGGBRAUTFRAGBRGBR

AUT

HKGGBR CANGBR ISRNLD

LUX

NOR

GBR

ESPDNK

FRA

DEUGBR HKG

ITA

HKGGBR

GBR

GBR HKGGBR PAN HKGGBRGBR FIN NLDHKGVIR

GBR

JPN

LIE

USAHKG

USA

SWE

NLD

GBR

DEUSWEDEUHKG

ITA

HKG

HKG

DEUNLD HKGFRA USAHKG HKGHKGDEUPRT

ITA

DEU

CHE

USAHKGNLD USAHKGTWNIRL

GBR

GBR FRO

USA

SGP HKGLUXGBR CAN MYSNLD

USA

DEUDEUESPITALUX NLDTWNGBR SGPGBRLUXDNK

JPN

CYPLUXSWE TWN

AUS

USA USA

DEU

LUXHKG FRA SGPAUSNLD

SGP

JPNITAGBR GBR

NLD

LUXSWE CHEGBRGBR VIRVIR JPNGBRLUX NLDUSA

USA

GBRLUX JPNAUSGBR JPNGBRGBR TWNBEL

GBR

GBRFRA HKGFIN HKGCAN VIR

INDSWE

GBR GBRDEUCHEHKGSWE

FRA

NLDHKGGBRGBR CHE PRTFRA

GBR

ITA DEU

USA

GBR NLDCHE MUS HKG

GBR

JPN

JPN

GBR

DNK

GBRHKG

VIR

GBR HKGGBR HKGAUS

ESP

HKGLUXIRL VIR FRAHKGHKG VIRDNK

LUX

JPN

CHE

BEL

DEU

DEU USA

BEL

LUX

FRA

GBRCHEGBR CANNLDLUX USAGBRGBRAUT NLD HKGCHE NLDLUX HKGGBRLUX USA USAUSA

SWE

NLDHKGFRAPRT

DEU

HKGDEULTULUX HKG

ITA

DEU HKGBHS FINGBR NLDCHE HKGHKGHKG BMU POL

DEU

CYM

HKGHKG

NLD

GBRLUX HKGHKG

FIN

DEU

ESP

FIN

SWEGBRNLDGBRLUX GBR

DEU

TWNAUT

USA

HKGBEL

FIN

SWE

HKGESP

GBR

NORESPLUXCHEHKG JPNGBR

CHE

HKG

SWE

GBRHKGSMRBELTURGBR

NLD

JPNVIR VIRLUX HKG

BRA

DEU

GBR

ITAGBR DNKDEU

DNK

GBRGBR NLD

NLD

INDGBRGBRGBR BELHKGGBR HKGGBRHKG

ESP

GBR

USAUSA USACHE BRACHE

CHE

HKGAUT

ESP

GBRFIN

JPN

HKGLUXUSA HKGDEU COLNLD

NLD

URYCHE

GBR

ESPCYM

NLD

TUR USAGBRURY HKGHKGHKG HKG

DEU

ESP

GBR HKGGBR HKGFIN USAGBR CHE HKGGBR HKGHKG GBRDEU

FRA

SWE

LIE HKG

SWE

HKGTWN

DEU

FRALUXITASWECHENLD CYMDEUVIR VIRHKG

AUT

JPN

MUSHKGDEU HKGLUXHKGUSALUXGBRESP CAN CANESP

JPN

NOR HKGGBRGBR HKGDNKHKG

NLD

FINGBRGBR HKGVIR TWNHKG BRAGBRCHE GBRGBRNLD

DEU

GBRGBR GBRDNK

FRA

BMUMUS CHEGBR

GBR

SWE DEUVIRGBR MUSHKGLUX ISRLIEUSADNK PER HKG

DEU

CANGBR

CHE

GBR CANUSAITA

AUS

NLDCYM DEUGBR NLD

FIN

GBRCYP HKGIRL

CHE

HKGHKGGBR GBRGBR HKGBEL IRLGBR VIR SWESWE LUXJPNHKG USALUX HKGCHEITADEU CAN HKG

FRA

CYM

CYMGBR DEU

ITA

CHE

GBRGBR SWEGBRPOL

CHE CHE

HKGCHE

DEU

BEL HKGUSA MYSDEULUX USADEU HKGLUX ITANLDSWE GBRHKGCHELUXROM GBRGBR NLDNLDSWEGBRAUTCYMLTUCHE

FRA

JPN

JPNVIRGBR HKGSWEEST MYSCYMHKG USAJPNCYM

DEU

LUX

SWE

HKG

FRA

TCA HKG HKG

JPN

CHENLD

NLD

JPN

FIN NLD DEUDEU

LUX

USACHEDEUHKGDEUCHEGBRLUX HKG USACYP

CHE

HKGHKGGBR HKGCANLIE

NLD

USA

ISL JPNUSAGBRGBRGBR GBRPRT

FRA

HKGHKG

FRA

CHE HKGHKGHKGGBRNLDGBR HKG

PRT

CHE

NLD

GBR HKGDEU USADEU CHEGBR

JPN

HKGLUX USAGBRGBR CANGBR

ITA

ZAF GBR HKGNLDGBRNLDGBRGBRNLD GBRNLDAUT HKGUSA HKGGBRPANUSA NLDHKGDEUGBR ITA CANSWE

GBR

LIE SGPNLDIRLNLD SWEGBRPRT

CHE

DEUNLD JPN

USA

HKG

ITA

DEUJPNSGP

LUX

CYP ESP

GBR

USAPAN USACHE HKGGBR

LUX

AUS

DEU

DEU

CHE

DEU HKGVENGBRGBR HKGGBR GBR ESPGBR JPNDEUNLDLUXHKG CANGBR USAGBRUSA VIR HKGGBR

CHE

USAUSA

GBR

FRA

DNKNLD LUXCHE

NLD

BEL

MYS

ARELUX

BRA

DEUVIR

DEU

GBRFINFIN GBRHKGCYPCHEGBRHKGLUX

CYM

HKG

DEU

FRA

USA

JPN

GBRESPCHE

HKGLUX

FRA

SGPESPCHEIRL HKGJPNGBRHKG HKGHKG

USA

HKGGBRSWE

DNK

SWEHKGGBR FRAESPGBR HKGTHAUSASWE

CHE

NLDISR

NLD

DEU

DEU

CYM

NLDUSABHSGBRGBR GBRGBR SGPKORLUX HKGIRLLUXBELUSA

GBR

USALUXGBR

FIN

CYM SWECHE SGP

DEU

SWEDEUPRT GBRDEU

DNK

HKGGBR HKGCHEESPPANGBR

DNK

TWN

DEU

HKGIRLGBRGBRGBREST

JPN

FRA

ZAF

NLD

DNK

DEUDNKESPCHEGBRCHE DEUGBRGBR HKGGBR

DEU

SWEGBR BMUHKGNLD BLZNLDUSA

GBR

CHE

GBR

FRA

LUX GBRHKGFRALUXNLD SWE VIRGBR

CHE USA

DNKSWE HKGGBR JPNGBR

NLD

SWE

GBR

NLD

SWE GBRGBR USACANLUX SGP

CHE

FRASWEGBR GBRGBRDEU HKGDEUNLD DEU HKGHKG

USA

DEU

ESP

ESP AUSNLD FIN

FRA

HKG CHEDEUGBRUSA HKGSGP

LUX

JPN

CHE

GBR

FRA

GBR

DEU

GBR

DEU

GBR

ESP

GBR JPN

CHE

SWEGBR HKGIRL HKG

NLD

SGP JPNGBR GBRLUX

GBR

ITA

GBRDEU DEU TWNGBR JPN

NLDCHE

IRL CHENORDEU HKGNLDGBR HKGGBR HKGHKGBRBSWE

MYS

HKG DNKCYPGBR

DEU

GBR

CHE

ISR

HKGITAHKG DNKDNKGBRUSAFIN GBR

DEU

USA

LUX

JPN

CHE

GBR NLD HKGGBRGBR HKG

GBR

JPN USADEU

FRA

DNK

LUX

FRA

HKGHKG USAGBRAUTITA

USA

FRAHKGCHEITA

GBR

AUT HKGSGP

BEL

USALUXHKG SWENLDDEU

DEU

GBRHKG

CHE

PANLIE HKG

USA

DEU HKGGBR GBRAUTLUX DNKITAAUT HKGGBR MUS

FRA

CHE

SWE

CAN

GBR

JPN

USAGBR USALUX

ITA

ITADNK HKGCHE ITAHKGFRASWE GBRGBR

DEU

CHNNLD AUSNLD HKGHKGGBR

BMU

DEUGBR USAGBRGBR GBRNLDGBRGBRUSAGBR DEUESP HKGLUXGBRCHE

NLD

FRA

USAGBRGBR

CHE

DEU HKG BMUAUTHKG

NLD

ITAGBRGBR

HKG

HKGEST HKGTWN BELSWE

DEU

LUX SAUHKG FRAGBR BELPOLLUX HKGHKG NLDHKG CANGBR HKG HKGHKGSWEDEU NLDGBRJPN

ZAF

USA CANGBR HKGNLD DNKGBR

DEU

USAKWT

AUT

GBR SGPDEUUSA

NLD ESP

HKGESPUSA USAGBRSMRITA HKGNLDITA

AUT

GBR TWNAUTGBR

NLDLUX

DEUFRALIE HKGDEU

FRA

HKG HKG

FIN

DNK

FRA

LUXGBRGBRCHECHENLD ITA FRAGBRESP GRLHKG HKGCHE DEUGBR

FRA

IRLNLD CANUSATWNDEUIRLCHEGBR

CHE

FIN USATTO HKGSWEGBR HKGGBRHKG HKGGBR AUTHKGGBRLUXGBR NLD

SWE

NLD

DEU

CYM HKGGBRDEUHKGGBRCHEIND CYMPOL

BMU LUX

NLDIRL HKGZAFGBR NLDMLT HKGCYMAUT USA HKG

FRA

DEUDNKVIR CHE

NLD

JPN HKGGBRNLD

LUX

GBR NOR

DNK

DEU

USAGBR

ITA

NLDNLD

LUX

NLDHKGSWE USA

JPN

GBR ITAPAN BEL HKG USALUX USAITA HKGJPNGBR BELIRL TWN

AUT

BELCHE HKG HKGJPNGBRGBRGBRUSAGBR HKGNLDFRAHKG

NLD

CHE CYM

DEU

LTU HKGUSAGBR GBR

ITA

BEL VIRBRALUXGBR PANGBR GBR HKGDEUGBR HKGUSACHEGBR DEU CHNITA USA AUSHKG HKG CAN

FRA

GBRITACHE NORGBR HKGFINHKGDEU HKGJPN HKG

BEL

NLDNLDHKG FRA

GBR

HKGGBR HKGHKGESP JPNHKGHKGBEL

GBR

CHEUSAGBRSWE USAGBR NLD

SWE

RUSDEU

BEL

DEUCHECHE NLD

DEU

HKGGBRGBRGBRHKGGBR GBRDEUHKG GBRFINGBRGBRFIN ITADEU HKG

ESPDEU

NLD

THA USA

CYM

DEUGBRNLD

NLD

CHE

DNKCANLUX FIN

DEU

GBRPRT BEL USALUXLUXCHE VIRNLD

LIE

AUT HKG

ITA

DEU ESP

DEU

GBRGBR HKGAREGBR FRABELDEUPOL SWELUX

FRA

HKGCHE

USA

GBRNLDGBR HKGNLD SWE

GBR

IRLGBRCHE

GBR

VIR

DEU

HKGGBRGBRUSALUX LUX ISR USALUXITA

FIN

BEL

SWE

CHE

ITA

VIRJPN

CHE

HKGBMUGBR GBRGBRBMUJPNLUXNOR JPNHKG HKG

FRA

HKG CHEHKG DEUISRSWEUSACHLTWN

DEU

USA

AUT

USA USANLDCRIDNKLUXLUX VIR

DEU

HKGGBR BELESP DNKLUX BHRCHE

GBR

SMRFRA GBR VIRDEUGBR

DNKSWE

GBRDEUDEUCHEHKGNLD FINUSASWE

LUX

HKG ZAFITA HKGNLDGBR HKGHUNLUXDNK

BEL

HKG

USA

HKGDNKGBR

NLD

USAGBR HKGDNKGBR HKG

FRA

DEU

HKGDNK LUX

ITA

BMUCHE HKGPRT HKGGBR PANDEUGBR HKGUSA

CHE

GBR NLDITANLDCHE

USA

USA

GBR GBRHKGCHE

CHE

HKGGBR FRAHKGGBR SGP BEL NLDHKGHKGVIRLUXSWEFRA IRL JPNGBR NLDESPNLD SWECAN

JPN

USA

GBR HKGDEUNLDSGPGBRNLDGBRGBR USA BRANLDCHE USAGBRGBRPRT

GBR

USA

CHE

HKGLUXHKG LUXGBR HKGBEL HKG

SWE

GBR MUSDNKCHE USA HKGGBR

NLD

GBRUSA JPNHKG CAN

ITA

GBRSGP

GBR USAUSAGBR USA TWNGBRLUX GBRSGPHKG HKG NLDUSAGBR PERCHEFRA

ITA

CHE

FRA

BEL

AUT

AUT

LUX

USA

NLDGBRLIE

USA

AUSHKGGBR JPNGBR

GBR

ITAVIR

FRA

DNK BEL BHRHKGGBR TWNUSA

GBR

JPN

BEL

CANCYMTWNBHSGBR GBRGBR

DEU

FRA

ESPBELFRAGBRGBRHKGLUXGBR HKGHKGLUX ITA HKGGBRNLDGBRSGP HKGITAGBR PANBMUGBR HKGHKG

CHE

JPNNOR

USAGBR

DNK

SWE

USA

USA USAGBR ITA

SWE

ISR GBR

GBR

HKGGBR DEUHKGGBR

FRA

HKG USAGBRDEU MACCAN PAKGBR GBRNLD HKGDEU IRL

GBR USA

ITA USALUX HKGGBRLUXAUTBEL GBRGBRGBR HKGLIE

CHE

VIR LUX

BEL

GBRSGPCANDEU

CHN

CHEAUTDEU USA

DNK

ITACHEHKG HKGTWNPRT DNKNLDHKG TWNZAF

DEU

DEUDEUFRA USA

AUS

MUS

GBR

USA

GBRAUT

FIN

HKGGBRGBRDEUGBR

VIR

HKGGBR

CHE

GBR HKGGBR HKGHKGGBR DNKVIRGBR LUXLUX HKG

DEU

FRA

DEU HKGHKGPRT LUXLIE HKGGBRSWEUSA DEU NLDHKGDEUCHEGBR HKGGBRHKG HKGGBRLUX CHE

NLD

HKGGBR DNKDEU HKG USAUSA BRA

FRA

HKG CANNLDGBRGBRFRADNKGBR FIN HKGHKGGBRGBRNLD HKGGBRDEU HKGHKGUSAGBR NLDSGP DEU HKG CANISR DEU HKG

NLD

DNK

NLDLUX HKGGBR

FRA

HKGNLD

DEU

GBR

GBR NLD JPN

LUX

GBR HKGHKG

NLD

SWEDEUDEUNLDDNK

LUX

HKG USAAUTNLDGBR DEUHKGGBR HKGGBR

CHE

VIRHKG GBRLUX DEU

DNK

COL USA

DEU

NLDGBR

SWE

SWE JPNGBR

DEU

HKGHKG

AUT

SGPGBR GBR

CAN

DEU SGP AUTSGP NORHKGROMLUXGBR DEU HKGNLD

DEU

AREVIRUSANLDSWECHESWE CANESP

CHE

VIRUSA

LUX

USA

CHE HKGHKGCHE BELLUX

ARE

HKGHKGHRVGBRDEU JPNPANSWEIND DEUGBRHKGUSA

CHE

HKG

SWE

GBR JPN USAAUTFRA

SWE

DNKGBRHKGDNK NLDHKGUSA

NLD

SWEGBR

AUT

ESP

ITA

GBR

FIN

USADEU CHE HKGCANGBR CYMSGPGBR

ITA

SWE HKG

ESP

CYMDEUDEUGBR BMULUX

CHE

DEUGBRCHE HKGGBR BMUAREUSA HKGIND

DEU

GBRLUX

USA

AUT

VIR

NLD CANKORHKG HKGUSA

LUX

HKG

CHE

BELBMUSGP HKG ITAGBR CHEFRAFRA HKG

FRA

AUT DEUHKGPRTGBRLUX

JPN

PANDEU JPNCHE

SWE

CHE FRAHKGLUXPRT HKGLUXPRTZAFGBRAUT ITASGP FRADEUDNK HKGBMUHKGNLD USAUSAGBRESPBELBEL CANNLDGBR NLD VIRIRLHKG DNK CANDNK

ITA

DNK

GBR CHE CANSWECHE AUSBMU SGPHKG

DEU

DEUITA FRAMEXITA HKG

DEU

CHEHUN CANGBR

DEU

MUSDEU GBRMYSHKGHKGNLD GBR

FIN

LUX USACANUSA USASWE HKGHKG AUSHKGSWE DNKITAHKGVIR HKGHKGGBR USAITAHKG

USA

USA

GBR HKGCHE USAHKG

USA

AUS

BEL

NLD

CAN

GBR

FIN

BELDEU

DEU

IND

GBR

LUXGBR HKG

FRA

LUX FRA

NLD

ITAVIRGBRDEUGBRNLD

GBR

ARGHKGHKGDEU HKG

NLD

USA USA

FRABEL ESPNLDLUXLUX

GBR

NLD

FRA

HKGVIR

01

02

03

0co

un

trie

s_

su

bs

0 2 4 6Log Employees

The figure shows the relationship between the size of MNC (horizontal axis) and the number offoreign countries they are active in (vertical axis). In the scatterplots, each observation is an MNC,labeled with the ISO3 code of the country where its headquarters is located. The left panel measuresthe firms’ size by the total number of subsidiaries it has (both domestic and foreign), while the rightpanel uses the total employees (both in domestic and foreign plants).

10

Figure B2: Distribution of reported SIC codes by plant, per country

01

23

4

me

an

of

nu

mb

erin

d

BLRBMUBRACOGCRIDZAEGYETHFINFRAGABGINGNQHKGHNDJORKENKORKWTMARMCOMKDMOZMRTNGAOMNPERQATSDNSLESURSYRTUNTZAUGAYEMJPNNORITAECUTWNGTMMEXCOLARGIRLSENURYBOLCHNCANINDAUSARECZEVENCMRLUXNICTTOESPGHABELUSAGBRNLDPHLTURCHLMYSDNKBGRESTPAKPANSWEDEUSGPIDNSAUTHAROMSVKPRTABWASMBGDCIVCYPDOMGEOGRDVNMZARAUTIRNLBNSLVNZLBENPNGRUSLIECHE

0.2

.4.6

.81

ABWASMBENBGDCYPGEOGRDZARCHENZLLIERUSAUTROMDOMPRTBGRCIVESTIRNPAKPANPNGCHLMYSDEUTURTHADNKVNMPHLESPSVKSWECMRLBNLUXNICSLVTTONLDIDNGBRSGPSAUUSABOLBELARECHNGHASENURYCANCZEVENAUSGTMINDARGCOLECUIRLTWNMEXNORITAJPNBLRBMUBRACOGCRIDZAEGYETHFINFRAGABGINGNQHKGHNDJORKENKORKWTMARMCOMKDMOZMRTNGAOMNPERQATSDNSLESURSYRTUNTZAUGAYEM

1 2 3 4 5 6

The figure describe the distribution of number of industries reported by establishment in the sample.The left panel shows the average number of reported industries across all subsidiaries per country,while the right panel shows, per country, the percentage of firms reporting one, two, three, four,five or six industries.

11

Figure B3: Histogram of SIC codes in the sample

0.0

02

.00

4.0

06

.00

8.0

1D

en

sity

2000 2500 3000 3500 4000sic1_est

The figure is an histogram of the SIC industries reported in the dataset. Each bin representsthe frequency of a particular SIC code within the manufacturing sector. Notice that the SICclassification is not fully continuos, what explains the zero values in the figure.

12

Figure B4: Number of different industries Vs. MNC size

20

40

60

80

Lo

g M

NC

diff.

in

du

str

ies

500010000Log MNC size (affiliates)

The figure plots the relationship between MNC size and total number of (different) industries theMNC is active in through its foreign affiliates. The figure reveals that larger MNCs (measured interms of number of subsidiaries) tend to make a larger number of different products.

13

Figure B5: Definition of Horizontal and Vertical

The diagram describes the methodology used to classify foreign subsidiaries as horizontal expansionsbased on their reported economic activity vis-a-vis the economic activity of the MNC in its homecountry.

14

C O*NET knowledge intensity measuresThis section describes characteristics of the knowledge intensity measures basedon Bahar (2018). Figure C1 presents the distribution of the knowledge intensitymeasure used in the paper: experience plus training (based on experience pluson-site and on-the-job training figures of workers in each industry). As opposedto the R&D investment based variables used in the literature, the distributionof the O*NET based variables is smoother, and behaves more like a normalprobability density function. Figure C2 presents the same graphs limiting thesample to manufacturing industries only.

[Figure C1 about here.]

[Figure C2 about here.]

Tables C1 presents the top and bottom ten products in the manufacturing di-vision (SIC codes 2000 to 3999) ranked by the knowledge intensity measure.

[Table C1 about here.]

C.1 Advantage over R&D MeasuresI find this measure correlates positively with other (proxies of) knowledge inten-sity measures used in the literature, such as the average R&D share of sales perindustry. Correlation coefficient is 0.13 with R&D intensity computed using theCompustat dataset and compiled by (Keller and Yeaple, 2013) and 0.20 withR&D intensity computed using the Orbis dataset and compiled by Nunn andTrefler (2008), using manufacturing NAICS 4-digit industries.35

The R&D based measures, however, have three main shortcomings that couldgenerate significant biases. First, these measures are concentrated in lower val-ues and have a long tail, as shown in Figure C3. This is because most firmswithin those industries have no R&D investment whatsoever. For these indus-tries in the lower end of the distribution, the intensity of their knowledge isalmost indistinguishable. Second, since these measures are computed by aver-aging across each industry the R&D share of sales reported by a (random ornot) sample of firms, they are likely to favor industries in which larger firms aremore prevalent. This might happen in industries for which the barriers to en-try are higher, and not necessarily knowledge intensive industries. Third, R&Dinvestment might not be equally accounted for across all industries.

[Figure C3 about here.]35It also correlate positively with other less popular measures that could proxy for knowledge

intensity or complexity. The correlation coefficient with the share of non-production workersin total employment, from the NBER-CES Manufacturing Industry Database (Becker el al.2013), is 0.68. Similarly, the correlation coefficient with the Product Complexity Index, de-veloped by Hausmann el al. (2011), is 0.49. These correlations were computed using SIC 4digits codes.

15

The O*NET based measures solve these issues. As shown in Figure C1,their distribution is smoother and closer to a two-tailed normal distribution.36Moreover, our measures do not rely on a sampling of particular firms, and theyuse the same standardized measure for all industries.

36According to the Skewness/Kurtosis test for normality, we cannot reject the hypothesisthat these measures are normally distributed.

16

Figure C1: Histogram O*NET-based KI (All Industries)

05

10

15

Pe

rce

nt

20 40 60 80 100KI (Experience+Training)

The figure shows the fitted distribution for the computed “experience plus training” O*NET-basedknowledge intensity measures for all industries. Industries are defined in SIC 1987 4-digit industries.

17

Figure C2: Histogram O*NET-based KI (Manufacturing Only)

05

10

15

Pe

rce

nt

40 50 60 70 80KI (Experioence+Training)

The figure shows the fitted distribution for the computed “experience plus training” O*NET-basedknowledge intensity measures for manufacturing industries only. Industries are defined in SIC 19874-digit industries.

18

Figure C3: Distribution R&D Measures

01

23

4D

en

sity o

f N

TR

DIn

t

0 .5 1 1.5 2Nunn & Trefler R&D Int

0.2

.4.6

.8D

en

sity o

f K

YR

DIn

t

0 2 4 6 8 10Keller & Yeaple R&D Int

The figure shows the fitted distribution for the R&D Intensity measures used in the literature.The left panel corresponds to the industry-specific R&D intensity computed using the Compustatdataset and compiled by Keller and Yeaple (2013) and the right panel corresponds to the samemeasure computed using the Orbis dataset and compiled by Nunn and Trefler (2008). Industriesare defined in NAICS 4-digit industries.

19

Tab

leC1:

Top

andbo

ttom

10man

ufacturing

prod

ucts,r

ankedby

KI

Ran

kSIC

Nam

eV

alue

Ran

king

byExp

erie

nce

+Tra

inin

g,Top

101

3669

Com

mun

ications

Equ

ipment,NEC

82.92

23663

Rad

ioan

dTelevisionBroad

castingan

dCom

mun

ications

Equ

ipment

82.92

33661

Telepho

nean

dTelegraph

App

aratus

(exceptconsum

erexternal

mod

ems)

81.45

43677

ElectronicCoils,T

ransform

ers,

andOther

Indu

ctors

79.97

53676

ElectronicResistors

79.97

63678

ElectronicCon

nectors

79.97

73675

ElectronicCap

acitors

79.97

83671

ElectronTub

es79.97

93672

Printed

Circu

itBoards

79.97

103674

Semicon

ductorsan

dRelated

Devices

79.97

Ran

king

byExp

erie

nce

+Tra

inin

g,B

otto

m10

459

2013

Sausages

andOther

Prepa

redMeatProdu

cts(excep

tlard

mad

efrom

purcha

sedmaterials)

36.89

458

2011

MeatPacking

Plants

36.89

457

2411

Logging

39.79

456

2077

Animal

andMarineFa

tsan

dOils

(animal

fats

andoils)

41.39

455

2053

Frozen

BakeryProdu

cts,

ExceptBread

41.53

454

2045

Prepa

redFlour

Mixes

andDou

ghs

41.53

453

2098

Macaron

i,Sp

aghetti,Vermicellian

dNoo

dles

41.53

452

2051

Bread

andOther

BakeryProdu

cts,

ExceptCoo

kies

andCrackers

41.53

451

2015

Pou

ltry

Slau

ghtering

andProcessing(pou

ltry

slau

ghtering

andprocessing

)42.72

450

2052

Coo

kies

andCrackers(unleavene

dbreadan

dsoft

pretzels)

45.04

The

tablepresents

thetopan

dbottom

10man

ufacturing

sectorsrank

edby

the“exp

erienc

eplus

training

”O*N

ET

basedkn

owledg

eintensitymeasure.

20

D Proxying for demand: knowledge transmissionin the proximity-concentration trade-off

The results on the determinants of horizontal expansion are not fully satisfactorysince they do not account for the demand factor (i.e. it is not possible to seethe all the locations where the MNCs faced a trade-off between exports andforeign affiliates and decided for the former). However, in order to explorefor the role of knowledge intensity in the likelihood of horizontal expansion wefollow the guidelines of Helpman, Melitz and Yeaple (2004) using BEA dataon American MNCs for years 1999-2011 to show that sales of foreign affiliatesdecreases with the knowledge intensity level of the industry (using my ownmeasures of knowledge intensity), after controlling for export volume for thoseindustries. That is, by including exports in the specification I aim to controlfor sector specific demand of American exporters in that location. Thus, thespecification I estimate is:

log(ForeignSaless,y) = βklog(ks)+βtlog(ts)+log(Exportss,y)+αy+εs,y (D1)

Where s indexes for industry and y for year. The dependent variable is thesales of foreign affiliates of US multinationals in the rest of the world for sectors, in millions of dollars. The right hand side includes the knowledge intensity(denoted by k) and the unit shipping cost (denoted by t) of sector s. It alsoincludes US export volumes of industry s to the world in millions of dollars.There are in total seven sectors, and each is defined as a 3-digit NAICS code.The specification also includes year dummies (denoted by αy). The results ofthis estimation are presented in Table D1.

[Table D1 about here.]

Column 1 estimates a linear regression where the dependent variable is thesales of foreign affiliates (in logs), controlling for total exports to the samedestination in the same industry code. Column 2 estimates a linear regressionwhere the dependent variable is the ratio of sales of foreign affiliates to exports,similarly to Helpman, Melitz and Yeaple (2004). Column 3 replicates column 2but excludes outliers in the sales to exports distribution. It can be seen that theestimator for βk is negative and statistically significant, regardless of whetherthe specification uses the total sales of foreign affiliates controlling for exports(column 1), or the ratio of foreign sales to exports (column 2 and 3). Similarly,βt is estimated to be positive and significant, as expected (i.e. industries withlarger trade cost will generate incentives for firms to create foreign affiliatesabroad to substitute for exports). Figure D1 plots the estimation of βk in thefirst column of Table D1.

[Figure D1 about here.]

21

Figure D1: Sales of Foreign Affiliates for American MNCs and Knowledge In-tensity

−.5

0.5

11

.5R

atio

Sa

les F

A t

o E

xp

ort

s

−.1 −.05 0 .05 .1Log Knowledge Intensity

Note: estimates control for trade costs and year fixed effects

BEA US Multinational Enterprises

The vertical axis is the log-sales of foreign affiliates and the horizontal axis is the log knowledgeintensity measure.

22

Table D1: Sales of Foreign Affiliates for American MNCs and Knowledge Inten-sity

Dependent Variable: Log Sales of Foreign Affiliateslog(SalesFA) Ratio Ratio (no outliers)

log(k) -1.2664 -1.5221 -1.0200(0.292)*** (0.650)** (0.298)***

log(t) 0.7290 1.6101 1.7935(0.104)*** (0.229)*** (0.095)***

log(exp) 1.1858(0.048)***

The table presents results for the estimation of Specification (D1) usingthe Bureau of Economic Activity’s data on American MNCs for years1999-2011. The left hand side variable is the log of foreign sales ofAmerican MNCs in each 3-digit NAICS code (column 1) or the ratio offoreign sales to exports (columns 2 and 3). Column 3 exclude outlierobservations in terms of the ratio. The variables in the right hand sideinclude the unit shipping cost associated with the industry, a knowledgeintensity measure and the exports of the US to the rest of the world ineach 3-digits NAICS category (column 1 only). All specifications includeyear fixed effects. Robust standard errors clustered at the year level arepresented in parentheses.∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01

23

E Regression discontinuity uniform weighting schemeTable E1 repeats the estimation of Specification (3), using a uniform weight-ing scheme which provides equal weight to all observations within the selectedbandwidth. Results are robust to those presented in Table 5 in the main bodyof the paper.

[Table E1 about here.]

24

Table E1: Regression discontinuity estimation, uniform weighting schemeDependent Variable: Foreign subsidiary’s knowledge intensity (log)

150 250 350 OptimalcloserTZ 0.0084 0.0060 0.0075 0.0066

(0.003)** (0.003)* (0.002)*** (0.002)***distTZzero -0.0000 -0.0000 -0.0000 -0.0000

(0.000)*** (0.000)*** (0.000)*** (0.000)***closerTZ × distTZzero -0.0000 0.0000 -0.0000 0.0000

(0.000) (0.000) (0.000) (0.000)

N 5232 9702 14179 12367Adj. R2 0.87 0.87 0.86 0.87

The table presents results for the estimation of Specification (3) using a sample of foreignsubsidiaries of MNCs estimated using several bandwidths for the running variable specified ineach column. The last column uses the optimal bandwidth computed using the methodologydescribed in Cattaneo et al. (2018) who build on the work by Imbens and Kalyanaraman(2012). The estimation uses a uniform weight scheme, giving same weight to all observationswithin the bandwidth. All specifications include MNC fixed effects and time zone line fixedeffects. Robust standard errors clustered at the MNC and time zone line level are presentedin parentheses.∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01

25