Policy Research Working Paper 9197 Determinants of Global Value Chain Participation Cross-Country Evidence Ana Fernandes Hiau Looi Kee Deborah Winkler Development Economics Development Research Group March 2020 Policy Research Working Paper 9197 Abstract The past decades witnessed big changes in international three decades. The evidence shows that factor endowments, trade with the rise of global value chains. Some countries, geography, political stability, liberal trade policies, foreign such as China, Poland, and Vietnam, rode the tide, while direct investment inflows, and domestic industrial capacity other countries, many in the Africa region, faltered. This are very important in determining participation in global paper studies the determinants of participation in global value chains. These factors affect participation in global value chains, based on empirical evidence from a panel value chains more than traditional exports. data set covering more than 100 countries over the past This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at afernandes@worldbank.org, hlkee@worldbank.org, and dwinkler2@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Determinants of Global Value Chain Participation: Cross-Country Evidence Ana Fernandesy Hiau Looi Keez Deborah Winklerx Keywords and JEL Classi cation: global value chain, factor endowments, trade policy, foreign direct investment, institutions; F13, F14, F23, O2 This paper is a substantial revision of the background paper, “Factors A¤ecting Global Value Chain Participation Across Countries,” prepared for the World Development Report 2020, Trading for Develop- ment in the Age of Global Value Chains. The authors thank Pol Antràs, Caroline Freund, Penny Goldberg, Aaditya Mattoo, and other colleagues for helpful comments. We are grateful to Alejandro Rojas for research assistance, Farid Toubal, Mary Hallward-Driemeier and Gaurav Nayyar for sharing data on English language and FDI. We acknowledge …nancial support from the WB’ s Multidonor Trust Fund for Trade and Develop- ment and the Strategic Research Program on Economic Development. The …ndings of this paper are those of the authors and do not necessarily represent the views of the World Bank, or its member countries. y Development Research Group, The World Bank, Washington, DC 20433, USA; Tel.: (202) 473-3983; e-mail: afernandes@worldbank.org. z Corresponding author. Development Research Group, The World Bank, Washington, DC 20433, USA; Tel.: (202) 473-4155; e-mail: hlkee@worldbank.org. x Macroeconomics, Trade and Investment Global Practice, The World Bank, Washington, DC 20433, USA; e-mail: dwinkler2@worldbank.org. 1 Introduction In the early 1990s, Argentina tried to develop a homegrown auto industry, hiding behind an s auto exports average tari¤ of more than 13 percent. Over the past two decades, Argentina’ have stagnated at a dismal 0.2 percent of global auto exports. Around the same period, s largest automakers, set up GM Poland to import General Motors (GM), one of the world’ Opel cars for the large Polish domestic market. In 1994, production activities of GM Poland s major auto exporting countries. Similar to started, and today, Poland is one of the world’ s electronics sector expanded sharply in less than a the auto industry in Poland, Vietnam’ s second largest smartphone exporter, producing 40 decade. Today, Vietnam is the world’ s global mobile phone products and employing 35 percent of its global percent of Samsung’ sta¤. Ten years ago, Vietnam barely exported electronics products. What sets Argentina, Poland, and Vietnam apart is their very di¤erent participation in global value chains (GVCs). In fact, the meteoric rises of Poland and Vietnam and the s World Trade Organization (WTO) accession faltering of Argentina are not unique. China’ in 2001 ushered a new wave of GVCs which gave rise to “Factory Asia” (Baldwin, 2016), while a large part of the Africa, South Asia and Latin America regions is being left behind with little integration into GVCs (see Figure 1). As discussed in Antràs (2016), what distinguishes GVC trade from traditional trade are the intense …rm-to-…rm interactions characterized by contracting and specialized products and investment. What factors determine GVC participation across countries? Do factors that a¤ect traditional trade have di¤erential impacts on GVC trade? This paper studies the determinants of GVC participation based on empirical evidence stamped from a panel data set covering more than 100 countries over the past three decades. The time period re‡ects the growing international fragmentation of production and the sample includes countries in all geographical regions and at all stages of development. This large cross-country variation makes the data set uniquely suitable to estimate the marginal impact of di¤erent potential determinants on GVC participation and assess their relative importance. While many papers explain the GVC phenomenon from various angles, the strategy of this paper is to provide a uni…ed empirical framework to test jointly the role of the di¤erent determinants highlighted 1 in the literature as being important for trade in general, or for GVC trade in particular.1 We focus on widely-used measures of backward GVC participation, mainly drawing on the EORA database, which capture the import content of exports, that is, how much imported materials are used in countries’ exports. We consider seven broad types of determinants emphasized in the trade literature: (i) factor endowments, (ii) geography, (iii) domestic industrial capacity, (iv) trade policy and foreign direct investment (FDI), (v) institutional quality, (vi) connectivity, and (vii) macroeconomic factors. However, establishing causality is challenging in the cross-country setting as some of the potential determinants, such as tari¤s and FDI in‡ows, are endogenous to GVC participa- tion. To this end, we rely on instrumental variables (IVs) estimation in the cross-country regressions, and a di¤erence-in-di¤erence framework following Rajan and Zingales (1998) in the cross-country cross-sector regressions. The use of IVs also mitigates biases in the construction of the GVC participation variables, which hinges on the use of international input-output tables that may have measurement problems. Our IV selections are guided by existing trade theories and strong empirical evidence. The results from our cross-country decadal panel and our cross-country IV regressions show that the key determinants of backward GVC participation are, in order of importance, factor endowments, geographical location, political stability, tari¤s and FDI in‡ows, and domestic industrial capacity. We show that our IVs have the expected coe¢ cients and sig- ni…cant explanatory power with high …rst-stage F-statistics. Our second-stage IV coe¢ cient estimates are larger in magnitude than the corresponding least squares coe¢ cient estimates suggesting that measurement errors and reverse causality are important in biasing the least squares results. Our …ndings are robust to the inclusion of alternative determinants and con- trols. Moreover, broad sector analysis suggests that the overall …ndings are largely driven by manufacturing. Cross-country cross-sector panel regression analysis in a di¤erence-in- di¤erence framework further con…rms that institutional quality, factor endowments, trade policies, FDI, and connectivity matter for GVC participation. Finally, we extend our analysis to show that the most determinants have larger impacts 1 For theoretical models on GVCs and international production fragmentation, please refer to Feenstra and Hanson (1997), Feenstra (1998), Antràs and Helpman (2004), Romalis (2004), Nunn (2007), Antràs and Helpman (2008), Chor (2010), Antràs (2016), and Antràs and De Gortari (2020). 2 on GVC trade than on traditional trade. We analyze forward GVC participation, which s export production. captures the domestic value added that is used in a bilateral partner’ Results again con…rm the importance of these factors in explaining GVC participations. This paper contributes to several strands of the trade literature. First, the paper re- lates to the empirical literature on the determinants of GVC participation.2 We expand the existing analyses in terms of country, time, and variable coverage, but also with regards to the methodology by addressing potential endogeneity concerns using IV and di¤erence- in-di¤erence estimations. By embedding the determinants in a uni…ed framework, we are able to estimate their marginal contributions and identify their relative importance to GVC participation. Second, the paper relates to the literature on the measurement of GVC par- ticipation.3 Our study uses an improved measure of GVC participation which avoids a double-counting problem.4 Finally, the paper contributes to the literature on international production sharing and GVCs.5 This paper is organized as follows. In Section 2, we de…ne the GVC participation measures and motivate our selection of determinants. We present the baseline empirical speci…cations in Section 3. The regression results on the determinants of GVC participation are shown in Section 4, followed by robustness checks in Section 5. Section 6 concludes. We refer the readers to the Appendix for the data descriptions and sources. 2 GVC Measures and Determinants 2.1 De…ning and Measuring GVC Participation From imports of pistons used as intermediates in car manufacturing in Morocco to Chilean exports of copper used in refrigerators produced by …rms in China and Mexico, GVC partic- 2 See Baldwin and Taglioni (2012), Brooks and Ferrarini (2012), Noguera (2012), Blyde (2014), Cheng et al. (2015), Kowalski et al. (2015), United Nations (2015), Buelens and Tirpak (2017), Balié et al. (2019), and Ignatenko et al. (2019). While the literature is vast, the studies so far establish strong correlations but no causal links. 3 See e.g., Hummels et al. (2001), Koopman et al. (2014), Johnson and Noguera (2017), and Miller and Temurshoev (2017). 4 The measure was introduced and developed in Borin and Mancini (2015, 2019). 5 See Baldwin (2012) for a discussion on GVC participation as a new industrialization strategy and Feenstra (1998) and Timmer et al. (2014) for reviews of the literature on foreign outsourcing and on GVCs. 3 ipation is multifaceted and diverse across countries. This paper mainly focuses on backward GVC participation measures that have been recently developed in the empirical literature, drawing on newly available international input-output tables, including the EORA database, the Trade in Value Added (TIVA) database and the World Input-Output Database (WIOD). s backward GVC participation measures the import content of its exports A country’ s total gross exports. Those imported inputs are predominantly relative to the country’ made up of foreign value added but can also contain domestic value added which has been previously exported. The import content of export measure was …rst introduced by Hummels et al. (2001) using national input-output tables and was more recently computed by Borin and Mancini (2015) and Wang et al. (2013, 2016) based on international input-output tables. This measure takes into account indirect e¤ects (known as the Leontief inverse matrix) where imported inputs are embodied in domestic output, sometimes at several stages, before being used as inputs for exports. s forward As a robustness check, we also consider forward GVC participation. A country’ s GVC participation measures the domestic value added in exports that is used by the country’ s total gross bilateral partner countries for export production as percent of the country’ exports. In other words, it captures the portion of domestic value added that is not directly consumed by the bilateral partner (which is the …nal stage in the value chain). For example, …nal apparel exports from Bangladesh that are exported to and consumed in the United States (US) would not be accounted for in this measure. The measure was developed in Borin and Mancini (2019) based on a bilateral source-based decomposition of exports.6 Both backward and forward GVC participation measures are such that their construction ensures that trade ‡ows cross at least two country borders –a key characteristic of a GVC. The country-level backward and forward GVC participation measures are additionally decomposed into measures for four broad sectors: agriculture, mining, manufacturing, and services. In order to obtain these measures, we split the country-level numerator into these 6 For a technical discussion of these and other GVC measures, see Aslam et al. (2017) and Borin and Mancini (2015, 2019). Koopman, et al. (2014) propose an alternative forward GVC participation measure: a country’s domestic value added that is used by third countries for export production. We do not use this measure as it double-counts the domestic value added component if it is used in the downstream processes of multiple countries (e.g., petroleum) and thus can counterintuitively exceed 1 as a share of the country’s gross exports. 4 s total exports. Using these alternative measures four sectors and divide each by the country’ by broad sector sheds light on which sectors are driving the aggregate cross-country results. Finally, both types of GVC participation measures are also available at the country-sector level for an analysis of cross-country cross-sector determinants of GVC participation. In addition to studying the “intensity”of GVC participation, as measured by the shares in gross exports, we are also interested in the levels of GVC participation (i.e., the numerator of the “intensity”) and in the levels of gross exports. Comparing the factors that a¤ect GVC participation shares with their in‡uence on GVC participation levels and on export levels indicates which determinants matter beyond traditional exports. The main data source for the GVC participation measures is the EORA database de- scribed in Lenzen et al. (2013), which covers 190 countries over the period 1990-2015 using a 26-sector harmonized classi…cation.7 The decomposition of country-level GVC participation measures into agriculture, mining, manufacturing, and services also draws on the EORA database and our cross-country cross-sector analysis relies on GVC participation measures for eight manufacturing sub-sectors in that database, listed in the Appendix.8 There are several limitations to using international input-output tables to measure GVC participation. A major limitation is related to the distorted statistics that international input-output tables are based on.9 In particular for lower-income countries that often do not produce national supply-use tables, the EORA data are based on interpolations and estimations, and therefore subject to measurement errors. There is also a concern regarding the lack of …rm-level heterogeneity in the conceptual measure of GVC participation which may also lead to errors.10 All these measurement errors in the EORA GVC data may bias 7 The number of countries in our estimating sample is at most 121 due to missing data on determinants and to the exclusion of countries with problematic (negative) values for the GVC participation measures (Liberia, Moldova, Mauritius, South Sudan, Sudan, and Zimbabwe). 8 In robustness checks we use as alternative sources the TIVA database 2016 and 2018 editions covering 64 countries over the periods 1995-2011 and 2005-2015, respectively, the WIOD 2013 release covering 40 countries in the period 1995-2011 and the WIOD 2016 release covering 43 countries in the period 2000-2014, described in Timmer et al. (2015, 2016). Further details, summary statistics and correlations among the GVC participation measures are provided in the Appendix. One interesting fact to note from the correlations is that the cross-country correlation between backward and forward GVC participation is negative. 9 United Nations (2019) describes the major limitations in detail and proposes solutions to help substan- tially improve the input-output coe¢ cients used in the current international input-output tables. 10 The GVC literature emphasizes governance of lead …rms and their relationship with suppliers as a major feature of GVC trade but such information is unavailable in current international input-output tables. 5 least-squares coe¢ cients toward zero in regressions. We will use IVs to address this issue. 2.2 Determinants of GVC Participation This section motivates the choice of determinants of GVC participation examined in the paper. We consider seven broad types of determinants: (i) factor endowments, (ii) geogra- phy, (iii) domestic industrial capacity, (iv) trade policy and FDI, (v) institutional quality, (vi) connectivity, and (vii) macroeconomic factors. The de…nitions, detailed data sources, summary statistics, and correlations of the variables are provided in the Appendix. 2.2.1 Factor Endowments Factor endowments are crucial in determining international specialization (Heckscher-Ohlin model) and may also shape the positioning of countries in GVCs.11 We focus on three types of endowments: (a) natural resources, (b) labor, which is separated into low-skilled and middle to high-skilled, and (c) capital. An abundance of natural resources in a country, such as copper and iron ore, is naturally linked to high forward GVC integration, since agricultural products and commodities are used in a variety of downstream production processes that typically cross several borders. Low-skilled labor in lower-income countries is often an entry point to downstream assembly-type stages of production associated with high content of s exports (high backward GVC participation) and exports of imported inputs in a country’ …nal goods (low forward GVC participation). But advancing to more skill-intensive tasks in the value chain increases forward GVC participation. Finally, these production processes require capital investments, hence capital endowment should induce GVC participation. 2.2.2 Geography Trade costs due to geography and distance can determine which country to import products s positioning in GVCs.12 In sequential (or snake-like) GVCs, from and can shape a country’ trade costs compound along the value chain and have a higher incidence on downstream 11 Empirical evidence on the in‡ uence of factor endowments on country-sector trade patterns is widely available following the study by Romalis (2004). 12 See, for example, Eaton and Kortum (2002) and Antràs and De Gortari (2020). 6 stages than on upstream stages. This may encourage more remote countries to specialize in upstream stages and more central countries to specialize in downstream stages. Ine¢ cient transport and logistics services and weak competition in these services amplify trade costs in many manufacturing GVCs with multiple border crossings and can o¤set other competitive advantages like low labor costs. Empirical evidence shows that bilateral GVC links are strongly positively correlated with geographic proximity and that countries’backward GVC participation is negatively associated with their higher distance to the closest manufacturing hub.13 We expect geography - measured by geographical distance to GVC hubs China, Germany, and the US - to play an important role for GVC participation. 2.2.3 Domestic Industrial Capacity It is well established by gravity models that countries with a larger domestic industrial ca- pacity have more traditional trade.14 However, for GVC trade, the relationship between domestic industrial capacity and GVC participation is not clear. On the one hand, to min- imize cross-hauling of semi-processed goods in di¤erent stages, countries often specialize in contiguous stages of production in GVCs. Countries with a larger domestic industrial capac- ity may have a larger set of contiguous stages which reduce the use of imported inputs relative to domestic inputs in their exports. This may lead to lower backward GVC participation. On the other hand, countries with a larger domestic industrial capacity may demand more …nal goods for domestic consumption, which may lead them to specialize in downstream stages of production embodying more foreign value added and thus could increase backward GVC participation. Finally, a larger domestic industrial capacity implies a larger domestic supplier base which reduces search frictions and facilitates the replacement of domestic sup- pliers in face of production disruptions and may increase domestic value added and forward GVC participation.15 Thus, the overall e¤ect of domestic industrial capacity - measured by domestic manufacturing value added - on GVC participation is ambiguous and can only be determined empirically. 13 See Kowalski et al. (2015) and Buelens and Tirpak (2017). 14 See, for example, Arkolakis et al. (2012). 15 See Kee (2015) for evidence of this mechanism in Bangladesh. 7 2.2.4 Trade Policy and Foreign Direct Investment Trade policy and FDI are important for traditional trade but they may play even larger roles for GVC trade, as intermediates and semi-…nished products cross international borders multiple times. Regulatory barriers on imports and exports such as tari¤s or quotas increase trade costs, with consequences for countries’participation and positioning in GVCs. Reduc- ing such barriers can have an ampli…ed bene…t for internationally fragmented production - especially when production stages are organized sequentially across borders - by lowering not only the price of …nal goods but also the input costs faced.16 Tari¤s imposed by partner countries can also increase the costs of exports.17 Emerging evidence shows that tari¤s on imports of …nal goods and intermediates and tari¤s faced in export markets are negatively associated with GVC participation.18 Deep preferential trade agreements (PTAs) go beyond traditional market access issues and include policy areas such as movement of capital, in- vestment, visas, and intellectual property rights (World Bank, 2019). A strong role of deep PTAs in fostering GVC participation is also shown in recent studies.19 Similarly, countries can attract FDI to overcome relative scarcity in capital, technology, and knowledge, and thus integrate into GVCs. When tight control over foreign production processes is necessary (perhaps because of weak contractual enforcement or weak protection s of intellectual property), lead …rms might prefer vertical integration of suppliers over an arm’ length relationship, resulting in FDI ‡ows and intra-…rm trade. Empirical evidence suggests that openness to FDI is positively associated with backward GVC participation.20 Trade 16 See Yi (2003, 2010) and Antràs and De Gortari (2020) for large magni…cation impacts of tari¤s when intermediates trade and multistage production are considered in general equilibrium trade models. Caliendo and Parro (2015) add to trade in intermediates also linkages across sectors and derive larger welfare gains from trade liberalization relative to models with no trade in intermediates and sectoral linkages. 17 Trade preferences given to Bangladesh by the EU induced greater …rm entry into exports to the EU and then growth in exports to all markets (Cherkashin et al., 2015). But evidence shows that in the long- run preferential access per se is insu¢ cient for export success. Complementary domestic policies are needed, including low import tari¤s, reduced regulatory burden, and enhanced connectivity (Fernandes et al., 2019a). 18 See Cheng et al. (2015) and Kowalski et al. (2015). The importance of lower tari¤s on intermediates to foster export performance is also con…rmed at a micro level (Bas and Strauss-Kahn, 2015; Pierola et al., 2018). 19 See Ore…ce and Rocha (2014), Kowalski et al. (2015), Johnson and Noguera (2017), and Laget et al. (2018). 20 See Kowalski et al. (2015) for FDI openness at the country level and Buelens and Tirpak (2017) for FDI stocks at the bilateral level. Cheng et al. (2015) show that higher FDI restrictiveness is related to lower GVC participation in low-tech manufacturing. 8 policies and FDI have also been identi…ed as important factors for moving up GVCs, based on …rm-level evidence for China and Bangladesh.21 Hence, we expect tari¤s and FDI (as well as deep PTAs) to be signi…cant in determining GVC participation. But these determinants may be subject to endogeneity concerns that we address in detail in Section 3.1. 2.2.5 Institutional Quality As discussed in Antràs (2016), what distinguishes GVC trade from traditional trade are the intense …rm-to-…rm interactions characterized by contracting and specialized products and investment. Weak contract enforcement is thus a signi…cant deterrent not only for tradi- tional trade, but also for GVC trade.22 Because the performance of a GVC depends on the strength of its weakest link, production delays driven by weak contract enforcement might be particularly harmful in GVCs. In addition, the presence of relationship-speci…c investments (e.g., the customization of products) and the exchange of large ‡ows of intangibles (such as technology, intellectual property and credit) reinforces the potential role of institutional quality as a signi…cant determinant of relational GVC participation. Some emerging evi- dence …nds a correlation between a stronger rule of law and stronger GVC integration.23 We expect institutional quality to be important in determining GVC participation. 2.2.6 Connectivity Transport costs remain, according to surveys of developing country suppliers, the main obstacle to entering, establishing, or upgrading in GVCs.24 The geographic centrality of a country can attract downstream production stages in GVCs. But geographic centrality is more related to centrality in the transport network than to distance. Thus, logistics and communication infrastructure, port and customs e¢ ciency, and information technology networks could be important for trade in general and especially for GVC trade in particular. 21 See Kee (2015) and Kee and Tang (2016). 22 A body of work establishes institutional quality as a comparative advantage factor in determining export patterns: Acemoglu et al. (2007), Levchenko (2007), Nunn (2007), Costinot (2009), and Chor (2010). 23 See Kowalski et al. (2015). 24 See OECD and WTO (2013). Hummels and Schaur (2013) …nd that a day of delay in transit due to di¤erent transport mode choice has a tari¤ equivalent of 0.6 to 2.1 percent, and the most sensitive ‡ ows are for a type of GVC trade, i.e., trade in parts and components. Similar magnitudes for the cost of a one-day delay in inland transit are found by Djankov et al. (2010). 9 The quality of the national road infrastructure also matters for timely delivery to global markets. Moreover, the use of the internet and a common language could also facilitate GVC participation. Studies show a stronger role of logistics performance for trade in parts and components than for trade in …nal goods, and provide evidence that unpredictable land transport keeps most Sub-Saharan African countries out of GVCs.25 Given that our sample includes a wide range of countries with diverse logistics infrastructure and language used, we are able to examine the impact of connectivity on GVC trade. 2.2.7 Macroeconomic Factors Macroeconomic factors, in particular related to real exchange rates, can play a role for GVC participation. The degree of …nancial development of countries is a source of comparative advantage but there is still limited understanding of whether …nancial factors a¤ect GVC participation.26 Our sample coverage of developed and developing countries allows us to study how these macroeconomic factors may impact trade and GVC participation. 3 Empirical Speci…cation We examine the factors that in‡uence GVC participation by exploiting variation across countries and over time in GVC participation and in determinants. We estimate the impact of country decadal averages of the determinants on country decadal average GVC participation by least squares between e¤ects (LS-BE). The main speci…cation is given by: Yct = 0 + 1 Xct + It + "ct ; (1) 25 See Ansón et al. (2017) for estimates of the sensitivity of bilateral trade in parts and components and in …nal goods to logistics performance. See Christ and Ferrantino (2011) for evidence on Sub-Saharan Africa. A positive correlation between broad infrastructure - including communication, electricity, roads, and power - and overall GVC participation in manufacturing is provided by Cheng et al. (2015). Linguistic proximity is shown to matter for bilateral GVC links in Buelens and Tirpak (2017) and Ignatenko et al. (2019). At a micro level, evidence on the role of unpredictability in imports’border clearance times, regional road density and internet access for …rm export performance are shown, respectively, by Vijil et al. (2019), Rodríguez-Pose et al. (2013) and Fernandes et al. (2019b). 26 See Beck (2003), Manova (2008), and Manova (2015). Exchange rate volatility is shown to be negatively related to bilateral GVC trade by Ignatenko et al. (2019). 10 where c and t stand for country and decade subscripts, respectively. Y is a measure of GVC participation in shares or in levels or a measure of gross exports, and X is a vector including the determinants described in Section 2.2, It are decade …xed e¤ects, and " is an independent and identically distributed error (i.i.d). We use LS-BE estimation for the cross-country panel regression speci…ed in Equation (1), where the panel includes up to three observations per country, each covering a decadal average. The coe¢ cients are identi…ed via cross-country variations in GVC participation and the determinants within a decade. There are two justi…cations for this approach. First, GVC participation and some de- terminants change very slowly within countries from year to year. This is the key reason why we do not rely on a country-year panel and within e¤ects estimation for Equation (1).27 Decadal averages of GVC participation and determinants exhibit more meaningful variation than year-to-year observations and they may wash out measurement issues in GVC partic- ipation due to errors in input-output tables (see Section 2.1). Moreover, the use of decadal averages allows to maximize country coverage as countries remain in the estimating sample s years. even if GVC participation or some determinants are observed only in a few of decade’ Second, in contrast to small within-country changes, GVC participation measures and deter- minants exhibit large cross-country variability. LS-BE estimation exploits this variability to identify more precisely the impacts of the di¤erent determinants on GVC participation. The decade …xed e¤ects in Equation (1) allow to account for technological shocks or the global …nancial crisis a¤ecting all countries. We consider the following dependent variables in Equation (1): (i) the share of backward GVC participation in gross exports, which captures the intensity of GVC trade relative to traditional exports; (ii) the level of backward GVC participation (logs); and (iii) gross exports (logs). A statistically signi…cant positive coe¢ cient of a factor in the GVC partici- pation share regression indicates that the factor has a stronger impact on the level of GVC participation relative to traditional exports. An insigni…cant coe¢ cient of a factor in the GVC participation share regression implies the factor does not have a di¤erential impact on GVC trade relative to traditional exports. We also decompose backward country-level 27 Most coe¢ cients in Equation (1) cannot be precisely estimated in a within e¤ects cross-country panel regression. 11 GVC participation measures into four broad sectors - agriculture, mining, manufacturing, and services - which are used as separate dependent variables in Equation (1). However, some determinants, namely tari¤s and FDI in‡ows, could be endogenous or simultaneously determined with GVC participation. There are two opposing forces at play regarding reverse causality from GVC participation to tari¤s and FDI in‡ows. First, coun- tries that participate in GVCs may lower tari¤s and attract FDI so GVC …rms can have access to cheaper imported inputs or better domestic inputs produced by FDI …rms. This force confounds with the direct impact of tari¤s and FDI on GVC participation and causes their LS-BE coe¢ cient estimates to be too large in magnitude. Second, countries that par- ticipate in GVCs may have political economy reasons to raise tari¤s and restrict FDI in‡ows in order to protect domestic …rms from competition. This force causes LS-BE estimates for tari¤s and FDI to be too small in magnitude. To add to ambiguous reverse causality, measurement errors in the construction of GVC participation variables due to the usage of international input-output tables may bias LS-BE estimates towards zero. Ultimately, de- pending on which force dominates, LS-BE estimates for the impacts of tari¤s and FDI on GVC participation could be biased upward or downward. 3.1 Instrumental Variables To address these endogeneity and measurement error biases, we rely on IVs which isolate the e¤ects of tari¤s and FDI in‡ows on GVC participation. To obtain consistent estimates, the IVs should be jointly signi…cant in explaining tari¤s and FDI in‡ows with meaningful …rst- stage coe¢ cients and F-statistics, and at the same time, the IVs should not be correlated with the second-stage regression errors in order to satisfy exclusion restrictions. In other words, the IVs should be relevant and exclusion restrictions require that these IVs only a¤ect GVC participation through tari¤s and FDI and do not have signi…cant direct impacts on GVC participation once tari¤s and FDI are included in the second-stage regressions, conditional on all other right-hand side variables.28 28 According to Wooldridge (2012), instrument exogeneity means that the IV should have no partial ef- fect on the dependent variable, after controlling for the right-hand side variables, and the IV should be uncorrelated with the omitted variables. 12 However, in this aggregate cross-country setting, it is highly challenging to …nd reasonable IVs that are both relevant and meet the exclusion restrictions. To this end, we rely on existing trade theories and on solid empirical evidence to guide our choice of meaningful IVs and we conduct econometric tests to establish the strength of our identi…cation. Below we describe the set of IVs that we use to explain average tari¤s imposed on manufacturing products and FDI in‡ows for more than 100 countries in our regressions. The detailed data sources for the IVs are provided in the Appendix. 3.1.1 Import Elasticity Classic trade theory since Bickerdike (1907) asserts that the optimal tari¤ set by a welfare- maximizing government for a country with market power is higher than zero, while the optimal tari¤ for a small open economy is zero. This positive relationship between tari¤s and the market power of a country that is well-established empirically has long been referred to as the terms-of-trade theory of trade policy.29 For our analysis we need a proxy for market power that is not a¤ected by GVC participation in order to satisfy the exclusion restriction. One way to assess the market power of an importing country is to estimate the export supply elasticity it faces as is done by Broda et al. (2008) for a set of 15 countries. Our choice is to use import demand elasticity estimates from Kee et al. (2008) that are available for more than 100 importing countries thus ensuring very good coverage of our sample.30 These elasticities represent the slope of the import demand function of the countries and are based on a GDP function that controls for countries’ factor endowments. While the elasticities are estimated at the country-HS 6-digit product level, our measure of the import demand s manufacturing sector is obtained as an average of the elasticities elasticity for each country’ of all its manufacturing products. Countries with more elastic import demand may impose lower tari¤s in order to minimize the deadweight loss which leads to a negative relationship between import elasticity and tari¤s. The exclusion restriction requires that the import elasticity does not have a direct impact 29 See Broda et al. (2008) and Bagwell and Staiger (2011) for evidence of this relationship for a sample of pre-WTO accession countries. 30 These elasticity estimates are used to study the e¤ect of WTO tari¤ commitments by Bagwell and Staiger (2011) and the general equilibrium impact of trade liberalizations in GTAP models. 13 on GVCs other than through tari¤s, conditioning on all other right-hand side variables. In theory, the import demand elasticity may depend on preferences, income and domestic substitutes. Given that the estimated elasticity is multilateral in nature (not bilateral partner country speci…c), and that we control for domestic industrial capacity and factor endowments which both may a¤ect domestic substitutes as well as institutional quality which may re‡ect income and preferences, it is reasonable to assume that the import elasticity meets the exclusion restriction. 3.1.2 Population Terms-of-trade theory also implies that countries with a larger total population may have more market power and thus impose higher tari¤s. Hence, total population is relevant in explaining tari¤s. We allow for the e¤ects of import elasticity and total population to be related, so their interaction term is also included as an additional IV. Moreover, total population plays a dual role in our IV strategy as more populous countries may attract larger FDI in‡ows to serve their domestic markets. Such FDI in‡ows will be either import- substituting or market-seeking and may lead to further exporting and GVC participation. Hence, total population is relevant for explaining FDI in‡ows. s total population, and populous countries GVC participation should not a¤ect a country’ (such as India and Brazil) do not necessarily have higher GVC participation. There may be concerns that population could a¤ect GVC participation through other channels, such as business environment, domestic suppliers and labor. However, since institutional quality, domestic industrial capacity and factor endowments are included in the regressions, the concerns may be mitigated. Thus, conditioning on these right-hand side variables, it is plausible that population a¤ects GVC participation only through tari¤s and FDI in‡ows and satis…es the exclusion restriction. 3.1.3 Statutory Corporate Tax Rates Foreign a¢ liates of a multinational company (MNC) are subject to corporate income taxes in the FDI host country. Higher corporate taxes in the host country reduce the after-tax return to investment for the MNC and hence may discourage stronger investment ‡ows. Gravity 14 regressions for MNC a¢ liates’location choices and investments drawing on economic geog- raphy models provide extensive evidence that high corporate income tax rates are negatively associated with FDI in‡ows.31 Using statutory corporate tax rates as an instrument for FDI in‡ows could satisfy the exclusion restriction, as a priori there is no clear relationship between corporate tax rates and GVC participation other than through the FDI channel, conditioning on institutional quality, domestic industrial capacity and endowments. 3.1.4 Transitional Economy Status The late 1980s ushered the fall of communism and gave rise to a group of countries that embraced market capitalism and abandoned central planning, collectively referred to as transitional economies (IMF, 2000). Most of these economies are the former Soviet Union and its satellite states, including Poland, Hungary, and Bulgaria, while other countries are in East Asia, such as China and Vietnam. These transitional economies opened up engaging in tari¤ and FDI liberalization, and several were subsequently successful in participating in GVCs. We use the transitional economy status as an instrument for tari¤s and FDI in‡ows, and since a priori GVC participation is not correlated with the transitional economy status of the countries, that status satis…es the exclusion restriction, conditioning on institutional quality, factor endowments and domestic industrial capacity. 3.1.5 Instrumental Variables Summary and Tests In summary, we have two endogenous variables in our reduced form speci…cation Equation (1): tari¤s and FDI in‡ows. We use …ve excluded exogenous variables as instruments: import elasticity, total population, the interaction between import elasticity and total population, statutory corporate tax rates, and transitional economy status. We test for weak IVs in the …rst-stage regressions, based on the Kleibergen-Paap Wald F-statistic, which indicates that these IVs jointly have signi…cant explanatory power for tari¤s and for FDI in‡ows if the F-statistic is higher than the critical value given by Stock and Yogo (2005). We also test that (the instrumented) tari¤s and FDI in‡ows can jointly explain GVC participation, based on 31 See e.g. Desai et al. (2004), Head and Mayer (2004), and Mutti and Grubert (2004). 15 the weak-instrument-robust Anderson-Rubin Wald test in the second-stage regressions. We expect the IV estimates of the coe¢ cients on tari¤s and FDI in‡ows to be larger in magnitude than the LS-BE estimates of those coe¢ cients if measurement errors and political economy forces dominate the confounding reverse causality forces.32 4 Main Results 4.1 Least Squares Between Regressions Starting with the determinants of backward GVC participation shares, Column (1) of Table 1 shows the baseline LS-BE estimates of Equation (1). Most coe¢ cients have the expected signs and are signi…cant. Factor endowments jointly matter for backward GVC participation, with a strong F-statistic of 6.84, which is signi…cant at the 99% level. Larger land and/or natural resources endowments are linked to lower backward GVC participation shares, while abundance in capital is associated with higher backward GVC participation shares. But geography and domestic industrial capacity also matter. A shorter distance to the GVC hubs is positively correlated with backward GVC participation. Countries with larger do- mestic industrial capacity (i.e., a potentially larger pool of domestic suppliers) exhibit lower backward GVC participation shares, as domestic inputs may be used to replace imports. Countries’ openness to trade and FDI is also a predictor for the intensity of backward GVC participation. Lower tari¤s and larger FDI in‡ows are associated with higher back- ward GVC participation shares across countries. Better institutional quality measured by a higher score in the political stability index is linked to higher backward GVC participa- tion. Exchange rate appreciation is unrelated to backward GVC participation, which is not surprising, given that appreciation stimulates imports by reducing its prices, but can also reduce export competitiveness due to higher export prices.33 The R-squared in Column (1) 32 We also tried using lagged tari¤s and FDI, as well as neighboring countries tari¤s and FDI as instruments. In these cases, the second-stage results are very similar, with IV estimates of the coe¢ cients on tari¤s and FDI being larger in magnitude than the corresponding LS-BE ones. Reed (2015) shows that lagged endogenous variables are valid IVs if the lagged endogenous variables do not a¤ect current period dependent variable and if the lagged endogenous variables are correlated with the current period endogenous variables. These results are available upon request. 33 This result is consistent with the …nding of Amiti et al. (2014) who show that low aggregate exchange rate 16 indicates that the di¤erent determinants considered explain more than half of the variation in backward GVC participation shares across countries. In Columns (2)-(6) of Table 1, we modify the baseline regression by including alternative or additional control variables or by changing the estimating sample. First, in Column (2) we replace exchange rate appreciation by a measure of misalignment whose e¤ect on GVC participation remains insigni…cant. Next, in Column (3) we include alternative measures of trade openness related to membership in PTAs (NAFTA, EU, MERCOSUR, ASEAN) and deep integration e¤orts. The estimated coe¢ cients show that countries’EU or ASEAN membership are linked to signi…cantly higher backward GVC participation shares. Third, to understand whether the patterns identi…ed are driven by a particular type of countries, we drop high-income countries as de…ned by the World Bank income classi…cation, from the estimating sample in Column (4).34 This sample exhibits substantially less cross-country variation but several patterns identi…ed in Column (1) still hold, namely the strong negative impact of tari¤s and distance to GVC hubs and the strong positive impact of political stability on backward GVC participation. Fourth, we add as a proxy for non-tari¤ trade costs and connectivity a measure of the time to import in Column (5) and a control for the importance of females in the labor force in Column (6). The baseline results in Column (1) remain robust. We also include other control variables, such as the credit to GDP ratio, measures of the quality of logistics infrastructure, internet infrastructure, the prevalence of spoken English in the regressions, but these variables were all insigni…cant.35 To understand which sectors within the economy are driving the aggregate results pre- sented in Table 1, we estimate Equation (1) separately for each of the backward GVC participation measures by broad sector - agriculture, mining, manufacturing, and services. pass-through and disconnect observed in the data is due to the fact that large exporters are simultaneously large importers, with high-market-share and high-import-intensity. 34 Since many countries change income status during our sample period, we use a time-varying World Bank income classi…cation to identify high-income countries. 35 These results are available upon request. A di¤erent type of robustness checks uses GVC participation measures based on alternative TIVA and WIOD databases. These databases’time, and importantly, country coverage is much more limited than EORA’ s, as they focus on high- and middle-income countries. Thus, the cross-country variation in GVC participation and in determinants is much smaller. But the broad patterns of results are similar to those in Tables 1 and 6 though weaker in signi…cance. The estimates that remain signi…cant for backward GVC participation are the negative impact of natural resource endowments, distance to GVC hubs and domestic industrial capacity and the positive impact of political stability. 17 The results are presented in Table 2 and show clearly that the …ndings at the aggregate level are largely driven by GVC participation in manufacturing. All the determinants identi…ed above in‡uence backward GVC participation in manufacturing in the same direction as in our baseline estimates. Low-skilled labor positively a¤ects GVC participation in agriculture, while middle or high-skilled labor negatively a¤ects GVC participation in mining. GVC participation in services responds positively to stronger political stability, but negatively to domestic industrial capacity and to the abundance of land. Overall the results of LS-BE regressions suggest that tari¤s and FDI in‡ows are important in determining GVC participation of a country. Factor endowments are also important, and so are political stability, proximity to GVC hubs as well as domestic industrial capacity. 4.2 Two-Stage Least Squares Regressions The estimates in Table 1 could be inconsistent due to potential endogeneity of tari¤s and FDI in‡ows and to potential measurement error in the GVC participation variable. To address these concerns, we adopt an IV strategy relying on import elasticity, total population, their interaction, corporate tax rates, and transitional economy status as our …ve IVs. We estimate the baseline between e¤ects speci…cation in Column (1) of Table 1 by two-stage least squares. The corresponding …rst-stage between regressions are presented in Columns (1) and (2) of Table 3. All IVs exhibit the expected signs and statistical signi…cance. Countries with more elastic import demand have lower tari¤s, but the negative impact of the import elasticity on tari¤s diminishes with country size. The separate impact of total population on tari¤s is positive and signi…cant at the 90% level, while countries with larger populations also attract more FDI in‡ows. But FDI in‡ows respond negatively to higher corporate tax rates. Finally, transitional economy status is associated with lower tari¤s. Column (1) of Table 4 presents the second-stage IV between estimates. Compared to the LS-BE estimates of Column (1) of Table 1, the estimates for tari¤s and FDI in‡ows are larger in magnitude, suggesting that measurement errors in GVC participation and po- litical economy-driven reverse causality are important in biasing the corresponding LS-BE estimates towards zero. Jointly, the null hypothesis that tari¤s and FDI in‡ows are not im- portant in explaining GVC participation is strongly rejected. Likewise, the null hypothesis 18 of endowments not being important in explaining GVC participation is also strongly re- jected. Similar to the results in Column (1) of Table 1, countries with larger endowments of land and/or natural resources exhibit signi…cantly lower backward GVC participation shares, while those with stronger abundance in capital exhibit signi…cantly higher GVC participa- tion shares. Moreover, countries located far from GVC hubs or those with larger domestic industrial capacity exhibit signi…cantly lower backward GVC participation. 4.2.1 Issues with Weak IVs and Statistical Inference A concern with the second-stage IV estimates is that they could be biased and the statistical inferences made above invalid if the IVs are weak. That is, if the IVs included cannot explain the two endogenous variables, tari¤s and FDI in‡ows, then the second-stage results are questionable. Unfortunately, the econometric theory of IV between estimators with two endogenous variables is not yet developed. The existing econometric theory focuses on cross-sectional IV estimation, with i.i.d. error terms as discussed in Andrews et al. (2019). To properly test for weak IVs, we collapse our cross-country decadal panel data s average over set into a single-period cross-country data set, by taking for each variable’ time within each country. In addition to enabling to properly test for weak IVs, this cross- sectional speci…cation allows for heteroskedasticity to be addressed. Columns (3) and (4) of Table 3 present the …rst-stage cross-sectional regressions, with standard errors robust to heteroskedasticity. Comparing Column (3) to Column (1) and Column (4) to Column (2) in Table 3, it is clear that the …rst-stage cross-sectional regressions are very similar to the …rst-stage between regressions. The …rst-stage F-statistics for tari¤s and FDI individually are strongly signi…cant. Jointly, the Kleibergen-Paap F-statistic for the …rst-stage is 10.39, which is larger than the critical value of 8.78, as given in Table 1 of Stock and Yogo (2005), and also larger than the rule of thumb of 10 widely used in empirical research relying on IV estimation. In Table 4, Column (2) presents the second-stage estimates of the cross-sectional regression which are very similar to those of the IV between regression in Column (1). The Anderson-Rubin Wald test for weak-instrument-robust inference for the second-stage is 3.43, which is signi…cant at the 99% level, rejecting the null hypothesis that tari¤s and FDI jointly are not important in determining GVC participation. The high Kleibergen-Paap rk LM 19 statistic of 22.07 also rejects the null hypothesis that the regression is under-identi…ed. For completeness and ease of comparison, Column (3) presents the second-stage cross-sectional regression LS estimates which are smaller in magnitude than the IV estimates. Finally, in Column (4) the …ve IVs are directly included in the second-stage cross-sectional regression and we cannot reject the null hypothesis that jointly these IVs are not signi…cant and do not have a direct impact on GVC participation. 4.3 Relative Importance of the Determinants Overall, the between and the cross-sectional IV estimates paint the same picture that tari¤s and FDI in‡ows are important in determining backward GVC participation of a country. In addition, factor endowments, political stability, geographical proximity to GVC hubs, and domestic industrial capacity are crucial in a¤ecting backward GVC participation. To assess the marginal contributions of the di¤erent determinants, we conduct a thought exercise described below, based on the IV cross-sectional estimation results in Column (2) of Table 4, and we report the results in Table 5. First, we partial out all the included exogenous variables such as distance, political stability, and factor endowments for a regression that includes only the endogenous variables: tari¤s and FDI. The R-squared of 0.056 gives the marginal contribution of tari¤s and FDI in explaining backward GVC participation. Next, we add back the other determinants one at a time and we assess the changes in the R-squared. After adding distance to GVC hubs, the R-squared increases to 0.145, indicating the marginal R- squared contribution of distance is 0.089 (=0.145-0.056). Adding all factor endowments, the new R-squared is 0.238, suggesting that their marginal contribution is 0.182 (=0.238-0.056). Overall, factor endowments can explain 43% of the backward GVC participation shares in the IV cross-sectional regression, which is the most important determinant, followed by distance (21%), political stability (18%), tari¤s and FDI (13%), and domestic industrial capacity (4%), while exchange rate appreciation has no explanatory power. 20 5 Extensions 5.1 Determinants of Forward GVC Participation Table 6 repeats the same exercises to understand the determinants of forward GVC par- ticipation shares. Column (1) presents the baseline LS-BE estimation for forward GVC participation, similar to Column (1) of Table 1. Unreported results show that …ndings are maintained adding alternative or additional controls or changing the sample as in Columns (2)-(6) of Table 1. Columns (2) and (3) of Table 6 reproduce for forward GVC participation shares the IV estimation as in Table 4. The corresponding …rst-stage results are the same as those in Table 3 for which we were able to reject the null hypothesis of weak IVs. The IV estimates in Columns (2) and (3) of Table 6 show that factor endowments play an important role, but in the opposite direction of what was found for backward GVC participation. Countries with stronger endowments of land and/or natural resources show signi…cantly higher forward GVC participation. These endowments lead to high forward participation because commodities are used in a variety of downstream production processes that typically cross several borders. In contrast to backward GVC participation, capital stock has no e¤ect on forward GVC participation, while labor matters. Countries with a stronger abundance in low-skilled labor exhibit lower forward GVC participation, whereas countries with a higher supply of medium- and high-skilled labor exhibit higher forward participation. As expected, shorter distance to GVC hubs, as well as larger domestic industrial capacity are associated with higher forward GVC participation. Forward GVC participation is higher in countries with higher tari¤s while it is not sig- ni…cantly linked to FDI in‡ows. Perhaps surprisingly, political stability has a negative and signi…cant e¤ect on forward GVC participation. This …nding is likely to re‡ect a sample composition e¤ect as countries with the highest forward GVC participation are low-income or fragile and con‡ict countries richly endowed with natural resources but receiving weak FDI in‡ows and having de…cient institutions.36 Exchange rate appreciation is unrelated to forward GVC participation, possibly as higher export prices hurt both domestic value added 36 The …ve countries with the highest average forward GVC participation shares in the 2010s are Libya, Democratic Republic of Congo, Guinea, Algeria, and Iraq. 21 embodied in bilateral partners’exports (numerator) as well as gross exports (denominator). Note that the IV estimates presented in Columns (2) and (3) are larger in magnitude for tar- i¤s, consistent with the previous …ndings, indicating that measurement errors and political economy-driven reverse causality are important. 5.2 GVC vs Traditional Trade In another extension, we examine whether the various determinants a¤ect GVC participation di¤erently from traditional exports. Table 7 shows the IV between estimates of Equation (1) using as the dependent variables backward and forward GVC participation in levels in Columns (1) and (2) and gross exports in Column (3). The IV cross-sectional estimates are presented in Columns (4)-(6) while LS-BE estimates are omitted due to space constraints. The noticeable patterns emerging from Table 7 are as follows. Several determinants are more important in explaining backward GVC participation in levels than traditional exports, as seen from the comparison of coe¢ cients in Column (1) to those in Column (3) or of coe¢ cients in Column (4) to those in Column (6). Speci…cally, tari¤s, distance, and land endowments a¤ect backward GVC participation more negatively than traditional trade, while domestic industrial capacity has a smaller positive impact on GVC trade than on traditional trade. This explains the negative e¤ects of land endowments, domestic industrial capacity, distance, and tari¤s on backward GVC participation shares in Table 1. By contrast, the positive e¤ects of capital endowments, political stability, and FDI in‡ows on backward GVC participation shares in Table 1 imply that these factors have stronger positive e¤ects on GVC trade than on traditional exports, as is con…rmed by comparing the e¤ects in Columns (1) and (3) or in Columns (4) and (6) of Table 7. Some determinants are more important in explaining forward GVC participation in levels than traditional exports, as evidenced by contrasting coe¢ cients in Columns (2) and (3) or in Columns (5) and (6) of Table 7. Domestic industrial capacity has a larger positive e¤ect on forward GVC participation than on traditional trade, while FDI has a smaller positive impact on GVC trade than on traditional trade. This explains the positive e¤ect of domestic industrial capacity and the negative e¤ect of FDI on forward GVC participation shares in Table 6. The signi…cant negative e¤ect of low-skilled labor, political stability, and distance 22 on forward GVC participation shares in Table 6 imply that such factors have more negative or smaller positive e¤ects on forward GVC trade than on traditional exports. 5.3 Cross-Country Cross-Sector Di¤erence-in-Di¤erence Analysis The cross-country regression models described in previous sections use IVs to establish causality between determinants and GVC participation. However, there may be a con- cern that the IV strategy is not su¢ cient given that some other determinants, such as factor endowments, institutions, and connectivity may also be endogenous. As an alternative ap- proach, we use country-sector data with year-to-year variability and apply the country-sector di¤erence-in-di¤erence regression model inspired by the Rajan and Zingales (1998) frame- s factor endowments and the work. This framework relies on interactions between a country’ intensity with which a sector uses a particular factor and has been used to link countries’ sectoral trade performance to various sources of comparative advantage: factor endowments (Romalis, 2004), institutions (Levchenko, 2007; Nunn, 2007; Costinot, 2009), and …nancial development (Beck, 2003; Manova, 2008).37 We use this di¤erence-in-di¤erences regression model to estimate the role of several determinants in explaining GVC participation and exports at the country-sector level: Ycst = 0 + 1 Xct 3 ms + 2 Zcst 3 + Ict + Is + "cst (2) where c; s and t stand for country, sector, and year subscripts, respectively. Y is a measure of GVC participation in levels or a measure of gross exports, X is a vector of country factor endowments, m is a set of sectoral intensities of use of these endowments from a benchmark country (the US), Z is a vector of country-sector trade policies, which include tari¤s imposed at home on imports of …nal goods, and on imports of intermediate inputs, and tari¤s imposed by destination markets and "; is an i.i.d. error.38 The speci…cation controls for country-year …xed e¤ects Ict that account for di¤erences across countries both in time-varying 37 Theoretical underpinnings for this framework in a trade context are provided by Chor (2010) who extends the seminal Eaton and Kortum (2002) trade model and estimates the importance of various sources of comparative advantage for trade patterns in an integrated framework. 38 The sectoral intensities and the tari¤s and FDI measures used are described in the Appendix. 23 macroeconomic conditions, such as country-level FDI in‡ows, political stability, and domestic industrial capacity, as well as in time-invariant factors like geography. The regression also controls for sector …xed e¤ects Is that account for sector-speci…c technology and productivity. The interaction terms allow the e¤ects of factor endowments on GVC participation or exports to di¤er across sectors according to their intensity of use. For example, the interaction term for skilled labor shows that an increase in skilled labor endowments may improve exports or GVC participation of the skilled labor-intensive sectors more than that of other sectors.39 This di¤erence-in-di¤erences model remedies the reverse causality that may exist in cross- country regressions in three ways. First, it is unlikely that strong sectoral GVC participation may cause changes to the country-level determinants. Second, the model assumes that sec- tors di¤er largely for technological reasons in their intensity of use of endowments, which allows results to be given a causal interpretation.40 Third, using a 3-year lag structure for determinants and including the country-year and sector …xed e¤ects mitigates the risk of re- verse causality. Moreover, the use of the lag structure allows GVC participation and exports to adjust slowly to changes in the determinants.41 Overall, we believe that this comparative advantage-type of speci…cation enables us to identify the causal impact of di¤erent determi- nants on GVC participation and exports at the country-sector level and thus complements our previous cross-country analysis. Table 8 provides the estimates for Equation (2). Columns (1) to (3) show that fac- s backward and forward GVC tor endowments and trade policy determine a country-sector’ participation levels as well as traditional exports. Speci…cally, countries relatively more endowed with skilled labor have a comparative advantage in exporting sectors that are skill- intensive, and they exhibit stronger backward and forward GVC participation. Countries with larger abundance in capital exhibit larger exports and stronger GVC participation in capital-intensive sectors. Countries with larger natural resources endowments export more 39 Our cross-country cross-sector speci…cation follows closely the Nunn (2007) and Chor (2010) comparative advantage models that explain trade ‡ ows in levels. We use it for GVC participation in levels but not for GVC participation as a share of gross exports as there is no conceptual basis to consider those measures in this framework. 40 This assumption is inspired by Rajan and Zingales (1998) and implies that there are no restrictions (other than cost) preventing access by certain sectors to skilled labor. 41 As a robustness check we use 5-year and 1-year lag structures and the results (available upon request) are qualitatively similar. 24 and have stronger GVC participation of natural-resource intensive sectors. Importantly, countries with stronger institutional quality have a comparative advantage in exports and in GVC participation of contract-intensive sectors. Importantly, conditioning on countries’ factor endowments, trade policies play an important role for GVC participation and exports. Sectors that face lower tari¤s in their destination markets or lower tari¤s on intermediate inputs exhibit stronger GVC participation and exports. To compare the strength of di¤erent determinants, Figure 2 shows the standardized beta coe¢ cients corresponding to the estimates in Table 8. Hypothetical examples are useful to provide an economic interpretation to the coe¢ cients. If Ghana increased its skilled labor share (7.5 percent in 2012) to the cross-country median (20 percent), its backward GVC participation level and exports would grow by 42 percent, and its forward GVC participation level would grow by 39 percent at the sample mean of sectoral skill intensity. If Mozambique improved its rule of law index to the cross-country median, its forward GVC participation and export levels would grow by 32 percent, and its backward GVC participation would grow by 29 percent at the sample mean of sectoral contractual intensity. Two robustness checks are performed on the cross-country cross-sector analysis. First, we add country-sector level FDI to vector Z. Given that, to the best of our knowledge, country-sector-year level FDI in‡ows data are not available for a wide range of countries, we use FDI announcement data from the fDi Markets Database for the period 2003-2015 as a proxy for FDI in‡ows, following the approach of Hallward-Driemeier and Nayyar (2019). Speci…cally, we cumulate the number of cross-border green…eld FDI projects announced from 2003 onwards for each country-sector. The estimates show that FDI has a positive and signi…cant marginal impact on country-sector GVC participation levels and on gross exports. Second, we show that countries with a higher internet penetration exhibit signi…cantly higher GVC participation and gross exports in sectors that use technology more intensively. These results are presented in Tables A7 and A8 of the Appendix. Overall the results from this di¤erence-in-di¤erences framework based on the cross- country cross-sector regressions are consistent with the previous results based on IV cross- country regressions, highlighting the importance of endowments, institutional quality, tari¤ and FDI, and connectivity in determining GVC participation. 25 6 Concluding Remarks The topic of GVC participation is not new, having been the object of much theoretical and empirical interest in recent years. The objective of this paper is to provide a uni…ed empirical framework to test jointly the role of di¤erent determinants highlighted in the literature as being important for trade in general, or for GVC trade in particular, based on data for more than 100 countries over the last three decades. Results based on our IV cross- s analysis suggest that factor country estimates and di¤erence-in-di¤erence country-sector’ endowments, geographical distance, political stability, trade policy and FDI, and domestic industrial capacity are all very important in explaining GVC participation. Some of these determinants, such as trade policy, FDI, geographical distance, and factor endowments a¤ect GVC trade more than traditional trade. These …ndings are robust to alternative controls and country samples, and to di¤erent measurements of GVC participation. By embedding the determinants in a uni…ed framework, we are able to estimate their marginal contributions and assess their relative importance for GVC participation. Given that GVCs are often characterized as trade with intense …rm-to-…rm interactions via contracting and specialized products and investment, it is perhaps not surprising that factors such as tari¤s and FDI, which a¤ect traditional trade would have an ampli…ed e¤ect on GVC trade. However other factors, such as domestic industrial capacity, are found to play a di¤erent role for GVC trade, due to a unique ability to cultivate domestic suppliers and allow countries, such as China, to climb up value chains (by increasing domestic value added). We hope that the patterns we identi…ed will be useful in guiding future theoretical and empirical studies of GVC formation and growth. References Acemoglu, D., Antràs, P., and Helpman, E. 2007. “Contracts and Technology Adoption,” Ameri- can Economic Review 97(3): 916-943. Amiti, M., Itskhoki, O., and Konings, J. 2014. “Importers, Exporters, and Exchange Rate Dis- connect,” American Economic Review 104(7): 1942-1978. Andrews, I., Stock, J., and Sun, L. 2019. “Weak Instruments in Instrumental Variables Regression: Theory and Practice,” Annual Review of Economics 11(1): 727-753. 26 Ansón, J., Arvis, J., Bo¤a, M., Helble, M., and Shepherd, B. 2017. “Time, Uncertainty, and Trade Flows,” Working Paper 673, Asian Development Bank, Manila, Philippines. Antràs, P. 2016. Global Production: Firms, Contracts and Trade Structure, Princeton University Press. Antràs, P., and De Gortari, A. 2020. “On the Geography of Global Value Chains,” Econometrica. Antràs, P., and Helpman, E. 2004. “Global Sourcing,” Journal of Political Economy 112(3): 552- 580. Antràs, P., and Helpman, E. 2008. “Contractual Frictions and Global Sourcing,”in Helpman, E., D. Marin, and T. Verdier (eds.) The Organization of Firms in a Global Economy. Cambridge, MA: Harvard University Press. Arkolaskis, C., Costinot, A., and Rodriguez-Clare, A. 2012. “New Trade Models, Same Old Gains?,” American Economic Review 102(1): 94-130. Aslam, A.,Novta, N., and Rodrigues-Bastos, F. 2017. “Calculating Trade in Value Added,” IMF Working Paper WP/17/178. Baldwin, R. 2012. “Trade and Industrialization after Globalization’s Second Unbundling: How Building and Joining a Supply Chain are Di¤erent and Why it Matters,” in Globalization in an Age of Crisis: Multilateral Economic Cooperation in the Twenty-First Century. University of Chicago Press. Baldwin, R. 2016. Globalization’s Three Unbundlings. Harvard University Press. Baldwin, R., and Taglioni, D. 2012. “Gravity Chains: Estimating Bilateral Trade Flows when Parts and Components Trade is Important,” NBER Working Paper No. 16672. Bagwell, K. and Staiger, R. 2011. “What Do Trade Negotiators Negotiate About? Empirical Evidence from the World Trade Organization,” American Economic Review 101(4): 1238-1273. Balié, J., Del Prete, D., Magrini, E., Montalbano, P., Nenci, S. 2019. “Does Trade Policy Impact Food and Agriculture Global Value Chain Participation of Sub-Saharan African Countries?,” American Journal of Agricultural Economics 101(3): 773-789. Bas, M., and Strauss-Kahn, V. 2015. “Input-Trade Liberalization, Export Prices, and Quality Upgrading,” Journal of International Economics 95(2): 250-262. Beck, T. 2003. “Financial Dependence and International Trade,” Review of International Eco- nomics 11: 296-316. Bickerdike, C. 1907. “Review of A.C. Pigou’s Protective and Preferential Import Duties,” Eco- nomic Journal 17 (65): 98-108. Blyde, J. (Ed.) 2014. Synchronized Factories: Latin America and the Caribbean in the Era of Global Value Chains. Inter-American Development Bank, Washington DC. 27 Borin, A., and Mancini, M. 2015. “Follow the Value Added: Bilateral Gross Export Accounting,” Economic Working Papers, Bank of Italy. Borin, A., and Mancini, M. 2019. “Measuring What Matters in Global Value Chains and Value- Added Trade,” Policy Research Paper 8804, World Bank, Washington, DC. Broda, C., Limao, N., and Weinstein, D. 2008. “Optimal Tari¤s and Market Power: The Evi- dence,” American Economic Review 98(5): 2032-2065. Brooks, D., and Ferrarini, B. 2012. “Vertical Gravity,” ADB Working Paper 303, Manila, Philip- pines. Buelens, C., and Tirpak, M. 2017. “Reading the Footprints: How Foreign Investors Shape Coun- tries’Participation in Global Value Chains,” Comparative Economic Studies 59(4): 561-584. Caliendo, L. and Parro, F. 2015. “Estimates of the Trade and Welfare E¤ects of NAFTA,” The Review of Economic Studies, 82(1): 1-44. Cheng, K., Rehman, S., Seneviratne, D., and Zhang, S. 2015. “Reaping the Bene…ts from Global Value Chains,”IMF Working Paper WP/15/204. Washington, DC: International Monetary Fund. Chor, D. 2010. “Unpacking Sources of Comparative Advantage: A Quantitative Approach,” Jour- nal of International Economics 82(2): 152-167. Christ, N., and Ferrantino, M. 2011. “Land Transport for Exports: The E¤ects of Cost, Time, and Uncertainty in Sub-Saharan Africa,” World Development 39(10): 1749–1759. Costinot, A. 2009. “On the Origins of Comparative Advantage,” Journal of International Eco- nomics 77: 255-264. Desai, M., Foley, C., and Hines Jr, J. 2004. “Foreign Direct Investment in a World of Multiple Taxes,” Journal of Public Economics 88(12): 2727-2744. Djankov, S., Freund, C., and Pham, C. 2010. “Trading on Time,” Review of Economics and Statistics 92(1): 166-173. Eaton, J., and Kortum, S. 2002. “Technology, Geography, and Trade,” Econometrica 70(5): 1741- 1779. Ekholm, K., Forslid, R., and Markusen, J. 2007. “Export-Platform Foreign Direct Investment,” Journal of the European Economic Association 5(4): 776-795. Feenstra, R. 1998. “Integration of Trade and Disintegration of Production in the Global Economy,” Journal of Economic Perspectives 12(4): 31-50. Feenstra, R., and Hanson, G. 1997. “Foreign Direct Investment and Relative Wages: Evidence From Mexico’ s Maquiladoras,” Journal of International Economics 42(3/4): 371-393. Fernandes, A., Forero, A., Maemir, H., and Mattoo, A. 2019a. “Are Trade Preferences a Panacea? The African Growth and Opportunity Act and African Exports,”Policy Research Working Paper 8753, World Bank, Washington, DC. 28 Fernandes, A., Mattoo, A., Nguyen, H., and Schi¤bauer, M. 2019b. “The Internet and Chinese Exports in the Pre-Ali Baba Era,” Journal of Development Economics 138: 57-76. Hallward-Driemeier, M. and Nayyar, G. 2019. “Have Robots Grounded the Flying Geese? Evi- dence from Green…eld FDI in Manufacturing,” Policy Research Paper 9097, World Bank, Wash- ington, DC. Head, K. and Mayer, T. 2004. “Market Potential and the Location of Japanese Investment in the European Union,” Review of Economics and Statistics 86(4): 959-972. Hummels, D., Ishii, J., and Yi, K-M. 2001. “The Nature and Growth of Vertical Specialization in World Trade,” Journal of International Economics 54: 75-96. Hummels, D., and Schaur, G. 2013. “Time as a Trade Barrier,” American Economic Review 103(7): 2935–59. International Monetary Fund. 2000. Transition Economies: An IMF Perspective on Progress and Prospects. Ignatenko, A., Raei, F., and Mircheva, B. 2019. “Global Value Chains: What are the Bene…ts and Why Do Countries Participate?,” IMF Working Paper No. 19/18. Johnson, R., and Noguera, G. 2017. “A Portrait of Trade in Value-Added over Four Decades,” Review of Economics and Statistics 99(5): 896-911. Kee, H. L., Nicita, A. and Olarreaga, M. 2008. “Import Demand Elasticities and Trade Distor- tions,” Review of Economics and Statistics 90(4): 666-682. Kee, H. L. 2015. “Local Intermediate Inputs, and the Shared Supplier Spillovers of Foreign Direct Investment,” Journal of Development Economics 112: 56-71. Kee, H. L., and Tang, H. 2016. “Domestic Value Added in Exports: Theory and Firm Evidence from China,” American Economic Review 106(6): 1402-1436. Kowalski, P., Gonzalez, J., Ragoussis, A., and Ugarte, C. 2015. “Participation of Developing Countries in Global Value Chains: Implications for Trade and Trade-Related Policies,” Trade Policy Papers 179, Organisation for Economic Co-operation and Development, Paris. Koopman, R., Wang, Z., and Wei, S. 2014. “Tracing Value-Added and Double Counting in Gross Exports,” American Economic Review 104(2): 459-94. Laget, E., Osnago, A., Rocha, N., and Ruta, M. 2018. “Deep Trade Agreements and Global Value Chains,” Policy Research Working Paper 8491, World Bank, Washington, DC. Lenzen, M., Moran, D., Kanemoto, K., and Geschke, A. 2013. “Building Eora: A Global Multi- regional Input-Output Database at High Country and Sector Resolution,” Economic Systems Research 25(1): 20-49. Levchenko, A. 2007. “Institutional Quality and International Trade,” Review of Economic Studies 74(3): 791-819. 29 Manova, K. 2008. “Credit Constraints, Equity Market Liberalizations and International Trade,” Journal of International Economics 76(1): 33-47. Manova, K. 2015. “Global Value Chains and Multinational Activity with Financial Frictions,” In J. Amador and F. di Mauro ed., The Age of Global Value Chains: Maps and Policy Issues. CEPR E-book, http://www.voxeu.org/content/age-global-value-chains-maps-and-policy-issues. Miller, R., and Temurshoev, U. 2017. “Output Upstreamness and Input Downstreamnes of Indus- tries/Countries in World Production,” International Regional Science Review 40(5): 443-475. Mutti, J. and Grubert, H. 2004. “Empirical Asymmetries in Foreign Direct Investment and Tax- ation,” Journal of International Economics 62(2): 337-358. Noguera, G. 2012. “Trade Costs and Gravity for Gross Trade and Value Added Trade,” UC Berkeley and Columbia University, Job Market Paper. Nunn, N. 2007. “Relationship-Speci…city, Incomplete Contracts, and the Pattern of Trade,” Quar- terly Journal of Economics 122(2): 569-600. OECD and WTO. 2013. Aid for Trade at a Glance 2013: Linking to Value Chains. Paris: Organ- isation for Economic Co-operation and Development. Ore…ce, G., and Rocha, N. 2014. “Deep Integration and Production Networks: An Empirical Analysis,” World Economy 37(1): 106-136. Pierola, M., Fernandes, A., and Farole, T. 2018. “The Role of Imports for Exporter Performance in Peru,” World Economy 41(2): 550-572. Rajan, R., and Zingales, L. 1998. “Financial Dependence and Growth,” American Economic Review 88(3): 559-586. Reed, W. 2015. “On the Practice of lagging variables to Avoid Simultaneity,” Oxford Bulletin of Economics and Statistics 77(6): 897-905. Rodríguez-Pose, A., Tselios, V., Winkler, D., and Farole, T. 2013. “Geography and the Drivers of Firm Exports in Indonesia,” World Development 44(C): 225-240. Romalis, J. 2004. “Factor Proportions and the Structure of Commodity Trade,” American Eco- nomic Review 94(1): 67-97. Stock, J. and Yogo, M. 2005. “Testing for Weak Instruments in Linear IV Regression.” Ch. 5 in J.Stock and D. Andrews (eds), Identi…cation and Inference for Econometric Models: Essays in Honor of Thomas J. Rothenberg, Cambridge University Press. Timmer, M., Erumban, A., Los, B., Stehrer, R., and de Vries, G. 2014 “Slicing Up Global Value Chains,” Journal of Economic Perspectives 28(2): 99-118. Timmer, M., Dietzenbacher, E., Los, B., Stehrer, R., and de Vries, G. 2015. “An Illustrated User Guide to the World Input–Output Database: the Case of Global Automotive Production,” Review of International Economics 23: 575-605. 30 Timmer, M., Los, B., Stehrer, R., and de Vries, G. 2016. “An Anatomy of the Global Trade Slow- down based on the WIOD 2016 Release,” GGDC research memorandum number 162, University of Groningen. United Nations. 2015. “Global Values and Interconnectedness of Asia-Paci…c Economies,”in Asia- Paci…c Trade and Investment Report 2015, Supporting Participation in Value Chains, 103-138. United Nations. 2019. Handbook on Accounting for Global Value Chains: GVC Satellite Accounts and Integrated Business Statistics, UN Statistics Division. Vijil, M., Wagner, L., and Woldemichael, M. 2019. “Import Uncertainty and Export Dynamics,” Policy Research Working Paper 8793, World Bank, Washington, DC. Wang, Z., Wei, S., Yu, X., and Zhu, K. 2016. “Characterizing Global Value Chains,” mimeo (available at http://www.cepr.org/sites/default/…les/Wang,%20Zhi.pdf). Wang, Z, Wei, S., and Zhu, K. 2013. “Quantifying International Production Sharing at the Bilat- eral and Sector Levels,” NBER Working Paper, No. 19677. Wooldridge, J. 2012. Introductory Econometrics: A Modern Approach, 5e, South-Western, Cen- gage Learning. World Bank. 2019. Trading for Development in the Age of Global Value Chains, World Develop- ment Report 2020. Washington, DC: The World Bank. Yi, K-M. 2003. “Can Vertical Specialization Explain the Growth of World Trade?” Journal of Political Economy 111(1): 52-102. Yi, K.M. 2010. “Can Multistage Production Explain the Home Bias in Trade?” American Eco- nomic Review 100(1): 364-393. Figure 1: GVC participation across world regions in 2010s Notes: Averages shown cover the period 2010-2015. Backward GVC participation measures the import content of exports relative to total exports. 31 Figure 2: Standardized coe¢ cients for determinants of country-sector GVC participation and exports Notes: Standardized beta coe¢ cients corresponding to the estimates in columns (1)-(3) of Table 8 are shown. ***, **, and * indicate signi…cance levels of 1%, 5%, and 10%, respectively. 32 Table 1: Dependent variable: Backward GVC participation shares (EORA) (1) (2) (3) (4) (5) (6) Regressions LS - BE LS - BE LS - BE LS - BE LS - BE LS - BE Avg. tari¤ rate (%) -0.006*** -0.006*** -0.004* -0.006*** -0.005** -0.006*** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) FDI in‡ows (log) 0.023** 0.023** 0.019* -0.009 0.017* 0.022** (0.010) (0.010) (0.011) (0.013) (0.010) (0.010) Distance to GVC hubs -0.104*** -0.103*** -0.111** -0.119*** -0.116*** -0.103*** (log) (0.036) (0.035) (0.048) (0.044) (0.038) (0.036) Political stability index 0.030* 0.029* 0.024 0.040** 0.034* 0.025 (0.016) (0.016) (0.019) (0.019) (0.017) (0.017) Domestic industrial capacity -0.027*** -0.028*** -0.027*** 0.000 -0.015* -0.027*** (log) (0.008) (0.008) (0.009) (0.011) (0.009) (0.008) Resources rents / GDP -0.003** -0.003*** -0.002** -0.002 -0.003*** -0.003** (%) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Capital / GDP (log) 0.044* 0.046* 0.025 0.033 0.034 0.050* (0.025) (0.025) (0.029) (0.026) (0.027) (0.025) Land / GDP (log) -0.020*** -0.019*** -0.014** -0.014 -0.019*** -0.020*** (0.006) (0.006) (0.007) (0.009) (0.007) (0.006) Med/High-skilled labor 0.012 0.012 0.005 -0.005 0.015 0.006 / GDP (log) (0.015) (0.015) (0.016) (0.016) (0.017) (0.016) Low-skilled labor 0.009 0.010 0.011 0.018 0.007 0.011 / GDP (log) (0.015) (0.015) (0.016) (0.016) (0.016) (0.015) Exch. rate 0.000 0.000 -0.000 0.214 0.000 appreciation (0.000) (0.000) (0.000) (0.403) (0.000) Misalignment 0.000 (0.000) NAFTA 0.046 (0.068) EU 0.079* (0.044) MERCOSUR 0.020 (0.071) ASEAN 0.133** (0.056) No. of trade partners -0.011 (log) (0.038) Depth of Agreement 0.009 (log) (0.023) Time to import 0.000 (log) (0.001) Female participation 0.001 (log) (0.001) Observations 290 290 266 194 191 290 R-squared 0.528 0.530 0.546 0.464 0.505 0.532 Number of countries 121 121 119 88 109 121 33 Decade …xed e¤ects YES YES YES YES YES YES Notes: LS standard errors in parentheses. Column (4) estimates are based on a sample excluding high-income countries; ***, **, and * indicate signi…cance levels of 1%, 5%, and 10%, respectively. Table 2: Dependent variable: Sectoral backward GVC participation shares (EORA) (1) (2) (3) (4) Regressions LS - BE LS - BE LS - BE LS - BE Sector Agri. Comm. Manu. Serv. Avg. tari¤ rate (%) -0.000 -0.000 -0.004*** -0.001 (0.000) (0.000) (0.002) (0.000) FDI in‡ ows (log) 0.001 -0.001 0.023*** 0.001 (0.001) (0.001) (0.009) (0.003) Distance to GVC hubs (log) -0.002 0.004 -0.098*** 0.005 (0.004) (0.004) (0.031) (0.010) Political stability index 0.001 -0.001 0.016 0.008* (0.002) (0.002) (0.014) (0.004) Domestic industrial capacity -0.001 0.001 -0.020*** -0.007*** (log) (0.001) (0.001) (0.007) (0.002) Rents from resources / GDP 0.000 0.001*** -0.004*** -0.000 34 (%) (0.000) (0.000) (0.001) (0.000) Capital / GDP (log) -0.002 0.001 0.056*** -0.003 (0.003) (0.003) (0.021) (0.007) Land / GDP (log) 0.000 0.001 -0.016*** -0.003* (0.001) (0.001) (0.005) (0.002) Med/High-skilled labor / GDP (log) 0.002 -0.003** 0.004 0.001 (0.002) (0.002) (0.013) (0.004) Low-skilled labor / GDP (log) 0.004** -0.001 0.004 0.006 (0.002) (0.002) (0.013) (0.004) Exch. rate appreciation 0.000 0.000 0.000 -0.000 (0.000) (0.000) (0.000) (0.000) Observations 290 290 290 290 R-squared 0.300 0.656 0.605 0.362 Number of countries 121 121 121 121 Decade …xed e¤ects YES YES YES YES Notes: LS standard errors in parentheses. ***, **, and * indicate signi…cance levels of 1%, 5%, and 10%, respectively. Table 3: First-stage regressions (1) (2) (3) (4) Regressions LS - BE LS - BE LS LS VARIABLES Avg. tari¤ rate (%) FDI in‡ ows (log) Avg. tari¤ rate (%) FDI in‡ows (log) Import elasticity -37.038*** 0.243 -38.720*** 0.782 (10.604) (1.864) (10.259) (1.880) Import elasticity*Population (log) 2.307*** -0.053 2.351*** -0.068 (0.606) (0.107) (0.594) (0.111) Population (log) -1.154 1.174*** -0.991 1.113*** (1.387) (0.244) (1.508) (0.332) Statutory Corporate Tax Rate -0.075 -0.025** -0.022 -0.041*** (0.071) (0.013) (0.111) (0.013) Transitional Economy Status -4.318*** 0.343 -4.641*** 0.408* (1.500) (0.264) (1.391) (0.239) Distance to GVC hubs (log) -0.826 0.385 -1.733 0.641* (2.144) (0.377) (1.925) (0.336) Political stability index -0.890 0.439*** -0.854 0.421*** (0.818) (0.144) (0.785) (0.157) Domestic industrial capacity -2.221** -0.127 -2.364* -0.072 (log) (0.937) (0.165) (1.289) (0.217) Rents from resources / GDP -0.016 -0.000 -0.042 0.007 35 (%) (0.053) (0.009) (0.041) (0.011) Capital / GDP (log) 0.997 0.091 0.784 0.170 (1.169) (0.205) (1.256) (0.200) Land / GDP (log) -0.029 -0.050 -0.003 -0.055 (0.298) (0.052) (0.355) (0.074) Med/High-skilled labor / GDP (log) -0.778 -1.093*** -1.233 -0.932*** (1.154) (0.203) (1.436) (0.257) Low-skilled labor / GDP (log) 0.112 -0.019 0.227 -0.065 (0.718) (0.126) (0.558) (0.105) Exch. rate appreciation 0.001 -0.001 0.001 -0.000 (0.004) (0.001) (0.002) (0.000) Observations 290 290 121 121 R-squared 0.480 0.863 0.465 0.851 Number of countries 121 121 121 121 Sanderson-Windmeijer F test NA NA 12.69*** 7.02*** Kleibergen-Paap Wald test for Weak IV NA 10.39** Decade …xed e¤ects YES YES NO NO Notes: LS standard errors in parentheses for Columns (1) and (2); Robust standard errors in parentheses for Columns (3) and (4); ***, **, and * indicate signi…cance levels of 1%, 5%, and 10%, respectively. Table 4: Dependent variable: Backward GVC participation shares (EORA) (1) (2) (3) (4) Regressions IV - BE IV LS LS Avg. tari¤ rate (%) -0.011*** -0.009*** -0.005*** -0.004** (0.004) (0.003) (0.001) (0.002) FDI in‡ ows (log) 0.038* 0.030* 0.018* 0.014 (0.020) (0.018) (0.010) (0.010) Distance to GVC hubs (log) -0.090** -0.112*** -0.118*** -0.080* (0.038) (0.041) (0.044) (0.046) Political stability index 0.017 0.031** 0.040*** 0.034** (0.019) (0.014) (0.014) (0.015) Domestic industrial capacity -0.040*** -0.031** -0.022** -0.023 (log) (0.015) (0.012) (0.008) (0.027) Rents from resources / GDP -0.003** -0.003*** -0.003*** -0.003** (%) (0.001) (0.001) (0.001) (0.001) Capital / GDP (log) 0.045* 0.033 0.034 0.030 (0.026) (0.024) (0.023) (0.023) Land / GDP (log) -0.021*** -0.019*** -0.019** -0.021*** (0.006) (0.007) (0.007) (0.007) Med/High-skilled labor / GDP (log) 0.018 0.009 0.005 -0.010 (0.016) (0.014) (0.014) (0.030) Low-skilled labor / GDP (log) 0.011 0.017 0.015 0.015 (0.016) (0.018) (0.019) (0.019) Exch. rate appreciation -0.000 -0.000 0.000 -0.000* (0.000) (0.000) (0.000) (0.000) Import elasticity 0.084 (0.250) Import elasticity X Population (log) -0.009 (0.014) Population (log) 0.024 (0.037) Statutory Corporate Tax Rate 0.000 (0.001) Transitional Economy Status 0.039 (0.034) Observations 290 121 121 121 R-squared 0.483 0.472 0.503 0.530 Number of countries 121 121 121 121 F stat for excluded IV 1.41 AR Wald test for weak IV robust inference 3.43*** KP rk LM statistic for underidenti…cation 22.07*** Decade …xed e¤ects YES NO NO NO Notes: LS standard errors in parentheses for Column (1); Robust standard errors for Columns (2)-(4); ***, **, and * indicate signi…cance levels of 1%, 5%, and 10%, respectively. 36 Table 5: Marginal Contributions (1) (2) (3) (4) Included Variables R-squared Marginal R-Squared Share of Contribution (%) Tari¤ + FDI 0.056 0.056 13 Tari¤ + FDI + Distance to Hubs 0.145 0.089 21 Tari¤ + FDI + Political Stability 0.132 0.076 18 37 Tari¤ + FDI + Dom. Ind. Cap. 0.073 0.017 4 Tari¤ + FDI + Endowments 0.238 0.182 43 Sum 0.42 100 Notes: R-squared in Column (2) is constructed based on the coe¢ cients of Column (2) of Table 4, by including only the variables listed in Column (1). The marginal R-squared in Column (3) measures the change in R-squared in Column (2) when compared to the R-squared of including only tari¤s and FDI equal to 0.056. Column (4) calculates the shareof contribution of the additional variables in the sum of all marginal R-squared. Table 6: Dependent variable: Forward GVC participation shares (EORA) (1) (2) (3) (4) (5) Regressions LS - BE IV - BE IV LS LS Avg. tari¤ rate (%) 0.001 0.004** 0.004* 0.000 -0.001 (0.001) (0.002) (0.002) (0.001) (0.001) FDI in‡ ows (log) -0.008 -0.006 -0.002 -0.006 -0.008 (0.006) (0.012) (0.012) (0.005) (0.005) Distance to GVC hubs (log) -0.042* -0.052** -0.043** -0.036* -0.044* (0.021) (0.023) (0.019) (0.019) (0.023) Political stability index -0.018* -0.011 -0.017* -0.022* -0.015 (0.010) (0.011) (0.010) (0.011) (0.011) Domestic industrial capacity 0.010* 0.011 0.006 0.007** -0.007 (log) (0.005) (0.009) (0.008) (0.003) (0.012) Rents from resources / GDP 0.002*** 0.002*** 0.002*** 0.002** 0.002** (%) (0.001) (0.001) (0.001) (0.001) (0.001) Capital / GDP (log) -0.012 -0.014 -0.010 -0.008 -0.004 (0.015) (0.015) (0.017) (0.017) (0.017) Land / GDP (log) 0.009*** 0.011*** 0.010*** 0.009** 0.009** (0.004) (0.004) (0.004) (0.004) (0.004) Med/High-skilled labor / GDP (log) 0.017* 0.016 0.019* 0.019* 0.011 (0.009) (0.010) (0.011) (0.011) (0.016) Low-skilled labor / GDP (log) -0.030*** -0.032*** -0.034*** -0.032*** -0.035*** (0.009) (0.009) (0.010) (0.010) (0.010) Exch. rate appreciation -0.000 0.000 0.000 -0.000 0.000* (0.000) (0.000) (0.000) (0.000) (0.000) Import elasticity -0.117 (0.113) Import elasticity X Population (log) 0.009 (0.007) Population (log) 0.003 (0.018) Statutory Corporate Tax Rate -0.000 (0.001) Transitional Economy Status -0.009 (0.019) Observations 290 290 121 121 121 R-squared 0.409 0.341 0.344 0.394 0.436 Number of countries 121 121 121 121 121 Decade …xed e¤ects YES YES NO NO NO Note: LS standard errors in parentheses for Columns (1) and (2); Robust standard errors for Columns (3)-(5); ***, **, and * indicate signi…cance levels of 1%, 5%, and 10%, respectively. 38 Table 7: GVC vs Traditional Trade (1) (2) (3) (4) (5) (6) Regressions IV - BE IV - BE IV - BE IV IV IV Dependent variables BGVC (log) FGVC (log) Exports (log) BGVC (log) FGVC (log) Exports (log) Avg. tari¤ rate (%) -0.069** -0.002 -0.044** -0.058** -0.004 -0.042** (0.028) (0.026) (0.020) (0.023) (0.022) (0.018) FDI in‡ ows (log) 0.492*** 0.292** 0.428*** 0.424*** 0.295** 0.421*** (0.151) (0.141) (0.111) (0.138) (0.130) (0.123) Distance to GVC hubs (log) -0.475* -0.303 0.073 -0.566** -0.268 0.046 (0.288) (0.269) (0.211) (0.221) (0.216) (0.154) Political stability index 0.088 -0.037 0.037 0.148 -0.064 0.047 (0.140) (0.130) (0.103) (0.123) (0.130) (0.094) Domestic industrial capacity 0.455*** 0.696*** 0.567*** 0.516*** 0.688*** 0.576*** (log) (0.111) (0.104) (0.082) (0.096) (0.088) (0.088) Rents from resources / GDP -0.004 0.022*** 0.032*** -0.006 0.022*** 0.031*** (%) (0.008) (0.007) (0.006) (0.007) (0.008) (0.009) Capital / GDP (log) 0.286 0.082 0.178 0.268* 0.122 0.180 39 (0.193) (0.181) (0.142) (0.158) (0.170) (0.127) Land / GDP (log) -0.128*** -0.013 -0.067* -0.119** -0.011 -0.063* (0.048) (0.045) (0.035) (0.047) (0.040) (0.037) Med/High-skilled labor / GDP (log) -0.076 -0.022 -0.070 -0.098 0.008 -0.065 (0.123) (0.115) (0.090) (0.115) (0.133) (0.080) Low-skilled labor / GDP (log) 0.038 -0.189* 0.018 0.029 -0.222* 0.005 (0.117) (0.110) (0.086) (0.101) (0.131) (0.071) Exch. rate appreciation 0.000 0.000 0.000 0.000* 0.000* 0.000** (0.001) (0.001) (0.000) (0.000) (0.000) (0.000) Observations 290 290 290 121 121 121 R-squared 0.920 0.925 0.950 0.924 0.923 0.949 Number of country 121 121 121 121 121 121 Decade …xed e¤ects YES YES YES NO NO NO Notes: LS standard errors in parentheses for Columns (1)-(3); Robust standard errors for Columns (4)-(6); BGVC is backward GVC participation in levels (logs) and FGVC is forward GVC participation in levels (logs); ***, **, and * indicate signi…cance levels of 1%, 5%, and 10%, respectively. Table 8: Country-Sector Analysis Dependent variables: Backward GVC participation Forward GVC participation Exports (log) (log) (log) (1) (2) (3) 3-year lag skilled labor endowment X skill intensity 0.087*** 0.082*** 0.087*** (0.005) (0.005) (0.005) 3-year lag capital endowment X capital intensity 0.002*** 0.002*** 0.002*** (0.000) (0.000) (0.000) 3-year lag nat. resource endoment X nat. resource intensity 0.103*** 0.107*** 0.111*** (0.008) (0.009) (0.009) 3-year lag rule of law index X contract intensity 1.087*** 1.202*** 1.182*** 40 (0.051) (0.050) (0.048) 3-year lag average output tari¤ -0.430 1.309*** 0.505* (0.301) (0.313) (0.287) 3-year lag average input tari¤ -1.977** -4.938*** -3.283*** (0.787) (0.881) (0.790) 3-year lag average market access tari¤ -6.185*** -6.866*** -5.959*** (0.354) (0.377) (0.360) Country*Year FE Yes Yes Yes Sector FE Yes Yes Yes Observations 14,387 14,387 14,387 R-squared 0.919 0.907 0.913 Notes: Robust standard errors in parentheses. ***, **, and * indicate signi…cance levels of 1%, 5%, and 10%, respectively. 1 On-line Appendix 1.1 Data for Cross-Country Analysis Table A1 presents the detailed de…nitions and data sources for the variables used in the analysis. Summary statistics for the country level variables are provided in Table A2 while their correlations are shown in Tables A3 and A4. The sectors in the EORA database are de…ned as follows. Agriculture includes …shing; mining and quarrying is a separate sector; manufacturing includes food and beverages; tex- tiles and wearing apparel; wood and paper; petroleum, chemicals and non-metallic minerals; metal products; electrical and machinery; transport equipment; and other manufacturing. Services covers recycling; electricity, gas and water; construction; maintenance and repair; wholesale trade; retail trade; hotels and restaurants; transport; post and telecommunica- tions; …nancial intermediation and business activities; public administration; and education, health and other services. We exclude from our analysis the following three sectors: private households; others; and re-exports and re-imports. Regarding factor endowments, all variables are measured as a ratio to real GDP. Low- skilled labor endowments follow the ILO’ s de…nition of low skill (skill level 1), based on the International Standard Classi…cation of Occupations. Likewise, medium/high-skilled labor refers to the ILO’ s de…nition of medium and high skill (skill levels 2-4). We use the real capital stock from the Penn World Tables to measure capital endowments. Regarding natural resource endowments, we use the country’ s land area as well as the rents from natural resources, both taken from the World Development Indicators (WDI). We use the average manufacturing tari¤ rate from WDI and the level of FDI in‡ ows from UNCTAD. Geography is captured by a country’ s total distance (in kilometers) to the main global GVC hubs, China, Germany, and the US from CEPII. For domestic industrial capacity we use manufacturing value added from WDI. To measure a country’ s institutional quality, we use the political stability index from the World Governance Indicators database. We also include a measure of exchange rate appreciation based on changes in the average nominal exchange rate from WDI. To assess the role of PTAs, we include indicator variables for a country’ s participation in PTAs: the North American Free Trade Agreement (NAFTA), the EU, ASEAN, and the Southern Common Market (MERCOSUR), the number of PTA partners and a measure of the depth of those PTAs, both taken from the Content of Deep Trade Agreements database of Hofmann et al. (2017). To capture aspects related to connectivity, we add alternatively, the time to clear imports from the Doing Business Database, the Logistics Performance Index’overall score, the share of the population that uses the internet from WDI and the share of people that speak English as a second language from Melitz and Toubal (2014). We also include a measure of domestic credit as percent of GDP from WDI to capture a country’ s …nancial development and the share of females in the total labor force from WDI. As an alternative to exchange rate depreciation we use a measure of exchange rate misalignment relative to an equilibrium exchange rate following Couharde et al. (2017). 1 1.2 Data for Cross-Country Cross-Sector Analysis For country-sector speci…cations, we estimate the impact of country-sector determinants in year t-3 on country-sector GVC participation in year t, e¤ectively covering the period 1999- 2015. Summary statistics for the EORA GVC participation shares and all the determinants at the country-sector level are provided in Table A5 and correlations among all determinants at country-sector level are provided in Table A6. We interact each country-year determinant with its corresponding sectoral intensity of use across manufacturing sub-sectors to obtain measures that vary at the country-sector- year level. For labor endowments we use the country-level share of high-skilled workers following the ILO’ s de…nition of low skill (skill levels 3 and 4) based on the International Standard Classi…cation of Occupations. We interact the country skilled labor endowment with a benchmark measure of skill intensity in the US de…ned as the ratio of skilled workers to unskilled workers by sector taken from the NBER-CES Manufacturing Industry Database. For capital endowments, we use the ratio of real capital stock to total employment at the country level from Penn World Tables which we interact with a benchmark measure of capital intensity in the US by sector taken from the NBER-CES Manufacturing Industry Database. To capture natural resources endowments, we interact the natural resources to GDP ratio from WDI with an indicator for natural resource intensive sectors based on Braun (2003). For country institutional quality we use the rule of law index from the World Governance Indicators database which is interacted with the measure of contract intensity proposed by Nunn (2007). The measure of contract intensity is determined by the share of specialized and customized intermediate inputs used in the production of the …nal good for each sector based on a U.S. input-output table.42 Additionally, our model controls for three types of tari¤ measures: tari¤s imposed at home on imports of the sector, tari¤s imposed at home on imports of intermediate inputs used by the sector, and tari¤s imposed in destination markets on exports of the sector. The tari¤s draw on a newly constructed database by Felbermayr et al. (2019), which is based on the UN-TRAINS and IDB databases, as well as on WITS import and export data. To compute the three types of tari¤ measures, we draw on tari¤ and trade data at the reporting country-HS6-digit-partner country-year level from WITS. First, to construct tari¤s imposed at home on imports of the sector we …rst aggregate tari¤s (taking the simple average) and imports (taking the sum) to the EORA sector level. We then sum imports across countries of origin to obtain import weights which are used to compute tari¤s imposed at home on imports of a sector as a weighted-average tari¤ at the reporting country-EORA sector-year level. Second, we construct tari¤s imposed at home on imports of intermediate inputs used by the sector as follows. We keep tari¤ data and import data only for HS 6-digit products classi…ed as intermediates according to the UN Broad Economic Categories (BEC) classi- 42 Nunn uses data on the fraction of each input that is not sold in an organized exchange nor reference- priced according to Rauch (1999) to construct the share of intermediate inputs that require specialized and customized business relationships. Nunn’ s contract intensity measure (based on Rauch’s liberal classi…cation) is available at the 3-digit ISIC revision 2 level which is concorded to the eight manufacturing sectors in the EORA database based on industries’verbal descriptions. 2 …cation.43 We then aggregate tari¤s (taking the simple average) and imports (taking the sum) to the EORA sector level. We then sum imports across countries of origin to obtain import weights which are used to compute an auxiliary measure of tari¤s on imports of BEC intermediates at the reporting country-EORA sector-year level. Then, based on the US Input-Output table for 2005 from WIOD which was aggregated to the eight EORA man- ufacturing sectors, we construct for each sector the share of its total intermediate inputs that originates in each of the eight EORA sectors. Finally, we combine these I-O table weights with the auxiliary tari¤s on imports of BEC intermediates to construct a weighted-average tari¤ on intermediate inputs at the reporting country-EORA sector-year level. Third, to construct tari¤s imposed in destination markets on exports of the sector, we aggregate tari¤s (taking the simple average) and exports (taking the sum) to the EORA sector level. Then, we sum exports across destination markets to compute export weights at the reporting country-EORA sector-year level which allows us to construct a weighted- average of the tari¤s the destination markets impose. Finally, we also rely on the fDi Markets Database which allows us to construct a proxy for FDI at the country-sector-year for a limited sample period 2003-2015, following the approach of Hallward-Driemeier and Nayyar (2019). Speci…cally, we de…ne a measure of the number of cross-border green…eld FDI projects announced cumulating it from 2003 onwards for each country-sector (summing across the projects from any origin country). References Braun, M. 2003. “Financial Contractibility and Asset Hardness,”UCLA, Mimeo. Couharde, C., Delatte, A-L, Grekou, C., Mignon,V., and Morvillier, F. 2017. “EQCHANGE: A World Database on Actual and Equilibrium E¤ective Exchange Rates,” CEPII Working Paper 2017-14. Felbermayr, G., Teti, F., and Yalcin, E. 2019. “Rules of Origin and the Pro…tability of Trade De‡ection,” Journal of International Economics 121: 103248. Hofmann, C., Osnago, A., and Ruta, M. 2017. “Horizontal Depth: A New Database on the Content of Preferential Trade Agreements,” Policy Research Working Paper 7981, World Bank, Washington, DC. Melitz, J., and Toubal, F. 2014. “Native Language, Spoken Language, Translation and Trade,” Journal of International Economics 93 (2): 351-363. Nunn, N. 2007. “Relationship-Speci…city, Incomplete Contracts, and the Pattern of Trade,” Quarterly Journal of Economics 122 (2): 569– 600. Rauch, J. 1999. “Networks Versus Markets in International Trade,”Journal of International Economics 48 (1): 7-35. 43 The UN BEC classi…es goods into capital, consumption and intermediate goods. 3 Table A1: Variable De…nitions and Data Sources Va ria b le n a m e D e …n itio n S o u rc e A v g . ta ri¤ ra te A p p lie d ta ri¤ ra te to m a nu fa c tu re d p ro d u c ts, w e ig hte d m e a n (in % ) WDI F D I in ‡ow s (lo g ) L o g a rith m o f n e t fo re ig n d ire c t inve stm e nt in ‡ow s (in m illio n s o f U S D ) U N C TA D D ista n c e to G V C hu b s (lo g ) L o g a rith m o f su m o f d ista n c e to C h in a , G e rm a ny a n d th e U S (c a p ita l c ity -to -c a p ita l c ity ) CEPII P o litic a l sta b ility in d e x P o litic a l S ta b ility a n d A b se n c e o f V io le n c e / Te rro rism : E stim a te Wo rld G ove rn a n c e In d ic a to rs D o m e stic in d u stria l c a p a c ity (lo g ) M a nu fa c tu rin g va lu e a d d e d (in c u rre nt U S D ) o b ta in e d by m u ltip ly in g n o m in a l G D P w ith th e sh a re o f WDI m a nu fa c tu rin g in va lu e -a d d e d L ow -sk ille d la b o r sh a re / re a l G D P (lo g ) L o g a rith m o f th e sh a re o f e m p loy m e nt o f sk ill le ve l 1 (low ) a c c o rd in g to th e Inte rn a tio n a l S ta n d a rd IL O C la ssi…c a tio n o f O c c u p a tio n s in to ta l e m p loy m e nt http s:/ / w w w .ilo .o rg / p u b lic / e n g lish / b u re a u / sta t/ isc o / isc o 0 8 / in d e x .htm M e d / H ig h -sk ille d la b o r sh a re / re a l G D P (lo g ) L o g a rith m o f th e sh a re o f e m p loy m e nt o f sk ill le ve ls 2 -4 (m e d -h ig h ) a c c o rd in g to th e Inte rn a tio n a l S ta n d a rd IL O C la ssi…c a tio n o f O c c u p a tio n s in to ta l e m p loy m e nt http s:/ / w w w .ilo .o rg / p u b lic / e n g lish / b u re a u / sta t/ isc o / isc o 0 8 / in d e x .htm R e nts fro m re so u rc e s % G D P To ta l n a tu ra l re so u rc e s re nts a s a p e rc e nta g e o f G D P WDI L a n d / re a l G D P (lo g ) L o g a rith m o f la n d a re a (sq . k m ) d iv id e d by re a l G D P (in c o n sta nt 2 0 1 0 U S D ) WDI C a p ita l / re a l G D P (lo g ) L o g a rith m o f re a l c a p ita l sto ck (in c o n sta nt 2 0 1 1 n a tio n a l p ric e s in m il. 2 0 1 1 U S $ ) d iv id e d by re a l G D P P e n n Wo rld Ta b le s 9 .0 , W D I (in c o n sta nt 2 0 1 0 U S D ) E x ch . ra te a p p re c ia tio n C h a n g e in n o m in a l e x ch a n g e ra te (in c re a se is a n a p p re c ia tio n ) WDI N A F TA In d ic a to r va ria b le e q u a l to 1 if th e c o u ntry is a m e m b e r o f N A F TA , a n d 0 o th e rw ise C o nte nt of D eep Tra d e A g re e m e nts EU In d ic a to r va ria b le e q u a l to 1 if th e c o u ntry is a m e m b e r o f th e E U , a n d 0 o th e rw ise C o nte nt of D eep Tra d e A g re e m e nts ASEAN In d ic a to r va ria b le e q u a l to 1 if th e c o u ntry is a m e m b e r o f A S E A N , a n d 0 o th e rw ise C o nte nt of D eep Tra d e A g re e m e nts M ERCO SUR In d ic a to r va ria b le e q u a l to 1 if th e c o u ntry is a m e m b e r o f M E R C O S U R , a n d 0 o th e rw ise C o nte nt of D eep Tra d e A g re e m e nts N b . o f P TA p a rtn e rs (lo g ) L o g a rith m o f th e nu m b e r o f P re fe re ntia l Tra d e A g re e m e nt (P TA ) p a rtn e rs C o nte nt of D eep Tra d e A g re e m e nts D e p th o f P TA s (lo g ) L o g a rith m o f th e nu m b e r o f p rov isio n s in d e e p P TA s a s d e sc rib e d in H o ¤m a n e t a l. (2 0 1 7 ) C o nte nt of D eep Tra d e A g re e m e nts E x ch . ra te m isa lig n m e nt R e a l e x ch a n g e ra te m isa lig n m e nt re la tive to a n e q u ilib riu m e x ch a n g e ra te b a se d o n th e m o d e l d e sc rib e d CEPII in C o u h a rd e e t a l. (2 0 1 7 ) C re d it / G D P D o m e stic c re d it to p riva te se c to r a s a p e rc e nta g e o f G D P WDI T im e to c le a r im p o rts (lo g ) L o g a rith m o f th e nu m b e r o f d ay s re q u ire d to im p o rt b a se d o n th e D o in g B u sin e ss 0 6 -1 5 m e th o d o lo g y D o in g B u sin e ss d a ta b a se L o g istic s p e rfo rm a n c e in d e x O ve ra ll sc o re o f th e lo g istic s p e rfo rm a n c e in d e x o b ta in e d a s a w e ig hte d ave ra g e o f th e c o u ntry sc o re s o n six L P I d a ta b a se ke y d im e n sio n s: e ¢ c ie n c y o f th e c le a ra n c e p ro c e ss (i.e ., sp e e d , sim p lic ity a n d p re d ic ta b ility o f fo rm a litie s) by b o rd e r c o ntro l a g e n c ie s, in c lu d in g c u sto m s; q u a lity o f tra d e a n d tra n sp o rt re la te d in fra stru c tu re (e .g ., p o rts, ra ilro a d s, ro a d s, in fo rm a tio n te ch n o lo g y ); e a se o f a rra n g in g c o m p e titive ly p ric e d sh ip m e nts; c o m p e te n c e a n d q u a lity o f lo g istic s se rv ic e s (e .g ., tra n sp o rt o p e ra to rs, c u sto m s b ro ke rs); a b ility to tra ck a n d tra c e c o n sig n m e nts; tim e lin e ss o f sh ip m e nts in re a ch in g d e stin a tio n w ith in th e sch e d u le d o r e x p e c te d d e live ry tim e . Inte rn e t u se P e rc e nta g e o f th e p o p u la tio n th a t u se s th e inte rn e t WDI E n g lish a s se c o n d la n g u a g e (% ) P e rc e nta g e o f th e p o p u la tio n th a t sp e a k s e n g lish a s a se c o n d la n g u a g e M e litz a n d To u b a l (2 0 1 4 ) Fe m a le la b o r fo rc e p a rtic . (% ) S h a re o f fe m a le s in to ta l la b o r fo rc e WDI Im p o rt e la stic ity A ve ra g e fo r e a ch c o u ntry o f th e im p o rt d e m a n d e la stic itie s e stim a te d by K e e , N ic ita , a n d O la rre a g a (2 0 0 8 ) K e e , N ic ita , a n d O la rre a g a (2 0 0 8 ) fo r a ll its m a nu fa c tu rin g p ro d u c ts d e …n e d a s th o se w ith H S 6 -d ig it c o d e s h ig h e r th a n H S 2 9 0 0 0 0 . 4 To ta l p o p u la tio n (lo g ) L o g a rith m o f th e to ta l nu m b e r o f in h a b ita nts. WDI C o rp o ra te in c o m e ta x e s (% ) S ta tu to ry c o rp o ra te in c o m e ta x ra te in p e rc e nta g e p o ints. Ta x Fo u n d a tio n (2 0 1 9 ) Tra n sitio n a l e c o n o m y sta tu s In d ic a to r va ria b le fo r c o u ntrie s th a t tra n sitio n e d fro m c o m m u n ism to a c a p ita list re g im e . IM F (2 0 0 0 ) H ig h -sk ille d la b o r sh a re (% ) S h a re o f e m p loy m e nt o f sk ill le ve ls 3 a n d 4 (h ig h ) a c c o rd in g to th e IL O Inte rn a tio n a l S ta n d a rd C la ssi…c a tio n o f O c c u p a tio n s in to ta l e m p loy m e nt http s:/ / w w w .ilo .o rg / p u b lic / e n g lish / b u re a u / sta t/ isc o / isc o 0 8 / in d e x .htm C a p ita l / la b o r (lo g ) R e a l c a p ita l sto ck (in c o n sta nt 2 0 1 1 n a tio n a l p ric e s in m il. 2 0 1 1 U S $ ) d iv id e d by to ta l e m p loy m e nt P e n n Wo rld Ta b le s 9 .0 R u le o f law in d e x R u le o f L aw : E stim a te Wo rld G ove rn a n c e In d ic a to rs S k ill inte n sity R a tio o f sk ille d w o rke rs to u n sk ille d w o rke rs a t th e 2 -d ig it le ve l o f S IC re v isio n 1 9 8 7 c o n c o rd e d to th e e ig ht N B E R -C E S M a nu fa c tu rin g In d u stry D a ta b a se se c to rs in th e E O R A d a ta b a se b a se d o n in d u strie s’ ve rb a l d e sc rip tio n s. C a p ita l inte n sity R a tio o f c a p ita l to e m p loy m e nt a t th e 2 -d ig it le ve l o f S IC re v isio n 1 9 8 7 c o n c o rd e d to th e e ig ht se c to rs in th e N B E R -C E S M a nu fa c tu rin g In d u stry D a ta b a se E O R A d a ta b a se b a se d o n in d u strie s’ ve rb a l d e sc rip tio n s. N a tu ra l re so u rc e inte n sity In d ic a to r va ria b le fo r n a tu ra l re so u rc e inte n sive se c to rs in th e E O R A d a ta b a se (w o o d a n d p a p e r; B ra u n (2 0 0 3 ) p e tro le u m , ch e m ic a ls a n d n o n -m e ta llic m in e ra ls; a n d m e ta l p ro d u c ts) C o ntra c t inte n sity S h a re o f sp e c ia liz e d a n d c u sto m iz e d inte rm e d ia te in p u ts u se d in th e p ro d u c tio n o f th e …n a l g o o d fo r e a ch in d u stry N u n n (2 0 0 7 ) 3 -d ig it IS IC re v isio n 2 in d u stry b a se d o n a U .S . in p u t-o u tp u t ta b le . S p e c ia liz e d a n d c u sto m iz e d inte rm e d ia te in p u ts a re th o se n o t so ld o n a n o rg a n iz e d e x ch a n g e n o r re fe re n c e -p ric e d a c c o rd in g to R a u ch (1 9 9 9 ) s re la tio n sh ip s. T h e 3 -d ig it IS IC re v isio n 2 le ve l is c o n c o rd e d to th e e ig ht se c to rs in th e E O R A d a ta b a se b a se d o n in d u strie s’ ve rb a l d e sc rip tio n s. Te ch n o lo g y inte n sity In d ic a to r va ria b le d e …n in g w h e th e r a n IS IC se c to r is te ch n o lo g y -inte n sive o r n o t, b a se d o n re se a rch U N ID O a n d d e ve lo p m e nt e x p e n d itu re in c u rre d in th e p ro d u c tio n o f m a nu fa c tu re d g o o d s. A ve ra g e o u tp u t ta ri¤ Ta ri¤s im p o se d a t h o m e o n im p o rts o f th e se c to r a re c o n stru c te d by …rst a g g re g a tin g ta ri¤s (ta k in g th e W IT S , T R A IN S a n d Fe lb e rm ay r, sim p le ave ra g e ) a n d im p o rts (ta k in g th e su m ) to th e E O R A se c to r le ve l. T h e n im p o rts a re su m m e d a c ro ss Te ti a n d Ya lc in (2 0 1 9 ) c o u ntrie s o f o rig in to o b ta in im p o rt w e ig hts w h ich a re u se d to c o m p u te ta ri¤s im p o se d a t h o m e o n im p o rts o f a se c to r a s a w e ig hte d -ave ra g e ta ri¤ a t th e re p o rtin g c o u ntry -E O R A se c to r-ye a r le ve l. A ve ra g e in p u t ta ri¤ Ta ri¤s im p o se d a t h o m e o n im p o rts o f inte rm e d ia te in p u ts u se d a re c o n stru c te d by …rst ke e p in g ta ri¤ d a ta W IT S , T R A IN S a n d Fe lb e rm ay r, a n d im p o rt d a ta o n ly fo r H S 6 -d ig it p ro d u c ts c la ssi…e d a s inte rm e d ia te s a c c o rd in g to th e U N B E C Te ti a n d Ya lc in (2 0 1 9 ) c la ssi…c a tio n . T h e sim p le ave ra g e o f ta ri¤s is ta ke n a n d im p o rts a re su m m e d a t th e E O R A se c to r le ve l. Im p o rts a re su m m e d a c ro ss c o u ntrie s o f o rig in to o b ta in im p o rt w e ig hts w h ich a re u se d to c o m p u te a n a u x ilia ry m e a su re o f ta ri¤s o n im p o rts o f B E C inte rm e d ia te s a t th e re p o rtin g c o u ntry -E O R A se c to r-ye a r le ve l. T h e n , b a se d o n th e U S In p u t-O u tp u t ta b le fo r 2 0 0 5 fro m W IO D w h ich w a s a g g re g a te d to th e e ig ht E O R A m a nu fa c tu rin g se c to rs, w e c o n stru c t fo r e a ch se c to r th e sh a re o f its to ta l inte rm e d ia te in p u ts th a t o rig in a te s in e a ch o f th e e ig ht E O R A se c to rs. F in a lly, w e c o m b in e th e se I-O ta b le w e ig hts w ith th e a u x ilia ry ta ri¤s o n im p o rts o f B E C inte rm e d ia te s to c o n stru c t a w e ig hte d -ave ra g e ta ri¤ o n inte rm e d ia te in p u ts a t th e re p o rtin g c o u ntry -E O R A se c to r-ye a r le ve l. A ve ra g e m a rke t a c c e ss ta ri¤ Ta ri¤s im p o se d in d e stin a tio n m a rke ts o n e x p o rts o f th e se c to r a re c o n stru c te d by …rst ta k in g th e sim p le W IT S , T R A IN S a n d Fe lb e rm ay r, ave ra g e o f ta ri¤s a n d su m m in g e x p o rts a t th e E O R A se c to r le ve l. T h e n , e x p o rts a re su m m e d a c ro ss Te ti a n d Ya lc in (2 0 1 9 ) d e stin a tio n m a rke ts to c o m p u te e x p o rt w e ig hts a t th e re p o rtin g c o u ntry -E O R A se c to r-ye a r le ve l w h ich a llow s u s to c o n stru c t a w e ig hte d -ave ra g e o f th e ta ri¤s th e d e stin a tio n m a rke ts im p o se . F D I p ro je c ts S u m o f a n n o u n c e d g re e n …e ld F D I p ro je c ts fro m a ll so u rc e c o u ntrie s fo r a g ive n c o u ntry -se c to r-ye a r, c u m u la tive sin c e 2 0 0 3 . S e c to rs a re c o n c o rd e d to th e 8 m a nu fa c tu rin g E O R A se c to rs. Table A2: Summary Statistics No. of obs. Average Median St. Dev. Summary statistics based on country averages in the 1990s, 2000s and in the 2010s EORA backward GVC participation share 290 23.39% 20.85% 13.21% EORA forward GVC participation share 290 19.16% 18.35% 6.85% EORA backward GVC participation (log) 290 7.54 7.30 2.42 EORA forward GVC participation (log) 290 7.44 7.32 2.35 Total exports (log) 290 16.12 16.06 2.22 Med/High-skilled labor / GDP (log) 290 -16.64 -16.84 1.33 Low-skilled labor / GDP (log) 290 -18.62 -18.62 1.50 Rents from resources / GDP 290 6.60 2.34 9.77 Land / GDP (log) 290 -12.83 -12.64 2.02 Capital / GDP (log) 290 -12.11 -12.13 0.46 Avg. tari¤ rate (%) 290 7.64 5.25 6.93 FDI in‡ ows (log) 290 6.92 7.02 2.08 Distance to GVC hubs (log) 290 10.02 10.07 0.30 Political stability index 290 0.01 0.02 0.87 Domestic industrial capacity (log) 290 22.51 22.39 2.26 Exch. rate appreciation 290 -13.55 0.00 220.77 Exch. rate misalignment 290 -7.59 0.59 66.49 NAFTA 289 0.02 0.00 0.15 EU 289 0.11 0.00 0.30 ASEAN 289 0.05 0.00 0.22 MERCOSUR 289 0.04 0.00 0.20 Nb. of PTA partners (log) 266 2.64 2.71 0.93 Depth of PTAs (log) 266 4.95 5.00 1.50 Time to clear imports (log) 191 2.97 2.98 0.67 Female labor force partic. (%) 290 40.67 43.69 8.85 Instruments Import elasticity 290 1.36 1.23 0.46 Import elasticity X Population (log) 290 22.44 19.75 9.30 Population (log) 290 16.20 16.10 1.69 Corporate tax rate (%) 290 28.72 30.00 8.92 Transition economy status 290 19.31% 0.00% 39.54% 5 Table A3: Correlations among country GVC measures and Exports EORA BGVC EORA FGVC EORA BGVC EORA FGVC Total exports part. share part. share part. (log) part. (log) (log) EORA BGVC part. share 1 6 EORA FGVC part. share -0.4492* 1 EORA BGVC part. (log) 0.3721* -0.026 1 EORA FGVC part. (log) 0.0836 0.2394* 0.9431* 1 Total exports (log) 0.1728* 0.098 0.9409* 0.9588* 1 Table A4: Correlations among country determinants M e d iu m L ow - R e nts Land C a p ita l A vg. FDI D ista n c e P o litic a l D o m e stic E x ch . E x ch . N A F TA EU ASEAN N M ERCO SU Rb. of D e p th T im e Fe m a le and sk ille d fro m / GDP / GDP ta ri¤ in - to sta - in d u s- ra te ra te P TA of to la b o r h ig h - la b o r re - (lo g ) (lo g ) ra te ‡ow s G VC b ility tria l a p p re - m is- p a rt- P TA s c le a r fo rc e sk ille d / GDP so u rc e s (% ) (lo g ) hu b s in d e x ca- c ia - a lig n - n e rs (lo g ) im - p a rtic . la b o r (lo g ) / GDP (lo g ) p a c ity tio n m e nt (lo g ) p o rts (% ) / GDP (lo g ) (lo g ) (lo g ) M e d iu m 1 a n d h ig h - sk ille d la b o r / G D P (lo g ) L ow - 0 .8 7 1 0 * 1 sk ille d la b o r / G D P (lo g ) R e nts fro m 0 .1 8 1 0 * 0 .1 1 0 9 1 re so u rc e s / GDP Land / 0 .6 4 7 1 * 0 .5 9 6 9 * 0 .3 9 4 2 * 1 G D P (lo g ) C a p ita l / 0 .4 7 1 8 * 0 .5 5 9 4 * 0 .1 1 0 9 0 .2 9 2 2 * 1 G D P (lo g ) A v g . ta ri¤ 0 .4 0 0 6 * 0 .4 0 4 3 * 0 .1 1 5 1 0 .2 7 6 6 * 0 .2 0 0 9 * 1 ra te (% ) FDI in - - - - - - - 1 ‡ow s (lo g ) 0 .5 6 0 8 * 0 .4 9 1 9 * 0 .1 2 2 0 * 0 .4 8 3 3 * 0 .2 0 2 9 * 0 .3 9 7 8 * D ista n c e to 0 .2 9 7 1 * 0 .4 1 4 5 * 0 .1 6 2 6 * 0 .3 4 5 6 * 0 .0 6 8 8 0 .2 3 3 6 * - 1 G V C hu b s 0 .1 7 3 7 * (lo g ) 7 P o litic a l - - - - - - 0 .2 1 0 6 * - 1 sta b ility 0 .6 5 0 1 * 0 .6 4 5 8 * 0 .2 4 3 1 * 0 .4 1 8 8 * 0 .3 3 2 1 * 0 .2 9 9 5 * 0 .2 4 2 7 * in d e x D o m e stic - - - - - - 0 .8 3 6 9 * - 0 .1 1 6 5 * 1 in d u stria l 0 .5 0 3 1 * 0 .4 4 8 8 * 0 .1 9 8 9 * 0 .4 4 9 5 * 0 .1 9 3 7 * 0 .2 9 4 7 * 0 .2 0 6 1 * c a p a c ity (lo g ) E x ch . ra te 0 .0 1 9 7 - 0 .0 3 0 6 - - - - - 0 .0 1 9 9 - 1 a p p re c ia - 0 .0 0 5 5 0 .0 2 6 3 0 .0 0 3 6 0 .0 9 2 5 0 .0 4 0 2 0 .0 7 7 2 0 .0 7 2 8 tio n E x ch . ra te - - - - - 0 .0 0 9 6 0 .1 2 8 6 * - 0 .1 4 9 9 * 0 .1 4 2 9 * - 1 m isa lig n - 0 .1 9 9 9 * 0 .2 3 1 7 * 0 .0 2 1 1 0 .1 8 8 5 * 0 .2 1 5 0 * 0 .1 0 0 6 0 .0 0 1 5 m e nt N A F TA - - - - - - 0 .3 1 0 8 * -0 .0 6 2 0 .0 5 1 4 0 .3 0 1 3 * 0 .0 1 0 2 0 .0 2 6 2 1 0 .1 6 8 6 * 0 .1 4 3 5 * 0 .0 6 4 3 0 .0 3 2 5 0 .1 4 4 8 * 0 .1 0 3 1 EU - - - - - - 0 .2 3 9 1 * - 0 .3 8 2 0 * 0 .2 5 4 9 * 0 .0 2 3 4 0 .1 1 0 1 - 1 0 .3 5 9 1 * 0 .4 1 1 9 * 0 .2 3 1 5 * 0 .3 3 6 1 * 0 .1 3 5 8 * 0 .2 6 5 9 * 0 .4 9 0 2 * 0 .0 6 2 8 ASEAN -0 .0 3 3 0 .0 0 3 7 - - 0 .1 9 0 0 * - 0 .1 6 9 6 * 0 .1 9 6 1 * - 0 .1 6 9 7 * 0 .0 1 4 9 0 .0 0 9 5 - - 1 0 .0 1 0 8 0 .2 0 0 9 * 0 .0 8 5 1 0 .0 0 3 6 0 .0 3 9 9 0 .0 9 1 4 M ERCO SUR - 0 .0 3 4 7 - 0 .1 0 6 9 -0 .0 4 8 0 .0 5 7 6 0 .0 5 6 8 0 .3 5 0 5 * - 0 .0 9 6 4 - 0 .0 3 1 7 - - - 1 0 .0 5 4 1 0 .0 8 0 3 0 .0 1 7 7 0 .2 5 6 5 * 0 .0 3 5 8 0 .0 8 1 9 0 .0 5 2 1 Nb. of - - - - 0 .0 7 8 4 - 0 .0 9 6 5 - 0 .1 7 9 0 * 0 .0 3 6 8 0 .1 0 2 4 0 .0 2 4 3 - 0 .4 4 2 4 * - - 1 P TA p a rt- 0 .1 4 5 2 * 0 .2 4 9 0 * 0 .0 8 2 4 0 .2 0 7 1 * 0 .2 1 0 9 * 0 .5 3 1 7 * 0 .0 4 6 8 0 .0 5 1 3 0 .3 4 7 3 * n e rs (lo g ) D e p th of - - - - - - 0 .2 1 1 5 * - 0 .3 3 2 4 * 0 .1 5 8 3 * 0 .0 4 2 1 0 .0 5 8 5 0 .0 2 1 3 0 .5 4 4 0 * - - 0 .8 6 7 9 * 1 P TA s (lo g ) 0 .2 7 7 8 * 0 .3 4 7 5 * 0 .2 4 7 2 * 0 .2 5 0 6 * 0 .1 0 8 9 0 .3 4 5 8 * 0 .5 4 6 2 * 0 .2 3 7 8 * 0 .1 3 8 5 * T im e 0 .7 1 1 3 * 0 .6 7 4 9 * 0 .3 7 0 5 * 0 .6 6 0 4 * 0 .3 3 9 3 * 0 .4 0 9 7 * - 0 .2 6 2 3 * - - - - - - - 0 .0 8 4 8 - - 1 to c le a r 0 .5 0 4 3 * 0 .5 6 0 5 * 0 .4 2 1 8 * 0 .2 5 1 2 * 0 .1 1 2 3 0 .1 7 4 9 * 0 .3 5 1 4 * 0 .1 8 5 5 * 0 .3 0 8 0 * 0 .3 6 5 1 * im p o rts (lo g ) Fe m a le 0 .1 3 6 1 * - - 0 .1 0 7 1 - - - - 0 .1 7 4 7 * - 0 .0 2 4 4 0 .0 7 0 3 0 .0 2 2 6 0 .1 6 1 8 * - 0 .0 0 4 7 0 .0 3 5 9 0 .1 8 5 2 * 0 .0 5 6 7 1 la b o r fo rc e 0 .0 1 8 2 0 .1 3 6 4 * 0 .1 7 9 6 * 0 .1 8 0 7 * 0 .0 6 1 1 0 .0 3 7 4 0 .1 3 2 4 * 0 .0 1 0 3 p a rtic . (% ) Table A5: Summary statistics on country-sector GVC participation measures and determinants Number of observations Average Median Standard deviation Summary statistics based on country-sector current year in period 1999-2015 EORA backward GVC participation (log) 14387 4.57 4.40 3.15 EORA forward GVC participation (log) 14387 3.96 3.71 3.10 Exports (log) 14387 5.94 5.85 3.04 Summary statistics based on country 3-year lags in period 1999-2015 High-skilled labor share (%) 14387 22.02 20.64 13.23 Capital / labor (log) 14387 11.24 11.44 1.30 Rents from resources / GDP 14387 23.51 23.42 2.17 Rule of law index 14387 0.09 -0.12 0.95 Internet use 8871 32.88 25.47 28.44 8 Summary statistics based on sector Skill intensity 13319 0.37 0.40 0.12 Capital intensity 13319 60.50 45.30 41.22 Natural resource intensity 14387 0.37 0.00 0.48 Contract intensity 13319 0.58 0.59 0.18 Technology intensity 8243 0.20 0.00 0.40 Summary statistics based on country-sector 3-year lags in period 1999-2015 Average output tari¤ 14387 0.07 0.06 0.07 Average input tari¤ 14387 0.06 0.05 0.06 Average market access tari¤ 14387 0.04 0.03 0.04 Number of FDI projects 3743 13.80 14.08 2.03 Note: Summary statistics are based on the estimating sample used in Table 8. Table A6: Correlations among country-sector determinants High- Capital Nat. Rule of Average Average Average Number Internet skill endow- resource law X output input market of FDI X IT endow- ment X endow- contract tari¤ tari¤ access projects intensity ment capital ment intensity tari¤ X skill intensity X nat. intensity resource intensity 9 High-skill endowment X skill intensity 1 Capital endowment X capital intensity 0.2715* 1 Nat. res. endow. X nat. res. intensity 0.0648* 0.6496* 1 Rule of law X contract intensity 0.6177* 0.0925* -0.0182* 1 Average output tari¤ -0.4490* -0.2446* -0.1917* -0.4113* 1 Average input tari¤ -0.4596* -0.2104* -0.0915* -0.4357* 0.8965* 1 Average market access tari¤ -0.1697* -0.0962* -0.1265* -0.1671* 0.3014* 0.2561* 1 Number of FDI projects 0.2217* 0.1205* -0.0745* 0.1628* -0.1366* -0.1765* -0.1171* 1 Internet X IT intensity 0.3326* 0.2441* -0.1586* 0.3585* -0.2109* -0.1875* -0.1404* 0.1599* 1 Table A7: Determinants of country-sector GVC participation shares adding FDI Backward GVC partic- Forward GVC partici- Exports (log) ipation (log) pation (log) (1) (2) (3) 3-year lag skilled labor / labor X sector skill intensity 0.100*** 0.073*** 0.102*** (0.010) (0.010) (0.009) 3-year lag capital / labor X sector capital intensity 0.002*** 0.001*** 0.001*** (0.000) (0.000) (0.000) 3-year lag nat. resource / GDP X nat. resource intensity 0.068*** 0.128*** 0.094*** (0.018) (0.018) (0.018) 3-year lag rule of law index X sector contract intensity 0.812*** 1.019*** 0.961*** (0.108) (0.098) (0.096) 10 3-year lag average output tari¤ 0.437 1.479*** 1.211*** (0.420) (0.414) (0.370) 3-year lag average input tari¤ -0.733 -1.336 -0.181 (0.936) (1.009) (0.916) 3-year lag average market access tari¤ -6.414*** -8.564*** -6.794*** (1.166) (1.423) (1.139) 3-year lag number of FDI projects 0.307*** 0.346*** 0.289*** (0.029) (0.030) (0.028) Country*Year FE Yes Yes Yes Sector FE Yes Yes Yes Observations 3,617 3,617 3,617 R-squared 0.877 0.893 0.882 Notes: Robust standard errors in parentheses. ***, **, and * indicate signi…cance at 1%, 5%, and 10% levels, respectively. Table A8: Determinants of country-sector GVC participation shares adding internet Dependent variable is: EORA backward GVC EORA forward GVC Exports (log) participation (log) participation (log) (1) (2) (3) 3-year lag skilled labor endowment X skill intensity 0.100*** 0.099*** 0.101*** (0.007) (0.007) (0.007) 3-year lag capital endowment X capital intensity -0.000 -0.000 -0.000 (0.000) (0.000) (0.000) 3-year lag nat. resource endoment X nat. resource intensity 0.062*** 0.043*** 0.061*** (0.012) (0.013) (0.013) 3-year lag rule of law index X contract intensity 1.180*** 1.319*** 1.240*** (0.107) (0.098) (0.101) 11 3-year lag average output tari¤ -1.592*** 0.656* -0.580* (0.380) (0.346) (0.349) 3-year lag average input tari¤ -0.984 -3.723*** -2.084** (0.867) (0.920) (0.861) 3-year lag average market access tari¤ -6.474*** -7.264*** -6.480*** (0.417) (0.420) (0.419) 3-year lag internet X IT intensity 0.009*** 0.010*** 0.010*** (0.002) (0.001) (0.001) Country*Year FE Yes Yes Yes Sector FE Yes Yes Yes Observations 8,911 8,911 8,911 R-squared 0.913 0.907 0.910 Notes: Robust standard errors in parentheses. ***, **, and * indicate signi…cance levels of 1%, 5%, and 10%, respectively.