Reading the Footprints: How Foreign Investors Shape Countries' Participation in Global Value Chains.

Buelens, Christian ; Tirpak, Marcel

INTRODUCTION

The fragmentation of production processes--following a firm's decision to outsource the production of inputs to an upstream supplier, whom it may possibly own--has become the norm nowadays, induced and facilitated by trade liberalisation, declining transport costs and improving communication technologies. As a result, many products undergo several value-enhancing intermediate stages of transformation and combination with other inputs prior to becoming final. When these sequential stages are separated by national borders, production processes (value chains) acquire an international (regional or even global), trade-creating, dimension, often referred to as global value chains (GVC). (1) Borders may indeed be crossed multiple times during such a process and both the final and the intermediate goods and services can consequently be regarded as "bundles" of multiple national and sectorial origins. (2)

Likewise, the mapping between the geographical location of a firm and its ownership has become blurred as a result of cross-border capital flows and foreign direct investment. A firm's decision to expand its activities abroad may reflect a variety of motives, including gaining access to new markets or taking advantage of differences in production costs (see, among others, Brainard, 1997; Helpman, 1984). The cross-border character of a production process thus provides a priori little indication as to the ownership of the different production units involved in it. Indeed, both the sourcing of inputs at arm's length or from foreign affiliates would be trade creating. Whereas the former requires a contractual agreement with a foreign supplier, a vertical investment is a prerequisite for the latter (either a greenfield investment, or a merger or acquisition). This decision ultimately depends on various factors, including firm and sector characteristics, trade and monitoring costs and risks linked to incomplete contracts.

In this article, we explore the link between foreign direct investment and countries' participation in global value chains. Our empirical analysis is based on an augmented gravity model framework applied to a newly constructed dataset combining GVC-related metrics and bilateral FDI stock for 40 developed and emerging economies. We find a positive relationship between the bilateral FDI stock on the one hand, and bilateral trade in both final and intermediate products, and the bilateral import content of exports on the other. Bilateral FDI thus affects both the volume and composition of trade flows. While we cannot formally distinguish between arm's length trade and affiliate sales, our results are consistent with the hypothesis that a substantial share of trade occurs within multinational firms and is the direct result of international sourcing. Our results also support the hypothesis that foreign investors contribute to shaping host economies' export structure and their participation in international production networks. Further, we show that traditional gravity variables play a significant role in explaining both the volume of GVC-related bilateral trade, as well as the bilateral import content of exports. We also present some evidence that cooperation costs--measured by linguistic and geographical proximity--are more relevant for trade that reflects production sharing.

This article contributes to the literature on the interaction between FDI and trade associated with cross-border production sharing. The decision of a firm to set up foreign affiliates and locate certain stages of its production process abroad, and hence becoming a multinational enterprise (MNE), depends on a number of factors including firm and sector characteristics (Antras and Helpman, 2004), production costs, trade and monitoring costs, risks linked to incomplete contracts (Markusen, 2002). FDI motives have traditionally been classified into two broad categories: horizontal or market-seeking FDI, which is primarily motivated by the possibility of supplying local markets (Markusen, 1984), and vertical FDI which exploits location advantages such as differences in production costs (Helpman, 1984). A concept that is closely related to the latter is "export platform" FDI, whereby foreign affiliates predominantly supply third markets, i.e. neither necessarily the home nor the host economy (Ekholm et al, 2007). FDI is thus prone to trade creation, both intra-firm and at arm's length.

The presence of foreign investment can affect host economies in multiple ways, possibly contributing to reshaping their economic and export structure. Amador and Cabral (2014) note that FDI is instrumental for the setting up of GVC run by multinational corporations. Nikolovova (2013), using data for EU27 countries, finds that an increase in FDI at sectoral level is associated with higher production and higher demand for intermediate goods, which suggest that FDI is primarily vertical and related to outsourcing activities. Alfaro and Charlton (2009) show that the share of vertical FDI is commonly underestimated and conclude that "intra-firm trade and foreign investment activity might be better explained by complex production processes involving several stages and decisions about not only where to source inputs, but also whether to source them from inside or outside the firm." Other effects of FDI pertinent to international production sharing include vertical productivity spillovers. Javorcik (2004) finds that these spillovers take place through backward linkages, i.e. contact between foreign affiliates and local suppliers, and are associated with partially owned foreign projects. Havranek and Irsova (2011), examining available studies on vertical productivity spillovers through a meta-analysis, conclude that spillovers through backward linkages generally dominate and greater spillovers are received by countries that are open to international trade and have relatively underdeveloped financial systems.

Many empirical papers focus on the Central and Eastern European countries (CEEC), which from a global perspective stand out both in terms of their degree of GVC participation and their high levels of inward FDI stock. (3) During their transition period towards becoming market economies, these countries received substantial FDI inflows and have established themselves as export platforms, notably vis-a-vis EU countries. The automotive industry constitutes the prime example of this (see IMF, 2013). As noted by Iossifov (2014), cross-border production sharing has recently intensified within the CEEC region, following the setting up and subsequent expansion of regional industrial clusters. Damijan et al. (2013) find that inward FDI in CEEC has altered their export composition and positively affected total factor productivity. They also fostered the development of production networks in CEEC leading to increased trade integration of the region (Kaminsky and Ng, 2005). Behar and Freund (2011) find that following the 2004 EU enlargement, intraEU trade in intermediate goods has become more sophisticated, and the role of countries joining the EU as suppliers of intermediate goods for the incumbent countries has increased.

While structural factors, such as proximity to consumer markets and natural resource endowments, play a role in the development of global value chains, Pathikonda and Farole (2016) argue that there is substantial scope for policies to contribute to this process. They distinguish between longer-term policies facilitating investment in capabilities (e.g. development of physical, human and institutional capital) and other structural policies (e.g. connectivity, market access, wage competitiveness), which help to shape the GVC participation over a shorter time horizon. Bajgar and Javorcik (2017), using data on Romanian manufacturing sector, provide evidence that the presence of MNEs has positive spillovers on the quality of exports by domestic firms. This occurs in particular via backward vertical spillovers, as MNEs demand inputs of higher quality from domestic firms. They conclude that policies promoting FDI would hence help stimulating product quality in host economies.

This article also relates to a series of recent studies using international input-output tables. The latter investigate, among others, the following issues: the correction of multiple counting of intermediate products in trade flows (Koopman et al., 2010); the measurement of foreign demand and trade elasticities (Bussiere et al, 2013); the reshuffling of the geographical and sectorial composition of the trade basket, affecting competitiveness indicators either based on gross trade weights, such as effective exchange rates (Bems and Johnson, 2012; Bayoumi et al, 2013), or on the export structure, such as the revealed comparative advantage (Koopman et al., 2010); the effect on bilateral trade balances (Nagengast and Stehrer, 2014); and spillover channels across countries and sectors (Acemoglu et al., 2015).

The remainder of the article is structured as follows. "Cross-border production sharing--stylised facts" section briefly discusses stylized facts on cross-border production sharing worldwide. "Empirical strategy" section presents our data sources, GVC metrics and empirical strategy as well as empirical results from a gravity model augmented by FD1 stock. "Robustness tests" section presents a robustness test and "Conclusions" section concludes.

CROSS-BORDER PRODUCTION SHARING--STYLIZED FACTS

Export openness varies considerably across the countries in our sample. Measured in gross terms, it ranges from around 10% in the USA to almost 90% in Ireland (see Figure 1). When considering value-added transfers, which capture the actual revenues arising from international trade, countries' export intensities, as well as their dispersion, narrow significantly. The proportion of value added exported to GDP remains nonetheless substantial, stretching from 10% of GDP in the USA to around 50% in Ireland. Many EU countries, and particularly some CEEC, rank high on both measures, making them the most open economies worldwide.

Three general observations can be made with regard to the wedge between gross and value-added exports. First, foreign inputs appear to constitute a pre-condition for the generation of domestic value added and thus trade revenues. Second, the relative ranking of countries in terms of openness remains largely unchanged when employing alternative measures of exports. Third, the share of foreign value added increases with gross export openness in a nonlinear way. The size of the wedge between gross exports and exported value added can be attributed to differences in the production processes across countries as well as differences in the product composition of their export basket. These differences in turn depend on multiple factors, inter alia on a country's size, geographic location, its production factor endowments and their allocation within the economy.

The most commonly used gauge of a country's participation in international production networks is the import content of exports. While the intensification of international production sharing and the faster rise of trade in intermediate goods compared to trade in final goods has been a global phenomenon (see Hummels et al., 2001), EU countries stand out in international comparison. Since the mid-1990s, the average share of import content has risen from above a quarter of gross exports to just under 40% prior to the outbreak of the global financial crisis (4) (see Figure 2).

Zooming in on the subset of exported intermediate goods that is exported further, post-processing, offers a complementary view on countries' integration in production chains. The domestic value added that is embedded in this subset of exports, referred to as forward linkage, will hence become import content of the trading partner's exports. It captures complementary activities between the exporting firms and its foreign downstream customers. It also measures countries' vertical specialisation (Hummels etai, 2001) and hence the degree of dependence on the demand faced by the firms in downstream countries processing its exported intermediary products. In a global comparison, EU countries, and in particular CEEC, stand out with forward linkages amounting to about 7% of their annual GDP (see Table 1, column h). Intermediate goods that are exported by CEEC and, once processed abroad, are exported to third countries, account for almost a quarter of their gross exports (Table 1, column d), which is a markedly higher share than for any other region worldwide.

EMPIRICAL STRATEGY

Data sources and GVC metrics

When deriving global value chain metrics, a distinction can be made between two broad concepts. The first concept is based on gross exports and consists in disaggregating the value added embedded in them either by origin or by destination. The breakdown by origin decomposes gross exports backwards (upstream) into domestic and a foreign value added, which in turn can be further disaggregated by country or sector of origin. The breakdown by destination decomposes gross exports forward (downstream) according to the future geographic itinerary of the value added (see Koopman et al, 2010). A second concept, introduced by Johnson and Noguera (2012), considers trade in value added. It thus establishes a direct link between the country where value originates ("value-added exporter") and the country where it is absorbed ("value-added importer"). This bilateral originator-absorber relationship is, however, artificial, as the country pair does not necessarily engage in actual trade, and the actual itinerary through which value added is effectively traded is lost. Our empirical analysis largely uses GVC metrics derived using the first concept, which allows for tracking value added in gross exports, while preserving information on the bilateral trade relationship.

We calculate the global value chain metrics from the World Input-Output Database (Timmer, 2012; Dietzenbacher et al, 2013), which contains international sector-by-sector input-output tables (5) in annual frequency for the period from 1995 to 2011. It covers 40 countries, including 27 EU member states, as well as a Rest-of-the-World aggregate and 35 sectors (see "Appendix"). Data on the bilateral inward FDI stock (i.e., equity capital, reinvested earnings and intra-company loans) are taken from the UNCTAD database. We obtain the gravity variables from the GeoDist database of CEPII (Mayer and Zignago, 2011; and Head and Mayer, 2014) and a linguistic similarity index from Toubal and Melitz (2014).

The empirical approach in this article builds on a gravity model framework. Gravity models, which have become a mainstay of international trade analysis, have a strong intuitive appeal and have proven to be empirically successful in "capturing the deep regularities in the pattern of international trade" (Shepherd, 2013). Traditionally, gravity models help to explain the magnitude of gross bilateral trade flows using country-specific and country-pair-specific characteristics --e.g. the bilateral distance and the presence of a common language--as well as potential frictions and/or catalysers of international trade (e.g. free trade area, common currency). In this article, we use the gravity model framework to analyse trade flows associated with global production sharing, as well as the geographical composition of the value added embedded in trade, focussing in particular on the role of bilateral FDI.

A gravity model with GVC-related trade flows

This section reviews the sensitivity of different measures of GVC-related trade flows with respect to traditional gravity variables. Following the export decomposition proposed by Koopman et al, (2010), we obtain the following six alternative dependent variables, which all capture different GVC-related characteristics: (1) total gross exports from country t to country j in year t ([X.sub.ijt]); (2) gross exports of final (consumer) goods ([Y.sub.ijt]); (3) gross exports of intermediate goods ([A.sub.ij] [X.sub.j,t]); (4) gross exports of intermediate goods, which remain in the country after going through the production process ([A.sub.ij] [X.sub.jj,t]); (5) gross exports of intermediate goods, which are embedded in country j's exports to country k ([A.sub.ij] [X.sub.jk,t]); (6) gross exports of intermediate goods, which are embedded in country j's exports to country i [[A.sub.ij][X.sub.ji,t]), i.e. "re-imports". (6)

We estimate a gravity model for each of the six trade flow variables, which enter the model in logarithmic transformation. In our first estimation (see Eq. 1), we regress [X.sub.ijt] on the following set of gravity variables: bilateral distance ([dist.sub.ij], in logs), contiguity ([contig.sub.ij]), a common language index ([cl.sub.ij]) (7) and common legal origin ([comleg.sub.ij]). We estimate the model using the ordinary least squares (OLS) estimator. The setup includes exporter-year ([[phi].sub.it]) and importer-year ([[phi].sub.jt]) fixed effects, that account for unobservable factors affecting trade at the level of the exporter and importer, respectively, in a given year. This specification ensures that the model contains multilateral resistance terms and is consistent with theoretical derivation of the standard gravity equation provided by Anderson and van Wincoop (2003). We then repeat the estimation, by replacing [X.sub.ijt] by the other trade flow variables. The estimates are reported in Table 2 (columns 1-6).

ln [X.sub.ijt] = [[delta].sub.0] + [[delta].sub.1] ln [dist.sub.ij] + [[delta].sub.2] [contig.sub.ij] + [[delta].sub.3][cl.sub.ij] + [[delta].sub.4] [comleg.sub.ij] + [[phi].sub.it] + [[phi].sub.jt] + [[epsilon].sub.ijt]. (1)

In the next specification (see Eq. 2), we also use the bilateral import content of gross exports, measured as the value added of country o (the originator) embedded in the total gross exports of country i (the exporter), [VAiX.sup.i.sub.o] (where o [not equal to] i) as dependent variable, to which we refer to as the "foreign footprint". This can be thought of as the value added generated in country o "exported" by country i. (8) Note that the bilateral import content from country o does not necessarily need to have been imported directly in full and may have entered country i embedded in the imports from third countries. This specification has some conceptual differences with respect to the traditional bilateral trade gravity model. In this specification, our interest rests on the origin of the import content in country's i overall exports (across all destinations)--which becomes the dependent variable, following a similar approach as in Rahman and Zhao (2013)-rather than on the bilateral trade flow between country i and country j. The gravity variables thus focus on the bilateral relationship between country i and country o. The estimates are reported in Table 2 (column 7).

In [VAiX.sup.i.sub.ot] = [[delta].sub.0] + [[delta].sub.1] ln [dist.sub.i0] + [[delta].sub.2] [contig.sub.io] + [[delta].sub.3] [cl.sub.io] + [[delta].sub.4] [comleg.sub.io] + [[phi].sub.it] + [[phi].sub.ot] + [[epsilon].sub.iot]. (2)

We also estimate a bilateral trade model including exporter and importer GDP in view of testing how countries' respective economic size--as proxies for supply and demand capacity--affects bilateral trade (Eq. 3). As in Eq. 1, we iterate through the six types of trade flows. In order to remain consistent with theory and account for the multilateral resistance terms, the pair-specific variables--distance, contiguity, language proximity and common legal origin--are corrected (9) by the method proposed by (Baier and Bergstrand, 2009). This method relies on a Taylor series approximation of the multilateral resistance terms and essentially requires weighting the bilateral variables. (10) The specification also includes year-fixed effects. Separately, we also modify Eq. 2 by including GDP of the exporter-country and the originating country (see Eq. 4). The estimates of Eqs. 3 and 4 are reported in "Appendix" (see Table 9).

ln [X.sub.ijt] = [[delta].sub.0] + [[beta].sub.1] ln [GDP.sub.it] + [[beta]].sub.2] ln [GDP.sub.jt] + [[delta].sub.1] ln [dist.sup.*.sub.ij] + [[delta].sub.2] [contig.sup.*.sub.ij] + [[delta].sub.3] [cl.sup.*.sub.ij] + [[delta].sup.4] [comleg.sup.*.sub.ij] + [[pji].sub.t] + [[epsilon].sub.ijt] (3)

ln [VAiX.sup.i.sub.ot] = [[delta].sub.0] + [[beta].sub.1] ln [GDP.sub.it] + [[beta]].sub.2] ln [GDP.sub.ot] + [[delta].sub.1] ln [dist.sup.*.sub.io] + [[delta].sub.2] [contig.sup.*.sub.io] + [[delta].sub.3] [cl.sup.*.sub.io] + [[delta].sup.4] [comleg.sup.*.sub.io] + [[pji].sub.t] + [[epsilon].sub.iot] (4)

Overall, we find that the coefficients on distance, linguistic similarity and common legal origin are statistically significant and their sign and magnitude are in line with the literature (e.g. Head and Mayer, 2014). The coefficient on contiguity is not significantly different from zero. Beyond these expected results, some important observations pertaining to GVC-related trade emerge: the elasticities on distance, providing a proxy for trade costs, vary depending on the type of trade flow considered, albeit slightly. Final goods are indeed somewhat less responsive to trade costs, measured by distance, than intermediate goods, which may reflect the lower degree of substitutability of final relative to intermediate goods. The estimated higher elasticity for intermediate products also points to a predominantly regional--as opposed to "global"--character of cross-border value chains, confirming the findings of (Baldwin, 2012). Furthermore, the coefficient on linguistic proximity turns out higher for intermediate goods than for final goods, underlining the importance of smooth communication and cooperation in shared production structures.

The regression results of the specification considering the bilateral import content of exports (Eqs. 2 and 4)--the "foreign footprint"--are reported in the last column of Tables 2 and 9, respectively. The foreign footprint captures the bilateral input reliance and reflects the complementarity in production for a given country pair. The statistical significance of the respective gravity variables is broadly comparable to that obtained in the bilateral trade gravity model, while the estimated elasticities on distance and on the legal origin are markedly lower. This suggests that besides their volume effect on bilateral trade flows, gravity variables also explain the gross export structure.

In a separate specification, (11) we use the bilateral import content of exports relative to gross exports as dependent variable. This estimation also includes countries' GDP as proxies for their economic size. The estimated coefficient on exporter's GDP is negative, implying that the larger (smaller) the exporting country, the less (more) foreign value added is likely to be embedded in its gross exports. This result underlines the propensity of small economies to source their inputs from abroad. It is therefore natural to see exports of small and open economies containing a high share of import content that predominantly originates in larger and neighbouring countries.

Gravity model augmented with foreign direct investment

To analyse the relationship between direct investment on various trade flows, we augment the standard gravity model with the bilateral FDI stock. In this way, we analyse how the latter affects the geographic pattern of international trade and the mode of countries' participation in global value chains. A multinational firm entering a country via a direct investment is likely to spur trade between their home country and the country they are investing in, both in final and intermediate products. This could result from higher intra-firm trade, but also from a more intensive trade with multinational firm's traditional input providers or input purchasers in their home countries.

We estimate a gravity model for gross exports from country i to country j in year t([X.sub.ijt]), regressing it on the bilateral inward FDI stock ([FD.sub.jit]), while controlling for the bilateral distance ([dist.sub.ij], in logs), contiguity ([contig.sub.ij],), the common language index ([cl.sub.ij]) and the common legal origin ([comleg.sub.ij]) (Eq. 5). The setup includes exporter-year ([[phi].sub.it]) and importer-year ([[phi].sub.jt]) fixed effects to account for unobservable factors affecting trade at the level of the exporter and importer, respectively, in a given year. We estimate the gravity model by ordinary least squares (OLS) and report an alternative specification using the nonlinear Poisson pseudo-maximum likelihood estimator as a robustness test (see "Robustness tests" section).

In [X.sub.ijt] = [[delta].sub.0] + [[delta].sub.1]ln [dist.sub.ij] + [[delta].sub.2] [contig.sub.ij] + [[delta].sub.3][cl.sub.ij] + [[delta].sub.4] [comleg.sub.ij] + [[delta].sub.5] ln [FDI.sub.jit] + [[phi].sub.it] + [[phi].sub.jt] + [[epsilon].sub.ijt]. (5)

The coefficient on the FDI stock is positive and statistically significant across all specifications, which confirms that foreign direct investment is positively associated with higher bilateral trade volumes, i.e. a country exports more to a country that has provided direct investment (see Table 3). The coefficients on distance, linguistic proximity and common legal origin remain significant with the expected sign, while the coefficient on contiguity turns significant compared to the model specification without FDI stock. In line with our previous results, we observe some variation in the coefficients of the respective explanatory variables when different types of exports are used as dependent variables. One observation that emerges for exports of intermediate goods is that the coefficient on the FDI stock is somewhat higher for intermediate goods with a lower degree of finalisation (Table 3, columns 5 and 6), consistent with the export platform FDI hypothesis. Overall, the estimated coefficients for FDI stock for exported final and intermediate products are broadly comparable, which does not allow us to discriminate whether horizontal or vertical motives behind foreign investment prevail.

We run the same set of regressions using bilateral imports as a dependent variable, [M.sub.ijt], in order to test for the complementarity of bilateral imports and the bilateral FDI stock (Eq. 6). The positive and significant coefficients on the FDI stock across different specifications suggest that foreign investments are associated with higher trade integration also on the import side (see Table 4). This is compatible with both the horizontal and the vertical motives for FDI. An example of the former would for instance be a retailer investing in a host country to distribute products for final consumption imported from the retailer's home country. An example of the latter would be the assembly and finalisation of intermediary products exported from the home to the host country and the sold further. More generally, the estimated positive relationship supports the hypothesis of substantial intra-firm trade between multinationals and their foreign affiliates, whereby these multinationals "carry" products that are then sold or processed in the host country, de facto establishing intermediate and/or final stages within a cross-border supply chain. The marginally higher coefficient on the bilateral FDI stock in the export equation relative to the import equation suggests that the vertical motive for FDI may be slightly dominating the horizontal one. The higher elasticity of foreign direct investment for the intermediate goods compared to the final goods in the import equation also supports this interpretation.

ln [M.sub.ijt] = [[delta].sub.0] + [[delta].sub.1] ln [dist.sub.ij] + [[delta].sub.2] [contig.sub.ij] + [[delta].sub.3][cl.sub.ij] + [[delta].sub.4] [comleg.sub.ij] + [[phi].sub.it] + [[phi].sub.jt] + [[epsilon].sub.ijt]. (6)

Finally, we investigate how the bilateral FDI stock affects the bilateral import content of gross exports ([VAiX.sup.i.sub.o]), again using an augmented gravity model (Eq. 7). Akin to Eq. 2, the bilateral relationship of interest is thus between the exporter and the originator--both of the import content and the FDI stock rather than the destination country, explaining the change in notation.

ln [VAiX.sup.i.sub.ot] = [[delta].sub.0] + [[delta].sub.1] ln [dist.sub.io] + [[delta].sub.2] [contig.sub.io] + [[delta].sub.3][cl.sub.ij] + [[delta].sub.4] [comleg.sub.ij] + [[delta].sub.5] ln [FDI.sub.oit] + [[phi].sub.it] + [[phi].sub.ot] + [[epsilon].sub.iot]. (7)

The regression results (see Table 5) show that the bilateral FDI stock is positively associated with the bilateral import content of exports. This suggests that the production structures established through the bilateral FDI not only affect the volume of bilateral trade, but also affect the composition of an FDI-host country's exports.

ROBUSTNESS TESTS

The empirical analysis presented above is based on gravity models estimated by ordinary least squares (OLS). Santos Silva and Tenreyro (2006) show that the parameters of log-linearised models (such as the gravity equation) estimated by OLS may under certain circumstances be biased and inconsistent. To surmount this problem, they propose a nonlinear Poisson PseudoMaximum Likelihood (PPML) estimator. In order to test for the robustness of our results obtained with OLS, we re-estimate the regressions with the PPML estimator. As shown in Table 6, which displays the results of Eq. 5 estimated by PPML, the estimates are broadly comparable in sign and statistical significance, hence supporting our conclusions. (12)

CONCLUSIONS

Cross-border production sharing has become the norm nowadays, and a growing share of firms participates in global value chains. The implications of this fragmented production mode have sparked a growing interest by academics and policy-makers alike. In this article, we investigate determinants of countries' participation in international production chains. More specifically, we analyse how foreign investors shape the host countries' export structure and their degree of GVC participation. We construct a new dataset, combining GVC participation metrics and bilateral FDI stocks. Using an augmented gravity model framework, we find a significant positive relationship between the bilateral FDI stock and bilateral trade in both final and intermediate products. We also find the bilateral FDI stock to have a compositional effect on exports by being positively related to the bilateral import content of exports (the "foreign footprint"). We interpret these results as evidence that the trade-generating effect of FDI primarily relates to intra-firm trade. Overall, our results indicate that foreign investors actively shape host economies' participation in international production networks and that this is reflected in both trade volumes and trade structure. Further, we show that traditional gravity variables play a significant role in explaining GVC-related bilateral trade, as well as the bilateral import content of exports. We also find some evidence that cooperation costs measured by linguistic and geographical proximity--seem more relevant for trade that reflects production sharing.

Acknowledgements

We are grateful for helpful comments and suggestions from Maja Ferjancic, Martin Schmitz, Fernando Zarzosa, the editor, Paul Wachtel, and four anonymous referees. We also thank participants at the 2015 Slovak Economic Association Meeting in Kosice, the 2016 INFER workshop in Bratislava, the 2016 European Trade Study Group in Helsinki, the 2016 Vienna Investment Conference, Center for Social and Economic Research (CASE) 25th Anniversary Conference in Warsaw and an ECB seminar for useful discussions and Giovanni Palmioli for his research assistance in early stages of this project.

This article should not be reported as representing the views of the European Central Bank (ECB) and/or European Commission (EC). The views expressed are those of the authors and do not necessarily reflect those of the ECB and/or the EC.

REFERENCES

Acemoglu, D, Akcigit, U and Kerr, W. 2015: Networks and the macroeconomy: an empirical exploration. NBER Working Paper Series, Working Paper 21344.

Alfaro, L and Charlton, A. 2009: Intra-Industry foreign direct investment. American Economic Review 99(5): 2096-2119.

Amador, J and Cabral, S. 2014: Global value chains surveying drivers and measures. ECB Working Paper Series 1739.

Anderson, JE and van Wincoop, E. 2003: Gravity with gravitas: a solution to the border puzzle. American Economic Review 93(1): 170.

Antras, P and Helpman, E. 2004: Global sourcing. Journal of Political Economy 112 (3):552-580.

Baier, SL and Bergstrand, JH. 2009: Bonus vetus OLS: A simple method for approximating international trade-cost effects using the gravity equation. Journal of International Economics 77(1):77-85.

Bajgar, M and Javorcik, B. 2017: Climbing the rungs of the quality ladder: FDI and domestic exporters in Romania, mimeo. http://users.ox.ac.uk/~econ0247/Romania.pdf.

Baldwin, R. 2012: WTO 2.0: Global governance of supply-chain trade. Centre for Economic Policy Research (CEPR) Policy Insight No. 64.

Baldwin, R and Taglioni, D. 2006: Gravity for dummies and dummies for gravity equations. NBER Working Paper 12516.

Bayoumi, T, Saito, M and Turunen, J. 2013: Measuring competitiveness: Trade in goods or tasks?. IMF Working Paper WP/13/100, May 2013.

Behar, A and Freund, C. 2011: Factory Europe? Brainier but not brawnier, World Bank, Other papers, https://doi.org/10.1596/24495.

Bems, R and Johnson, RC. 2012: Value-added exchange rates. NBER Working Papers 18498.

Bems, R, Johnson, RC and Yi, K-M. 2012: The great trade collapse. NBER Working Paper 18632.

Brainard, S. 1997: An empirical assessment of the proximity-concentration tradeoff between multinational sales and trade. American Economic Review 87: 520-544.

Buelens, C and Tirpak, M. 2017: Reading the footprints: How foreign investors shape countries' participation in global value chains. ECB Working Paper Series 2060.

Bussiere, M, Callegari, G, Ghironi, F, Giulia, S and Yamano, N. 2013: Estimating trade elasticities: Demand composition and the trade collapse of 2008-2009. American Economic Journal: Macroeconomic 5(3): 118-151.

Damijan, J, Kostevc, C and Rojec, M 2013: Global supply chains at work in central and eastern European countries: Impact of FDI on export restructuring and productivity growth. LICOS Discussion Paper 332/2013.

Dietzenbacher, E, Los, B, Stehrer, R, Timmer, M and de Vries, G 2013: The construction of the world input-output tables in the WIOD project. Economic Systems Research 25(1): 71.

Ekholm, K, Forslid, R and Markusen, JR. 2007: Export platform foreign direct investment. Journal of the European Economic Association 5(4): 776-795.

Ferrantino, MJ and Taglioni, D. 2014: Global value chains in the current trade slowdown. Economic Premise, Number 137, World Bank.

Havranek, T and Irsova, Z. 2011: Estimating vertical spillovers from FDI: Why results vary and what the true effect. Journal of International Economics 85:234-244.

Head, K and Mayer, T. 2014. Gravity equations: Workhorse, toolkit, and cookbook. In: Gopinath, G, Helpman, E and Rogoff, K. (eds). Handbook of International Economics. Elsevier, vol. 4, pp. 131-195.

Helpman, E. 1984: A simple theory of international trade with multinational corporations. Journal of Political Economy 92(3): 451-471.

Hummels, D, Ishii, J and Kei-Mu, Y. 2001: The nature and growth of vertical specialization in world. Journal of International Economics 54(1): 75-96.

IMF. 2013: IMF Multi-Country Report: German-Central European Supply Chain--Cluster Report, August 2013. IMF.

Iossifov, P. 2014: Cross-border production chains and business cycle co-movement between Central and Eastern European Countries and Euro Area Member States. ECB Working Paper Series 1628.

Javorcik, B. 2004: Does foreign direct investment increase the productivity of domestic firms? In search of spillovers through backward linkages. American Economic Review 94(3): 605-627.

Johnson, RC and Noguera, G. 2012: Fragmentation and trade in value added over four decades. NBER Working Paper 18186.

Kaminsky, B and Ng, F. 2005: Production disintegration and integration of central Europe into global markets. International Review of Economics and Finance 14: 377-390.

Koopman, R, Powers, W, Wang, Z and Wei, S-J 2010: Give credit where credit is due: tracing value added in global production chains. NBER Working Paper 16426.

Markusen, J. 1984: Multinationals, multi-plant economies, and the gains from trade. Journal of International Economics 16(3-4): 205-26.

Markusen, J. 2002: Multinational Firms and the Theory of International Trade. Cambridge: MIT Press.

Mayer, T and Zignago, S. 2011: Notes on CEPll's distances measures: The GeoDist Database.

Nagengast, AJ and Stehrer, R. 2014: Collateral imbalances in intra-European trade? Accounting for the differences between gross and value added trade balances. ECB Working Paper 1695.

Nikolovova, P. 2013: Sourcing patterns of FDI activity and their impact on the domestic economy. Czech Journal of Economics and Finance 63: 288-302.

Noguera, G. 2012: Trade costs and gravity for gross and value added trade. Job Market Paper. https://sites.google.com/site/guillermonoguera/home/files/Noguera_JMP.pdf?attredirects = 0.

Pathikonda, V and Farole, T. 2016: The capabilities driving participation in global value chains. World Bank Policy Research Paper No. 7804.

Rahman, J and Zhao T. 2013: Export performance in Europe: What do we Know from the supply links?. IMF Working Paper WP/13/62.

Santos Silva, JMC and Tenreyro, S. 2006: The log of gravity. The Review of Economics and Statistics 88(4): 641-658.

Shepherd, B. 2013: The Gravity Model of International Trade: A User Guide. ARTNeT Books and Research Reports, United Nations Economic and Social Commission for Asia and the Pacific (ESCAP).

Stehrer, R. 2013: Accounting relations in bilateral value added trade, wiiw Working Papers 101.

Timmer, M. 2012: The World Input-Output Database (WIOD): Contents, Sources and Methods. WIOD working paper nr. 10.

Toubal, F and Melitz, J. 2014: Native language, spoken language, translation and trade. Journal of International Economics 93(2): 351-363.

APPENDIX Table 7: Country sample World a + b + c 40 EU a = a1 + a2 27 CEEC al 10 Bulgaria, Czech Republic, Estonia, Hungary, Lithuania, Latvia, Poland, Romania, Slovakia, Slovenia Other EU a2 17 Austria, Belgium, Cyprus, Germany, Denmark, Spain, Finland, France, United Kingdom, Greece, Ireland, Italy, Luxembourg, Malta, Netherlands, Portugal, Sweden EMEs b 9 Brazil, China, Indonesia, India, South Korea, Mexico, Russia, Taiwan, Turkey Advanced c 4 Australia, Canada, Japan, USA Source: WIOD, UNCTAD Table 8: Summary statistics Variable Obs Mean Std. Dev. FDI stock (in logs) 9381 6.14 2.89 Gross exports (in logs) 9381 7.22 2.14 Of final goods (in logs) 9381 6.12 2.20 Of intermediate goods (in logs) 9381 6.73 2.15 Which are consumed post-processing 9381 6.33 2.17 (in logs) Which are exported post-processing 9381 5.44 2.13 (in logs) Which are shipped back post-processing 9381 0.94 3.41 (in logs) Import content of exports (in logs) 9381 -1.58 1.85 Distance (in logs) 9381 7.80 1.13 Contiguity 9381 0.09 0.28 Language similarity 9381 0.18 0.17 Common legal origin 9381 0.24 0.43 Variable Min Max FDI stock (in logs) -0.69 13.04 Gross exports (in logs) -0.79 12.93 Of final goods (in logs) -2.75 12.29 Of intermediate goods (in logs) -2.10 12.26 Which are consumed post-processing -2.16 12.15 (in logs) Which are exported post-processing -5.02 10.23 (in logs) Which are shipped back post-processing -12.75 10.47 (in logs) Import content of exports (in logs) -7.36 3.06 Distance (in logs) 4.09 9.84 Contiguity 0.00 1.00 Language similarity 0.00 0.99 Common legal origin 0.00 1.00 Note: The summary statistics refer to the sample used in Tables 3, 4, 5, and 6. Source: WIOD, UNCTAD, GeoDist, and Toubal and MeLitz (2014) Table 9: GDP-weighted gravity model for exports, OLS Variables Gross exports Final goods Intermediate goods (1) (2) (3) Log GDP- 0.807 *** 0.835 *** 0.799 *** exporter (0.0153) (0.0168) (0.0156) Log GDP- 0.720 *** 0.725 *** 0.718 *** importer (0.0148) (0.0162) (0.0154) Distance -1.103 *** -1.057 *** -1.143 *** (0.0802) (0.0872) (0.0808) Contiguity 0.175 0.211 0.168 (0.198) (0.209) (0.201) Language 0.670 ** 0.536 0.722 ** (0.331) (0.358) (0.332) Common legal 0.290 *** 0.322 *** 0.284 *** origin (0.108) (0.115) (0.109) Observations 26,514 26,506 26,514 R-squared 0.761 0.722 0.750 Year FE YES YES YES Intermediate goods, which are processed and 'Foreign Variables Consumed Exported Shipped footprint' further back (4) (5) (6) (7) Log GDP- 0.790 *** 0.807 *** 1.528 *** 0.598 *** exporter (0.0153) (0.0186) (0.0282) (0.0135) Log GDP- 0.770 *** 0.570 *** 0.564 *** 0.826 *** importer (0.0153) (0.0176) (0.0285) (0.0138) Distance -1.134 *** -1.165 *** -2.333 *** -0.876 *** (0.0785) (0.0931) (0.163) (0.0692) Contiguity 0.172 0.0793 0.274 0.217 (0.195) (0.239) (0.406) (0.192) Language 0.707 ** 0.718 * 1.309 * 0.637 ** (0.320) (0.396) (0.674) (0.324) Common legal 0.287 *** 0.276 ** 0.558 ** 0.174 * origin (0.104) (0.129) (0.219) (0.101) Observations 26,514 26,514 26,508 26,520 R-squared 0.767 0.658 0.688 0.773 Year FE YES YES YES YES *** p < 0.01, ** p < 0.05, * p < 0.1. Note: The panel includes annual observations for 40 countries over the period 1995-2011. The models include robust variance estimates yielding heteroscedasticity-consistent standard errors, while distance is used as a clustering variable. Robust standard errors in parentheses.

Electronic supplementary material The online version of this article (https://doi.org/10.1057/ s41294-017-0036-2) contains supplementary material, which is available to authorized users.

(1) The terms global value chains, global supply chains, international production chains, international/cross-border production sharing are used interchangeably in the literature and in the present article.

(2) In many cases, fragmenting the production process is unavoidable, notably when natural resources are involved (in that case exports are fully conditional on imports). But fragmentation and the blurring of the 'made in' attribute have also become a reality for most other products, which are likely to contain (directly or indirectly) some fraction of foreign value added.

(3) Buelens and Tirpak (2017) report a stronger association between the FDI stock and GVC participation for CEEC as compared to other countries. Furthermore, sector-level FDI in the CEEC region is positively related to the import-intensity of the sector's exports, which in turn supports the export platform FDI hypothesis.

(4) In the wake of the global financial crisis the average share of import content of exports declined, as the 2009 global trade collapse weighed disproportionately on trade in intermediate products. As trade recovered, the import content in exports rebounded. This reflects the fact that the 2009 trade collapse resulted from a severe adverse shock to final demand, affecting in particular the production of input-intensive and more complex durable goods, for which multiple counting of trade is particularly acute. Inventory adjustments and credit supply constraints further exacerbated the drop in final demand during the crisis (see, among others, Bems et al, 2012; and Ferrantino and Taglioni, 2014).

(5) It should be emphasised that international input-output tables are themselves estimates based on a number of assumptions and simplifications. For example, all firms in an industry are assumed to use the same input combination and thus the same technology; or multi-product firms are typically classified within the sector of primary production, which may distort the imputed industry technology.

(6) Note that all trade flows considered (1-6) contain both domestic and foreign value added. Total trade flows (1) can be broken down into final (2) and intermediate trade (3), while trade flows (4-6) are subsets of intermediate trade (3). The [A.sub.ij] term refers to a sub-matrix of technical (input-output) coefficients, which specify in which proportion inputs from country i enter country j's production process.

(7) The common language index takes into account the linguistic proximity of two languages, even if they are formally distinct. All other things equal, a higher linguistic similarity should facilitate cooperation via lower interpretation and communication costs.

(8) In other words, it is a backward (upstream) link of country I to country 0, or, equivalently, the forward (downstream) link of country 0 via country t.

(9) Restricting the potential determinants of bilateral trade flows to country- and pair-specific variables of the two countries involved only, would disregard the fact that countries (generally) have more than one bilateral trade partner and that other bilateral trade relationships may create or divert trade--failure to do so would yield a "naive" version of the gravity model with omitted variable and award the "gold medal" of gravity model errors (Baldwin and Taglioni, 2006).

(10) Specifically, the transformation for the bilateral distance is given by ln [dist.sup.*.sub.ij] = [1/N [N.summation over (j=1)] ln [dist.sub.ij]) + 1/N ([N.summation over (i=1)] ln [dist.sub.ij] - 1/[N.sup.2]([N.summation over (i=1)] [N.summation over (j=1)] ln [dist.sub.ij])]. The transformation for contiguity is similar.

(11) See Electronic Supplementary Material of this article (https://doi.org/10.1057/s41294-0170036-2).

(12) Further results from alternative specifications, including (i) a baseline gravity model for imports estimated by PPML and (ii) baseline gravity models for exports and imports using truncated time sample, excluding the crisis and post-crisis periods are reported in the Electronic Supplementary Material to this article (https://doi.org/10.1057/s41294-017-0036-2).

CHRISTIAN BUELENS (1) & MARCEL TIRPAK (2)

(1) European Commission, Brussels, Belgium. E-mail: Christian.buelens@ec.europa.eu

(2) European Central Bank, Frankfurt, Germany. E-mail: marcel.tirpak@ecb.europa.eu

Caption: Figure 1: Export openness, 2008

Caption: Figure 2: Evolution of the import content of exports Table 1: Gross exports broken down by product category and origin, 2008 Gross exports Final Intermediate goods goods a b EU 37.8 62.2 CEEC 37.4 62.6 Rest-EU 37.9 62.1 EMEs 35.1 64.9 Advanced 29.7 70.3 Memo item World 32.9 67.1 Gross exports Intermediate goods. which are processed and Consumed Exported Shipped further back c d e EU 42.4 18.4 1.4 CEEC 40.6 21.5 0.5 Rest-EU 42.6 18.0 1.5 EMEs 45.9 17.9 1.1 Advanced 52.4 14.9 3.1 Memo item World 48.4 16.3 2.3 Gross exports Domestic Foreign Forward value value linkage added added (as % of GDP) f g h EU 68.5 31.5 5.3 CEEC 60.0 40.0 6.9 Rest-EU 69.5 30.5 5.2 EMEs 75.0 25.0 4.4 Advanced 82.9 17.1 2.2 Memo item World 73.7 26.3 3.9 Note: Percentage of gross exports unless noted otherwise. For definition of regions, please see 'Appendix'. Source: WIOD, authors' calculations Table 2: Baseline gravity model, OLS Gross exports Final goods Intermediate goods Variables (1) (2) (3) Distance -1.055 *** -1.003 *** -1.096 *** (0.0513) (0.0551) (0.0524) Contiguity 0.216 0.259 0.207 (0.150) (0.161) (0.152) Language 0.674 *** 0.538 ** 0.723 *** (0.240) (0.255) (0.246) Common legal origin 0.301 *** 0.336 *** 0.296 *** (0.0658) (0.0696) (0.0672) Observations 26,514 26,506 26,514 R-squared 0.900 0.894 0.890 Exporter-year FE YES YES YES Importer-year FE YES YES YES Intermediate goods, which are processed and Consumed Exported further Variables (4) (5) Distance -1.091 *** -1.107 *** (0.0533) (0.0526) Contiguity 0.208 0.129 (0.152) (0.155) Language 0.709 *** 0.718 *** (0.249) (0.242) Common legal origin 0.297 *** 0.291 *** (0.0683) (0.0672) Observations 26,514 26,514 R-squared 0.891 0.885 Exporter-year FE YES YES Importer-year FE YES YES Intermediate goods, which are processed and 'Foreign Shipped footprint' back Variables (6) (7) Distance -2.234 *** -0.834 *** (0.103) (0.0346) Contiguity 0.360 0.252 ** (0.308) (0.118) Language 1.318 *** 0.637 *** (0.496) (0.181) Common legal origin 0.583 *** 0.185 *** (0.135) (0.0460) Observations 26,508 26,520 R-squared 0.882 0.924 Exporter-year FE YES YES Importer-year FE YES YES *** p < 0.01, ** p < 0.05, * p < 0.1. Note: Estimates of Eq. 1 (columns 1- 6) and Eq. 2 (column 7). The panel includes annual observations for 40 countries over the period 1995-2011. The models include robust variance estimates yielding heteroscedasticity-consistent standard errors, while distance is used as a clustering variable. Robust standard errors in parentheses. Table 3: Bilateral exports Variables Gross exports Final goods Intermediate goods (1) (2) (3) Distance -0.713 *** -0.654 *** -0.736 *** (0.0415) (0.0458) (0.0413) Contiguity 0.289 *** 0.336 *** 0.289 *** (0.101) (0.111) (0.103) Language 0.491 *** 0.388 * 0.529 *** (0.189) (0.220) (0.189) Common legal origin 0.166 *** 0.164 ** 0.167 *** (0.0582) (0.0644) (0.0582) FDI stock 0.161 *** 0.167 *** 0.165 *** (0.0140) (0.0152) (0.0142) Observations 9381 9381 9381 R-squared 0.931 0.927 0.924 Exporter-year FE YES YES YES Importer-year FE YES YES YES Intermediate goods, which are processed and Variables Consumed Exported Shipped back further (4) (5) (6) Distance -0.736 *** -0.724 *** -1.556 *** (0.0423) (0.0438) (0.0795) Contiguity 0.291 *** 0.231 ** 0.509 ** (0.102) (0.106) (0.205) Language 0.534 *** 0.496 *** 0.914 ** (0.191) (0.188) (0.430) Common legal origin 0.163 *** 0.150 ** 0.305 ** (0.0582) (0.0602) (0.123) FDI stock 0.163 *** 0.170 *** 0.314 *** (0.0144) (0.0147) (0.0268) Observations 9381 9381 9381 R-squared 0.926 0.916 0.905 Exporter-year FE YES YES YES Importer-year FE YES YES YES *** p < 0.01, ** p < 0.05, * p < 0.1. Note: Estimates of Eq. 5. The panel includes annual observations over the period 2000-2011. The models include robust variance estimates yielding heteroscedasticity-consistent standard errors and bilateral distance used as a clustering variable. Robust standard errors in parentheses. Table 4: Bilateral imports Variables Gross imports Final goods Intermediate goods (1) (2) (3) Distance -0.756 *** -0.717 *** -0.779 *** (0.0424) (0.0469) (0.0425) Contiguity 0.292 *** 0.286 *** 0.304 *** (0.103) (0.110) (0.104) Language 0.363 0.377 0.349 (0.240) (0.270) (0.231) Common legal origin 0.149 ** 0.180 ** 0.147 ** (0.0657) (0.0732) (0.0643) FDI stock 0.148 *** 0.139 *** 0.158 *** (0.0144) (0.0157) (0.0149) Observations 9381 9381 9381 R-squared 0.927 0.923 0.920 Exporter-year FE YES YES YES Importer-year FE YES YES YES Intermediate goods, which are processed and Variables Consumed Exported Shipped back further (4) (5) (6) Distance -0.770 *** -0.783 *** -1.557 *** (0.0427) (0.0472) (0.0788) Contiguity 0.316 *** 0.206 * 0.525 ** (0.103) (0.109) (0.206) Language 0.351 0.300 0.778 * (0.236) (0.237) (0.445) Common legal origin 0.145 ** 0.149 ** 0.327 *** (0.0645) (0.0679) (0.125) FDI stock 0.156 *** 0.165 *** 0.321 *** (0.0149) (0.0163) (0.0272) Observations 9381 9381 9381 R-squared 0.923 0.911 0.922 Exporter-year FE YES YES YES Importer-year FE YES YES YES *** p < 0.01, ** p < 0.05, * p < 0.1. Note: Estimates of Eq. 6. The panel includes annual observations over the period 2000-2011. The models include robust variance estimates yielding heteroscedasticity-consistent standard errors, while distance is used as a clustering variable. Robust standard errors in parentheses. Table 5: Bilateral import content of exports Variables (1) (2) Distance -0.747 *** -0.624 *** (0.0347) (0.0322) Contiguity 0.327 *** 0.300 *** (0.104) (0.0893) Language 0.509 ** 0.316 (0.211) (0.192) Common legal origin 0.156 *** 0.0903 * (0.0576) (0.0519) FDI stock 0.116 *** (0.0118) Observations 9381 9381 R-squared 0.933 0.940 Exporter-year FE YES YES Originator-year FE YES YES *** p < 0.01, ** p < 0.05, * p < 0.1. Note: Estimates of Eq. 7. The panel includes annual observations over the period 2000-2011. The models include robust variance estimates yielding heteroscedasticity-consistent standard errors, while distance is used as a clustering variable. Robust standard errors in parentheses. Table 6: FDI augmented gravity model for exports; PPML Variables Gross Final Intermediate exports goods goods (1) (2) (3) Distance -0.499 *** -0.508 *** -0.504 *** (0.0309) (0.0358) (0.0311) Contiguity 0.424 *** 0.444 *** 0.407 *** (0.0813) (0.0861) (0.0868) Lanquaqe -0.004 -0.359 * 0.162 (0.169) (0.194) (0.170) Common 0.230 *** 0.288 *** 0.214 *** legal (0.0636) (0.0700) (0.0641) origin FDI stock 0.145 *** 0.134 *** 0.148 *** (0.0214) (0.0249) (0.0201) Observations 9381 9381 9381 R-squared 0.928 0.931 0.925 Exporter- YES YES YES year FE Importer- YES YES YES year FE Intermediate goods, which are processed and Variables Consumed Exported Shipped further back (4) (5) (6) Distance -0.497 *** -0.463 *** -0.976 *** (0.0315) (0.0371) (0.0699) Contiguity 0.417 *** 0.285 *** 0.581 *** (0.0853) (0.0948) (0.171) Lanquaqe 0.149 0.355 * -0.502 (0.168) (0.185) (0.387) Common 0.175 *** 0.204 *** 0.854 *** legal (0.0658) (0.0684) (0.142) origin FDI stock 0.147 *** 0.152 *** 0.416 *** (0.0209) (0.0211) (0.0485) Observations 9381 9381 9381 R-squared 0.931 0.856 0.983 Exporter- YES YES YES year FE Importer- YES YES YES year FE *** p < 0.01, ** p < 0.05, * p < 0.1. Note: Estimates of Eq. 5. The panel includes annual observations over the period 2000-2011. The models include robust variance estimates yielding heteroscedasticity-consistent standard errors, white distance is used as a clustering variable. Robust standard errors in parentheses.