Policy Research Working Paper 10321 Deep Trade Agreements and Firm Ownership in GVCs Peter H. Egger Gerard Masllorens Development Research Group & Macroeconomics, Trade and Investment Global Practice February 2023 Policy Research Working Paper 10321 Abstract This paper focuses on the effect of preferential trade agree- that preferential trade agreements boost vertical interna- ments and their depth on firm ownership, in particular, tional investment links (both backward and forward) while along global value chains. It measures shareholder-affiliate reducing horizontal investment. Deep preferential trade ownership links at the country-sector-pair level to discern agreements stimulate investment particularly for sector between vertical and horizontal links. The findings show pairs, where a high input specificity prevails. This paper is a product of the Development Research Group, Development Economics and the Macroeconomics, Trade and Investment Global Practice. 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 pegger@ethz.ch and gmasllorens@ethz.ch. 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 Deep Trade Agreements and Firm Ownership in GVCs , and Gerard Masllorens‖ Peter H. Egger¶ Keywords: PTAs; PTA depth; Foreign ownership; Foreign direct investment ¶ ETH Zurich, CEPR, CESifo, WIFO; Email: pegger@ethz.ch; Address: Leonhardstrasse 21, 8092 Zurich, Switzerland. ‖ ETH Zurich. Email: gmasllorens@ethz.ch; Address: Leonhardstrasse 21, 8092 Zurich, Switzer- land. This paper has benefited from support from the World Bank’s Umbrella Facility for Trade trust fund financed by the governments of the Netherlands, Norway, Sweden, Switzerland and the United Kingdom. 1 1 Introduction Preferential Trade Agreements (PTAs) are a key instrument to conduct trade policy and the main one to extend a preferential treatment to trading partners in the long run. The proliferation of PTAs, particularly since the 1990s, has been well documented (Hofmann et al., 2017). Naturally, this led to a vast literature on the effects of such agreements. Pri- marily, these studies focus on the tariff-reduction effect of PTAs and find a positive association between PTAs and trade flows (Baier and Bergstrand, 2009; Egger et al., 2011; Caliendo and Parro, 2014; Anderson and Yotov, 2016). An eminent literature has studied normative as well as positive aspects of PTAs as preferential tariff agreements. However, in particular the rising importance of services – where tariffs do not apply and only non-tariff barriers (NTBs) – and the efforts to standardize provisions around, declarations of, and the measurement of NTBs at the time of the Uruguay Round at the World Trade Organization have put non-tariff aspects in PTAs into the limelight. This could be seen as the wake of research around non-tariff aspects in PTAs and the literature on the depth of PTAs. Work on the depth of PTAs is much younger than that on exclusively tariff- reducing PTAs. A small body of theoretical work established normative insights into deep PTAs (Bagwell and Staiger, 2001; Maggi and Ossa, 2021; Grossman et al., 2021; Parenti and Vannoorenberghe, 2022). In parallel to the emergence of theoretical work, substantial efforts were made to delineate the key non-tariff and non-trade provisions in PTAs (see, e.g., Hofmann et al., 2017). The measurement of PTA content led to new empirical work on the determinants and effects of PTAs, which essentially meant parting with their bi- nary characterization. Hence, data on the depth of PTAs enabled research to look into their heterogeneous effects beyond tariff reductions (see the various chapters in Mattoo et al., 2020). While non-goods-trade provisions in “new” PTAs, namely ones that were signed since the 1990s, are frequent and key, much of the work on the consequences of PTAs still focuses on heterogeneous depth-related effects of PTAs on goods trade (see Egger and Nigai, 2015; Aichele et al., 2016; Mulabdic et al., 2017; Mattoo et al., 2022). Some other work focuses on services trade (see Egger and Wamser, 2013a; Gootiiz et al., 2020; Borchert and Di Ubaldo, 2021), global value chains (see Bruhn, 2014; Orefice and Rocha, 2014; Berger et al., 2016; Ruta, 2017; Laget et al., 2020), and foreign direct investment (see Egger and Wamser, 2013b; Osnago et al., 2017, 2019; Kox and Rojas-Romagosa, 2020). This paper contributes to the literature on the effect of deep PTAs on firm-to- 2 firm ownership at the country-and-sector-pair level. Accordingly, it addresses effects at the interface of ones on direct investment as well as global value chains (GVCs). Specifically, we analyze effects of entering deep PTAs in a unique data set on the frequency of shareholder-affiliate links across all pairs of 209 countries and 38 sectors over 9 years between 2007 and 2015. Global input-output tables permit assigning to every shareholder sector and coun- try whether it is up the stream or down the stream of an affiliate sector and country. Hence, every shareholder-affiliate link can be classified as horizontal (within the same sector) or vertical and then forward (the shareholder being up the stream of the af- filiate) or backward (the shareholder being down the stream of the affiliate). Theoretical work on the activity of multinational firms provides guidance as to the expected effect of PTA membership on foreign ownership (see Markusen, 2002; Egger et al., 2007): whereas lower preferential tariffs should reduce the propensity of horizontal ownership, they should increase the propensity of vertical ownership (in both the forward and backward directions). On average, positive effects of PTAs on foreign direct investment appear to dominate (Orefice and Rocha, 2014; Osnago et al., 2017; Kox and Rojas-Romagosa, 2020; Laget et al., 2020). This points to a relative dominance of vertical ownership links, consistent with the findings of Alfaro and Charlton (2009). However, the evidence is implicit only, because, as Kox and Rojas-Romagosa (2020) put it, ”we cannot separate the FDI data between horizontal and vertical FDI ”. In the light of the latter, the present study provides three innovations. First, it provides a new measurement by focusing on the extensive margin of investment in terms of shareholder-affiliate ownership links. Second, it differentiates those links as to be horizontal versus vertical (and then forward versus backward) in terms in the light of GVC data. Third, it identifies parameters and provides insights based on a very large panel data set covering all pairs among 209 countries and 38 sectors over a period of 9 years. The latter permits conditioning on a host of unobservable factors in a high-dimensional fixed-effects design. The latter ensures that the effects of PTAs and their depth can be identified from the time variation in the data – i.e., the new membership in PTAs – only. The key insights from our study are the following. First, entering a PTA raises the number of new foreign ownership links. Second, the latter is completely driven by vertical links, i.e., ones in the forward or backward integration direction. The effects tend to be somewhat stronger in the forward than in the backward direction. The propensity of horizontal ownership links declines with the formation of PTAs. The effects of PTAs on vertical ownership links increase with a higher PTA depth. Finally, PTA effects on vertical investment are stronger, if the specificity of inputs 3 for a sector pair the shareholder and the affiliate belong in is higher on average. 2 Data In this paper we use a unique combination of data sets that allow us to explore the effects of PTA on firm ownership. First, we obtain the data on firm ownership from the Bureau van Dijk’s OR- BIS data set. Our main explanatory variable -PTAs- comes from the Deep Trade Agreement Dataset prepared by the World Bank, which also includes a detailed text analysis of every treaty’s content. Finally, we use World Input-Output Tables from the WIOD to obtain different measures of GVC organization. 2.1 Firm Ownership Data The ORBIS data set extensively compiles firm level data such as annual accounts and ownership structure for the period 2007-2018. For the purpose of this analysis the most relevant information is the ownership structure. In this data set a link is defined as an ownership relation of any kind (regardless of the share of ownership) between a parent firm located in country j and sector s and an affiliate located in country i and sector r. To clean the data set, we drop the duplicated entries and also those observations with relevant information missing such as country or sector. Furthermore, we keep a panel of incumbents (observed during the full sample) and entrants (firms born during the sample). Finally, we aggregate these data at the country-sector-to-country-sector level and fill in the 0s. In the process we create a new variable called number of connected rs firms (CFij,t ), that counts the number of firms in country i and sector r that are owned by firms from sector s in country j . Note that given the number of countries (209) and sectors (38) this data set is huge. More concretely, we have 209 × 209 × 38 × 38 = 63 million observations per year, which represent almost 600 million observations in total. Given the size of the sample and to avoid computational problems, we have to focus separately on the frequency and the propensity of any ownership as two types of extensive foreign investment margins. First, we focus on the (non-zero) ownership rs counts and use log(CFij ). Moreover, we use a binary variable indicating the existence rs or not of any ownership link, 1(CFij,t ) as a dependent variable. 4 Figure 1: Log of Number of Connected Firms. 2.2 PTA Data The Deep Trade Agreement Dataset of the World Bank is the most comprehensive database of PTAs. It includes 279 treaties from 1958 to 2015. Even more interest- ingly, this database also includes a full text analysis of each treaty and a detailed codification on the inclusion of 52 different provisions. Given our empirical specification, which will include country pair fixed effects, we can only identify the effects of those treaties that came into force during the period 2007-2015. Nevertheless, it is the case that this includes almost half of the treaties (112 new treaties) that generate around 4,000 new dyadic relations (see table 1) that involve a rich variety of countries (see table 2). 5 Table 1: PTAs coming into force 2007-2015. Year PTAs Total depth Core depth WTO-X depth 2007 11 0.415 0.833 0.193 2008 17 0.354 0.784 0.126 2009 18 0.366 0.809 0.132 2010 12 0.359 0.801 0.125 2011 11 0.423 0.808 0.219 2012 15 0.390 0.763 0.192 2013 12 0.466 0.801 0.289 2014 10 0.508 0.844 0.329 2015 6 0.455 0.852 0.245 In our empirical analysis, we define different measures to account for PTAs. One measure is simply an indicator variable, P T Aij,t , that equals 1 if there is a PTA in force between countries i and j in year t. This measure of PTAs is, however, too broad and does not take into consideration the intrinsic heterogeneities between different treaties. To account for differences between treaties, therefore, we make use of the rich set of provisions coded in the database and define various variables to measure the depth of every PTA. More concretely, it is possible to classify the different provisions into 2 groups: (i) WTO+ which includes provisions already covered by the WTO (14 provisions) and (ii) WTO-X which includes those provisions that go beyond the current WTO mandate (38 provisions). Moreover, there are some provisions that have been recog- nized in previous studies (Baldwin, 2008; Damuri et al., 2012) as being more relevant than others. This group of provisions are named ”core” and include WTO+ plus competition policy, investment, movement of capital and intellectual property rights (18 provisions). Hence, we define three different depth measures: 52 Provisionp p=1 Total Depth = 52 18 Provisionc Core Depth = c=1 (1) 18 34 Provisionx WTO-X Depth = x=1 34 where Provision indicates in a binary way whether a given provision is included in the agreement or not. 6 Table 2: New dyadic PTA relations by country in period 2007-2015 (top 15 countries of 122). Country New dyadic relations Total depth Core depth WTO-X depth Romania 95 0.577 0.905 0.403 Bulgaria 95 0.577 0.905 0.403 Rest of EU (each country) 87 0.602 0.907 0.441 Moldova 46 0.641 0.804 0.555 Croatia 44 0.629 0.933 0.469 Korea, Rep. 43 0.440 0.903 0.195 Peru 43 0.620 0.926 0.458 Montenegro 41 0.274 0.775 0.009 Yugoslavia 40 0.259 0.704 0.024 Bosnia and Herzegovina 39 0.253 0.702 0.015 Colombia 39 0.642 0.922 0.494 Costa Rica 38 0.753 0.939 0.654 Honduras 37 0.758 0.913 0.676 El Salvador 34 0.788 0.920 0.719 Guatemala 34 0.784 0.923 0.710 2.3 Global Value Chains Data The WIOD data set is a widely used Global input-output tables source. We use the the information contained in the 2016 release which covers 43 countries and 56 ISIC Rev. 4 two-digit (primary production, manufacturing, and services). In order to match the WIOT data with the information contained in ORBIS, we aggregate the 56 sectors up so as to obtain 38 sectors. Moreover, we group the countries in 22 major world regions1 according to the detailed United Nations geoscheme and substitute coefficients for those countries in ORBIS which are not specifically contained in WIOT by the respective annual group average. For a more formal account of the WIOT-data construction for our purpose, let us closely follow the notation in Antr` as and Chor (2018) and define a world economy with J countries (indexed by i or j ) and S sectors (indexed by r or s). Also let us rs use Zij,t as total value of inputs used by country j ’s sector s originating from country r r i’s sector r in year t; Fi,t and Yi,t as the total value of the final goods sold and the gross output by industry r in country j , respectively. These basic definitions serve to define three measures which are informed by and reflective of a country-sector pair’s positioning in the global value chain. These 1 Northern America, Central America, Caribbean, South America, Northern Africa, Western Africa, Middle Africa, Eastern Africa, Southern Africa, Southern Europe, Western Europe, North- ern Europe, Eastern Europe, Western Asia, Central Asia, Southern Asia, Eastern Asia, Southeast- ern Asia, Australia and New Zealand, Micronesia, Polynesia, Melanesia 7 measures are the following. rs Input coefficient. Given that Zij,t is measured in U.S. dollars. It is useful to define a currency-free input coefficient ars rs s ij,t = Zij,t /Yj,t . Moreover we can aggregate ars ij,t across supplying countries to obtain J J rs i=1 Zij,t ars j,t = ars ij,t = s (2) i=1 Yj,t The latter measures the normalized inputs used by sector-s of country j in its production sourced from sector-r output (regardless of its geographic origin) in year t. In what follows we associate a high input coefficient of the parent firm to backward integration. Output coefficient. By the same token, we can define brs rs r ij,t = Zij,t /Yi,t as a currency-free output coefficient. In turn, this can also be aggregated across using countries j to obtain J J rs rs rs j =1 Zij,t bi,t = bij,t = r (3) j =1 Yi,t The latter measures the normalized output sold by country i’s sector-r geared toward sector-r (regardless of the country) at year t. In what follows we associate a high output coefficient of the parent firm to forward integration. Upstreamness. While the previous two measures provide information about the connectedness of two sectors for a given (making or using) country, they are not immediately informative about the overall relative positioning of any sector and country or any sector-to-sector link and country in the global value chain. To deter- mine the general upstreamness of a country-sector pair, we can iteratively make use of ars ij,t to obtain: S J S J S J r r Yi,t = Fi,t + ars s ij,t Fj,t + ars st t ij,t ajk,t Fk,t + ... (4) s=1 j =1 s=1 j =1 t=1 k=1 This equation shows that the output of a country-sector can be expressed as that supplied directly for final consumption, plus that supplied in the production for final consumption in all country-sectors, plus that supplied to a supplier for final consumption in all country-sectors, etc. From equation 4 we can define a measure of upstreamness as: S J S J S J r r Fi,t s=1 j =1 ars s ij,t Fj s=1 j =1 t=1 k=1 ars st t ij,t ajk,t Fk,t Ui,t = 1× r +2× r +3× r +... > 1 Yi,t Yi,t Yi,t (5) 8 Higher values of Uir represent a larger degree of upstreamness. r When defining the JS × 1 vectors U and Y which have typical elements Ui,t r and Yi,t , respectively, as well as the JS × JS matrix A which has typical elements ars ij,t , and the JS × JS identity matrix I, the vector of upstreamness can be elegantly obtain as U = [I − A]−1 Y ⊘ Y, (6) where ⊘ indicates an elementwise division. Downstreamness. Analogously, using brs r ij,t and V Ai,t as the value added created by sector r in country i we can define a measure of the general downstreamness of a country-sector pair: S J rs S J S J r V Ari,t s=1 j =1 bij,t V As j s=1 j =1 t=1 rs st k=1 bij,t bjk,t V At k,t Di,t = 1× r +2× r +3× r +... > 1 Yi,t Yi,t Yi,t (7) r Higher values of represent a larger degree of downstreamness. Di For the analysis at hand, we compute (ars rs r r j,t , bi,t , Ui,t , , Di,t ) for all years in the WIOD and we take the average.2 Table 3: Descriptive Statistics Positive Ownership-link Sample Ownership Propensity Sample Variable Mean Standard Deviation Mean Standard Deviation rs log(CFij,t ) 0.896 1.277 0.009 0.094 PTAij,t 0.457 0.498 0.303 0.459 TotalDepthij,t 0.310 0.363 0.163 0.283 CoreDepthij,t 0.421 0.474 0.24 0.391 WTO-XDepthij,t 0.251 0.314 0.122 0.241 Downstreamnesr i (affiliate) 2.173 0.456 2.293 0.558 Downstreamness j (parent) 2.131 0.437 2.269 0.542 Upstreamnesr i (affiliate) 2.337 0.633 2.262 0.772 Upstreamness j (parent) 2.344 0.623 2.252 0.763 Backward(ars j,t ) 0.039 0.065 0.014 0.036 rs Forward(bi,t ) 0.039 0.067 0.014 0.04 Input specificityr (affiliate) 0.612 0.152 0.571 0.161 Input specificitys (parent) 0.626 0.150 0.571 0.161 BITij,t 0.288 0.453 0.331 0.471 2 The WIOT distinguishes three components of gross output – namely, intermediate uses, final uses, and net inventories – instead of just two (intermediate and final uses). Therefore, we follow Antr`as et al. (2012) in applying a ”net inventory” correction. 9 3 Empirical Analysis: PTA Effects on Firm-to- Firm Ownership in GVCs In this section we explain our empirical strategy and present the main results of rs the analysis. It will be useful to introduce the generic dependent variable Yij,t ∈ rs rs rs {log (CFij,t , 1(CFij,t > 0)}, where log (CFij,t is defined only for positive ownership rs counts for each observation, and 1(CFij,t > 0)} is a binary indicator, which is unity for (any) positive ownership counts and zero else. We will refer to the variation in rs rs log (CFij,t and 1(CFij,t > 0)} as to be informative about the positive count (the extent) and the propensity of any foreign ownership, respectively. Note that in the data the count of all observations {rs, ij, t} is 100, 828, 240. Of the latter, positive firm-to-firm ownership counts exist for only 985, 731 observations. rs In the interest of computational feasibility, we will employ Yij,t generally in linear regressions, irrespective of whether we focus on the positive counts or the propensity of any firm-to-firm ownership. In what follows, we will report on the result based on regressions of the form rs Yij,t = PTA-Measuresij,t βPTA-Measures + GVC-Measuresrs i,t βGVC-Measures + PTA-Measuresij,t × GVC-Measuresrs i,t βInteract + βBIT BITij,t 2015 + r βDomestic,t Domesticij,t + ηij + γ rs + ωi,t s + νj,t + ϵrs ij,t , (8) t=2007 where P T A − M easuresij,t is a vector of various measures on PTAs as introduced above and depending on the specification, GV C − M easuresrs ij,t is a vector of GVC measures of upstreamness/downstreamness or input-output coefficients, BITij,t is the binary indicator for the presence of a ratified BIT between countries i and j , and Domesticij,t is an indicator which is unity whenever i = j in year t. All parameters β r are regression coefficients, {ηij , γ rs , ωi,t s , νj,t } are fixed effects, and ϵrsij,t is a disturbance term. We will generally only report on the parameters β , and they will always be identified using the high-dimensional set of fixed effects mentioned above. 3.1 PTAs and Upstream versus Downstream Ownership in GVCs In this subsection, we employ the aforementioned measures of upstreamness and downstreamness in GVC-Measuresrs j,t . Note that these measures of the positioning of a shareholder or an affiliate in GVCs are country-sector-indexed each. Hence, 10 we measure for every shareholder and affiliate country and sector its degree of up- streamness as well as downstreamness. Due to the country-sector variation of the aforementioned measures, their main effects will be absorbed by the country-sector- time fixed effects. However, their interaction effects with PTA-Measuresij,t can be identified. As indicated above, we will present results regarding the firm-to-firm ownership at two types of extensive margins: the frequency of ownership links and the propensity of any ownership link. Table 4 presents the results for the ownership-link frequency. In the first column, we employ a binary indicator for PTA membership, while in the remaining columns we employ alternative measures of PTA depth. Clearly, the results suggest that PTAs have a positive and stronger effect on vertical ownership links (that is, if either the shareholder or the affiliate is situated up or down the stream of the value chain of each other). The effects appear to be bigger for up- stream ownership links, which is possible, because ownership may be more or less concentrated.3 The positive effects on vertical ownership tend to be bigger with deeper PTAs. However, there is one exception with regard to the latter: deep PTAs appear to have a weaker effect on the acquisition of upstream affiliates. Table 4: Frequency of Ownership Links: Upstreamness and Downstreamness within GVCs rs log(Number of Firm-to-Firm Connections (CFij,t )) PTA Total depth Core depth WTO-X depth PTA-Measuresij,t −1.338∗∗∗ −1.917∗∗∗ −1.442∗∗∗ −2.122∗∗∗ (0.047) (0.072) (0.052) (0.085) PTA-Measuresij,t × upstreamness affiliater i 0.025∗∗∗ 0.022 0.020∗ 0.024 (0.009) (0.014) (0.01) (0.017) PTA-Measuresij,t × upstreamness parents j 0.045∗∗∗ 0.072∗∗∗ 0.051∗∗∗ 0.087∗∗∗ (0.009) (0.013) (0.009) (0.015) PTA-Measuresij,t × downstreamness affiliater i 0.230∗∗∗ 0.375∗∗∗ 0.263∗∗∗ 0.435∗∗∗ (0.013) (0.021) (0.015) (0.025) PTA-Measuresij,t × downstreamness parents j 0.313∗∗∗ 0.429∗∗∗ 0.324∗∗∗ 0.479∗∗∗ (0.013) (0.02) (0.014) (0.023) BITij,t 0.018∗ 0.018∗ 0.018 0.018∗ (0.011) (0.011) (0.011) (0.011) Country-pair FE ✓ ✓ ✓ ✓ Industry-pair FE ✓ ✓ ✓ ✓ Shareholder-country-industry-year FE ✓ ✓ ✓ ✓ Subsidiary-country-industry-year FE ✓ ✓ ✓ ✓ Domestic-year FE ✓ ✓ ✓ ✓ Obs. 985,731 985,731 985,731 985,731 R2 0.568 0.568 0.568 0.568 Standard errors are clustered at country-industry pairs level and reported in parentheses. Downstreamness have been scaled by 10−3 * p < 0.1, ** p < 0.05, *** p < 0.01 In Table 5 we use the same structure as in Table 4 but now focus on the propensity 3 Several shareholders may be involved in one affiliate. Conversely, one shareholder my hold ownership in several affiliates, etc. 11 of any ownership links from one country-sector pair to another one. Table 5 suggests that firms located up the stream of the chain tend to integrate more likely when a PTA comes into force. The opposite appears to be true for firms down the stream of the value chain. Both of these results appear to be intensified by the depth of the PTA, i.e., a deeper PTA increases more integration up the stream and decreases more integration down the stream of the chain vis-a-vis a shallower PTA. Table 5: Ownership Propensity: Upstreamness and Downstreamness within GVCs rs 1(Number of Firm-to-Firm Connections (CFij,t )) PTA Total depth Core depth WTO-X depth PTA-Measuresij,t 0.007∗∗∗ 0.019∗∗∗ 0.012∗∗∗ 0.023∗∗∗ (0.000) (0.001) (0.000) (0.001) PTA-Measuresij,t × upstreamness affiliater i 0.002∗∗∗ 0.005∗∗∗ 0.003∗∗∗ 0.007∗∗∗ (0.000) (0.000) (0.000) (0.000) PTA-Measuresij,t × upstreamness parents j 0.003∗∗∗ 0.008∗∗∗ 0.005∗∗∗ 0.009∗∗∗ (0.000) (0.000) (0.000) (0.000) PTA-Measuresij,t × downstreamness affiliater i −0.003∗∗∗ −0.009∗∗∗ −0.005∗∗∗ −0.011∗∗∗ (0.000) (0.000) (0.000) (0.000) PTA-Measuresij,t × downstreamness parents j −0.005∗∗∗ −0.013∗∗∗ −0.008∗∗∗ −0.015∗∗∗ (0.000) (0.000) (0.000) (0.000) BITij,t −0.001∗∗∗ −0.001∗∗∗ −0.001∗∗∗ −0.001∗∗∗ (0.000) (0.000) (0.000) (0.000) Country-pair FE ✓ ✓ ✓ ✓ Industry-pair FE ✓ ✓ ✓ ✓ Shareholder-country-industry-year FE ✓ ✓ ✓ ✓ Subsidiary-country-industry-year FE ✓ ✓ ✓ ✓ Domestic-year FE ✓ ✓ ✓ ✓ Obs. 102,828,240 102,828,240 102,828,240 102,828,240 R2 0.200 0.200 0.200 0.200 Standard errors are clustered at country-industry pairs level and reported in parentheses. Downstreamness have been scaled by 10−6 * p < 0.1, ** p < 0.05, *** p < 0.01 3.2 PTAs and Vertical Integration Effects on Ownership in GVCs In this subsection, we employ the variation in input and output coefficients for each country pair and sector. Hence, in contrast to the country-sector-variant measures of upstreamness and downstreamness above, we utilize country-sector-to-sector data on the relative input and output dependence in this subsection. Recall that a higher input coefficient indicates a larger degree of backward integration (as the affiliate is a bigger supplier to the parent), while a higher output coefficient indicates a larger degree of forward integration (the affiliate is a bigger user of the parent). We can identify the main effect on the respective input and output coefficients apart from their interaction effects with PTA-Measuresij,t . Again, we will present effects on the positive counts and the propensity of any firm-to-firm ownership in separate tables. 12 Table 6: Positive Ownership Counts: Vertical integration in Global Value Chains rs log(Number of Firm-to-Firm Connections (CFij,t )) PTA Total depth Core depth WTO-X depth Backwardrs rs j (aj,t ) 0.017 −0.031 −0.013 −0.046 (0.104) (0.102) (0.103) (0.101) Forwardrs j (brs j,t ) 0.448∗∗∗ 0.407∗∗∗ 0.420∗∗∗ 0.406∗∗∗ (0.111) (0.109) (0.110) (0.108) PTA-Measuresij,t 0.022∗∗ 0.045∗∗ 0.019∗ 0.065∗∗ (0.010) (0.021) (0.012) (0.031) PTA-Measuresij,t × Backwardrs j −0.175∗ −0.138 −0.135 −0.122 (0.100) (0.136) (0.106) (0.155) PTA-Measuresij,t × Forwardrs j 0.161∗ 0.372∗∗∗ 0.243∗∗ 0.467∗∗∗ (0.095) (0.130) (0.101) (0.149) BITij,t 0.013 0.013 0.013 0.013 (0.011) (0.011) (0.011) (0.011) Country-pair FE ✓ ✓ ✓ ✓ Industry-pair FE ✓ ✓ ✓ ✓ Shareholder-country-industry-year FE ✓ ✓ ✓ ✓ Subsidiary-country-industry-year FE ✓ ✓ ✓ ✓ Domestic-year FE ✓ ✓ ✓ ✓ Obs. 990,033 990,033 990,033 990,033 R2 0.564 0.564 0.564 0.564 Standard errors are clustered at country-industry pairs level and reported in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01 Table 6 reports on the ownership-count effects of PTA-Measuresij,t as such and interacted with the vertical-integration measures. The table is horizontally organized in four columns of results, where the first column is devoted to the binary measure of PTA and the rest to their depth. The binary PTA indicator carries a positive (semi-elasticity) coefficient, and big- ger interdependence in the forward integration direction also boosts the number of ownership links. However, we do not find evidence of a strong and robust interaction between PTA membership and the backward integration direction. Whereas the first column of Table 6 did not acknowledge the heterogeneity of PTAs depending on their depth, we will do so in columns 2 to 4 by defining PTA-Measuresij,t as to contain one of the elements in {Total depthij,t , Core depthij,t , WTO-X depthij,t }, always used in the main effects as well as in the interactions with the GVC measures. These results suggest that PTA membership raises the number of ownership links, in particular, in the forward-integration direction. Moreover, forward integration becomes more attractive with deeper PTAs. Next, we turn to the results regarding the propensity of any ownership links being formed where there were none prior to a PTA. Table 7 is structured in the same way as Table 6, but it involves the binary ownership indicator as a dependent variable. The results suggest a relatively stronger influence of backward integration than of forward integration for the propensity of any ownership. When taking the main effects and the interaction terms together and evaluating the increase of GVC-Measuresrs i,t in one standard deviation, the overall 13 effect of PTAs is positive. Interestingly, conditional on PTA membership, BITs tend to reduce the propen- sity of any ownership in this table. Table 7: Ownership Propensity: Vertical integration in Global Value Chains rs 1(Number of Firm-to-Firm Connections (CFij,t )) PTA Total depth Core depth WTO-X depth Backwardrs rs j (aj,t −0.006∗∗∗ −0.017∗∗∗ −0.015∗∗∗ −0.013∗∗∗ (0.002) (0.002) (0.002) (0.002) Forwardrs rs j (bj,t ) 0.039∗∗∗ 0.034∗∗∗ 0.035∗∗∗ 0.037∗∗∗ (0.002) (0.002) (0.002) (0.002) PTA-Measuresij,t −0.004∗∗∗ −0.008∗∗∗ −0.005∗∗∗ −0.009∗∗∗ (0.000) (0.000) (0.000) (0.000) PTA-Measuresij,t × Backwardrs j 0.120∗∗∗ 0.287∗∗∗ 0.190∗∗∗ 0.343∗∗∗ (0.003) (0.005) (0.004) (0.007) PTA-Measuresij,t × Forwardrs j 0.085∗∗∗ 0.202∗∗∗ 0.131∗∗∗ 0.244∗∗∗ (0.002) (0.004) (0.003) (0.006) BITij,t −0.001∗∗∗ −0.001∗∗∗ −0.001∗∗∗ −0.001∗∗∗ (0.000) (0.000) (0.000) (0.000) Country-pair FE ✓ ✓ ✓ ✓ Industry-pair FE ✓ ✓ ✓ ✓ Shareholder-country-industry-year FE ✓ ✓ ✓ ✓ Subsidiary-country-industry-year FE ✓ ✓ ✓ ✓ Domestic-year FE ✓ ✓ ✓ ✓ Obs. 111,505,680 111,505,680 111,505,680 111,505,680 R2 0.193 0.194 0.194 0.194 Standard errors are clustered at country-industry pairs level and reported in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01 Table 7 suggests that, across all three measures of PTA depth, the propensity of any ownership link being present rises, and more strongly so in the backward rather than the forward-integration direction. In Tables 8 and 9 we further scrutinize on the ownership margin results, using triple interaction terms between the PTA-depth measures, the GVC measures, and a measure of the specificity of inputs. The latter reflects the degree to which inputs are customized and cannot be easily substituted for when traded between firms, and it is defined by following Rauch (1999) in classifying the average products belonging to a supplying sector at stake.4 In discussing the results, we focus on the coefficient on the interaction effect between PTA depth and input specificity and the triple-interaction terms. rs We start presenting the results where the dependent variable is the log(CFij,t ), i.e., the ownership count, in Table 8. This table suggests that a high specificity of inputs – which are supplied by the shareholder or the affiliate, depending on the direction of ownership – increases the integration frequency in the forward direction, while it reduces it in the backward direction. 4 Rauch (1999) product classification is based on the SITC 5-digit level. The association be- tween 5-digit SITC-product categories and 2-digit ISIC sectors as used to cluster firms follows the concordance tables provided by UNCTAD. 14 Table 8: Positive Ownership Counts: Vertical Integration in GVCs and Input Speci- ficity rs log(Number of Firm-to-Firm Connections (CFij,t )) PTA Total depth Core depth WTO-X depth Backwardrs rs j (aj,t 1.035∗∗∗ 0.962∗∗∗ 1.014∗∗∗ 0.944∗∗∗ (0.362) (0.355) (0.358) (0.352) Forwardrs j (brs j,t ) −0.293 −0.461 −0.408 −0.531 (0.381) (0.376) (0.379) (0.373) Specificity affiliate × intput coefficient 1.093∗ 1.317∗∗ 1.249∗∗ 1.435∗∗ (0.631) (0.621) (0.626) (0.613) Specificity parent × output coefficient −1.464∗∗ −1.449∗∗ −1.495∗∗ −1.459∗∗ (0.61) (0.598) (0.603) (0.590) PTA-Measuresij,t 0.532∗∗∗ 0.744∗∗∗ 0.540∗∗∗ 0.856∗∗∗ (0.036) (0.055) (0.039) (0.067) PTA-Measuresij,t × Backwardrs j 0.528 0.982∗∗ 0.617∗ 1.259∗∗ (0.346) (0.476) (0.369) (0.544) PTA-Measuresij,t × Forwardrs j −1.566∗∗∗ −1.974∗∗∗ −1.485∗∗∗ −2.249∗∗∗ (0.34) (0.471) (0.361) (0.541) PTA-Measuresij,t × Input specificity affiliater −0.218∗∗∗ −0.354∗∗∗ −0.249∗∗∗ −0.416∗∗∗ (0.039) (0.058) (0.043) (0.068) PTA-Measuresij,t × Inputspecificity parents −0.583∗∗∗ −0.753∗∗∗ −0.571∗∗∗ −0.843∗∗∗ (0.038) (0.055) (0.041) (0.065) PTA-Measuresij,t × Input specificity affiliater × Backwardrs j −1.474∗∗ −2.197∗∗∗ −1.532∗∗ −2.649∗∗∗ (0.614) (0.839) (0.651) (0.955) PTA-Measuresij,t × Input specificity parents × Forwardrs j 3.020∗∗∗ 4.054∗∗∗ 3.004∗∗∗ 4.682∗∗∗ (0.62) (0.859) (0.658) (0.983) BITij,t 0.013 0.013 0.013 0.012 (0.011) (0.011) (0.011) (0.011) Country-pair FE ✓ ✓ ✓ ✓ Industry-pair FE ✓ ✓ ✓ ✓ Shareholder-country-industry-year FE ✓ ✓ ✓ ✓ Subsidiary-country-industry-year FE ✓ ✓ ✓ ✓ Domestic-year FE ✓ ✓ ✓ ✓ Obs. 990,033 990,033 990,033 990,033 R2 0.565 0.565 0.565 0.565 Standard errors are clustered at country-industry pairs level and reported in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01 Table 9 indicates that a higher input specificity raises the marginal effect of any PTA membership and of PTA depth for the propensity of there being any ownership links. The effect tends to be generally bigger in the forward than in the backward integration direction. Hence, overall, PTAs – and particularly deep ones – raise the propensity of any firm integration, specifically in the forward integration direction, and even more so when the inputs supplied by the parent to the affiliate are more specific and customized than otherwise. 4 Conclusion In this paper, we investigate the effects of PTAs (and their depth) on firm ownership. Thanks to a unique and novel data set that measures counts of ownership links at a country and sector pair level, we can uncover interesting heterogeneities arising when a PTA comes into force. In particular, given the structure of our data, we are the first to look at sector-specific characteristics when it comes to ownership along 15 Table 9: Ownership Propensity: Vertical Integration in GVCs and Input Specificity rs 1(Number of Firm-to-Firm Connections (CFij,t )) PTA Total depth Core depth WTO-X depth Backwardrs rs j (aj,t 0.022∗∗∗ 0.025∗∗∗ 0.023∗∗∗ 0.026∗∗∗ (0.006) (0.006) (0.006) (0.006) Forwardrs rs j (bj,t ) −0.006 −0.003 −0.003 −0.005 (0.007) (0.007) (0.007) (0.007) Specificity affiliate × intput coefficient −0.058∗∗∗ −0.086∗∗∗ −0.078∗∗∗ −0.081∗∗∗ (0.011) (0.011) (0.011) (0.011) Specificity parent × output coefficient 0.079∗∗∗ 0.064∗∗∗ 0.066∗∗∗ 0.073∗∗∗ (0.013) (0.013) (0.013) (0.013) PTA-Measuresij,t −0.018∗∗∗ −0.043∗∗∗ −0.027∗∗∗ −0.052∗∗∗ (0.000) (0.001) (0.000) (0.001) PTA-Measuresij,t × Backwardrs j −0.002 0.026 0.010 0.035∗ (0.008) (0.016) (0.011) (0.020) PTA-Measuresij,t × Forwardrs j −0.022∗∗ −0.084∗∗∗ −0.058∗∗∗ −0.091∗∗∗ (0.009) (0.018) (0.012) (0.022) PTA-Measuresij,t × Input specificity affiliater 0.008∗∗∗ 0.021∗∗∗ 0.012∗∗∗ 0.027∗∗∗ (0.000) (0.001) (0.000) (0.001) PTA-Measuresij,t × Inputspecificity parents 0.017∗∗∗ 0.040∗∗∗ 0.026∗∗∗ 0.048∗∗∗ (0.000) (0.001) (0.001) (0.001) PTA-Measuresij,t × Input specificity affiliater × Backwardrs j 0.232∗∗∗ 0.489∗∗∗ 0.338∗∗∗ 0.574∗∗∗ (0.016) (0.033) (0.022) (0.040) PTA-Measuresij,t × Input specificity parents × Forwardrs j 0.205∗∗∗ 0.556∗∗∗ 0.369∗∗∗ 0.651∗∗∗ (0.018) (0.037) (0.024) (0.045) BITij,t −0.001∗∗∗ −0.001∗∗∗ −0.001∗∗∗ −0.001∗∗∗ (0.000) (0.000) (0.000) (0.000) Country-pair FE ✓ ✓ ✓ ✓ Industry-pair FE ✓ ✓ ✓ ✓ Shareholdert-country-industry-year FE ✓ ✓ ✓ ✓ Subsidiary-country-industry-year FE ✓ ✓ ✓ ✓ Domestic-year FE ✓ ✓ ✓ ✓ Obs. 111,505,680 111,505,680 111,505,680 111,505,680 R2 0.194 0.195 0.195 0.195 Standard errors are clustered at country-industry pairs level and reported in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01 the value chain. Overall, we find a positive effect of PTAs (and their depth) on firm foreign own- ership both for the frequency as well as the propensity of any ownership. Moreover, for the downstream, ownership frequency is increased by more than the upstream one. On the other hand, the propensity of there being any upstream ownership at all increases by more with PTAs than the propensity of any downstream ownership does. A second set of results is related to the direction of integration within GVCs. More concretely, after combining our ownership data with input-output coefficients from input-output tables, we are able to differentiate between horizontal and vertical and, for the latter, between forward versus backward investment. Regarding posi- tive ownership links, we only find a mild positive effect of PTAs on horizontal and vertical forward integration. The strongest effects materialize for the propensity of any ownership. At this margin, we find a clear negative impact of PTAs on horizon- tal integration and a positive effect on vertical integration in both the forward and backward directions. Finally, we shed light on the role of the specificity of inputs in conjunction with PTA membership for vertical integration. We find that a higher input specificity 16 induces a larger positive effect of PTAs on the frequency of forward integration, while the opposite is true for backward integration. A higher input specificity raises the propensity of any integration in both the forward and the backward direction. 17 References Aichele, Rahel, Gabriel J Felbermayr, and Inga Heiland, “TTIP and intra- ucher f¨ European trade: boon or bane?,” Jahrb¨ okonomie und Statistik, ur National¨ 2016, 236 (6), 639–664. Alfaro, Laura and Andrew Charlton, “Intra-industry foreign direct investment,” American Economic Review, 2009, 99 (5), 2096–2119. Anderson, James E and Yoto V Yotov, “Terms of trade and global efficiency effects of free trade agreements, 1990–2002,” Journal of International Economics, 2016, 99, 279–298. Antr`as, Pol and Davin Chor, On the measurement of upstreamness and down- streamness in global value chains, Taylor & Francis Group, as, Pol, Davin Chor, Thibault Fally, and Russell Hillberry, “Measuring Antr` the upstreamness of production and trade flows,” American Economic Review, 2012, 102 (3), 412–416. Bagwell, Kyle and Robert W. Staiger, “Reciprocity, non-discrimination and preferential agreements in the multilateral trading system,” European Journal of Political Economy, 2001, 17 (2), 281–325. The World Trade Organization 2001. Baier, Scott L and Jeffrey H Bergstrand, “Estimating the effects of free trade agreements on international trade flows using matching econometrics,” Journal of international Economics, 2009, 77 (1), 63–76. Baldwin, Richard, “Big-Think Regionalism: a Critical Survey,” CEPR Discussion Papers 6874, C.E.P.R. Discussion Papers June 2008. Berger, Axel, Dominique Bruhn, Andrea Bender, Julia Friesen, Katha- rina Kick, Felix Kullmann, Robert Roßner, and Svenja Weyrauch, Deep preferential trade agreements and upgrading in global value chains: the case of Vietnam number 92, Studies, 2016. Borchert, Ingo and Mattia Di Ubaldo, “Deep Services Trade Agreements and their Effect on Trade and Value Added,” 2021. Bruhn, Dominique, “Global value chains and deep preferential trade agreements: promoting trade at the cost of domestic policy autonomy?,” Technical Report 2014. 18 Caliendo, Lorenzo and Fernando Parro, “Estimates of the Trade and Welfare Effects of NAFTA,” The Review of Economic Studies, 11 2014, 82 (1), 1–44. Damuri, Yose Rizal et al., “21st century regionalism and production sharing practice,” Technical Report, Graduate Institute of International and Development Studies 2012. Egger, Peter and Georg Wamser, “Effects of the endogenous scope of preferen- tialism on international goods trade,” The BE Journal of Economic Analysis & Policy, 2013, 13 (2), 709–731. and , “Multiple faces of preferential market access: their causes and conse- quences,” Economic Policy, 2013, 28 (73), 143–187. and Sergey Nigai, Effects ofdeep versus shallow trade agreements in general equilibrium, Cambridge University Press, , Mario Larch, and Michael Pfaffermayr, “On the welfare effects of trade and investment liberalization,” European Economic Review, 2007, 51 (3), 669–694. , , Kevin E. Staub, and Rainer Winkelmann, “The Trade Effects of Endogenous Preferential Trade Agreements,” American Economic Journal: Eco- nomic Policy, August 2011, 3 (3), 113–43. Gootiiz, Batshur, Giulia Jonetzko, Joscelyn Magdeleine, Juan Marchetti, and Aaditya Mattoo, Services Grossman, Gene M, Phillip McCalman, and Robert W Staiger, “The “New” Economics of Trade Agreements: From Trade Liberalization to Regulatory Con- vergence?,” Econometrica, 2021, 89 (1), 215–249. Hofmann, Claudia, Alberto Osnago, and Michele Ruta, “Horizontal depth: a new database on the content of preferential trade agreements,” World Bank Policy Research Working Paper, 2017, (7981). Kox, Henk LM and Hugo Rojas-Romagosa, “How trade and investment agree- ments affect bilateral foreign direct investment: Results from a structural gravity model,” The World Economy, 2020, 43 (12). Laget, Edith, Alberto Osnago, Nadia Rocha, and Michele Ruta, “Deep trade agreements and global value chains,” Review of Industrial Organization, 2020, 57 (2), 379–410. 19 Maggi, Giovanni and Ralph Ossa, “The political economy of deep integration,” Annual Review of Economics, 2021, 13, 19–38. Markusen, James R, Multinational firms and the theory of international trade, MIT press, 2002. Mattoo, Aaditya, Alen Mulabdic, and Michele Ruta, “Trade creation and trade diversion in deep agreements,” Canadian Journal of Economics/Revue cana- economique, 2022, 55 (3), 1598–1637. dienne d’´ , Nadia Rocha, and Michele Ruta, Handbook of deep trade agreements, World Bank Publications, 2020. Mulabdic, Alen, Alberto Osnago, and Michele Ruta, “Deep integration and UK–EU trade relations,” in “The Economics of UK-EU Relations,” Springer, 2017, pp. 253–282. Orefice, Gianluca and Nadia Rocha, “Deep integration and production net- works: an empirical analysis,” The World Economy, 2014, 37 (1), 106–136. Osnago, Alberto, Nadia Rocha, and Michele Ruta, “Do Deep Trade Agree- ments Boost Vertical FDI?,” The World Bank Economic Review, 04 2017, 30 (Supplement1 ), S 119 − −S 125. , , and , “Deep trade agreements and vertical FDI: The devil is in the de- economique, 2019, 52 tails,” Canadian Journal of Economics/Revue canadienne d’´ (4), 1558–1599. Parenti, Mathieu and Gonzague Vannoorenberghe, “A simple theory of deep trade integration,” 2022. Rauch, James E., “Networks versus markets in international trade,” Journal of In- ternational Economics, June 1999, 48 (1), 7–35. Ruta, Michele, “Preferential trade agreements and global value chains: Theory, ev- idence, and open questions,” World Bank Policy Research Working Paper, 2017, (8190). 20 Appendix 21 Table A1: Sector Description Section Division Description A 01-03 Agriculture, forestry and fishing B 05-09 Mining and quarrying C 10-12 Manufacture of food products, beverages and tobacco products C 13-15 Manufacture of textiles, wearing apparel and leather products C 16 Manufacture of wood and of products of wood and cork, except furniture; etc. C 17 Manufacture of paper and paper products C 18 Printing and reproduction of recorded media C 19 Manufacture of coke and refined petroleum products C 20 Manufacture of chemicals and chemical products C 21 Manufacture of basic pharmaceutical products and pharmaceutical preparations C 22 Manufacture of rubber and plastic products C 23 Manufacture of other non-metallic mineral products C 24 Manufacture of basic metals C 25 Manufacture of fabricated metal products, except machinery and equipment C 26 Manufacture of computer, electronic and optical products C 27 Manufacture of electrical equipment C 28 Manufacture of machinery and equipment n.e.c. C 29 Manufacture of motor vehicles, trailers and semi-trailers C 30 Manufacture of other transport equipment C 31-32 Manufacture of furniture; other manufacturing C 33 Repair and installation of machinery and equipment D 35 Electricity, gas, steam and air conditioning supply E 36-39 Water supply; sewerage, waste management and remediation activities F 41-43 Construction G 45-47 Wholesale and retail trade; repair of motor vehicles and motorcycles H 49-53 Transportation and storage I 55-56 Accommodation and food service activities J 58-63 Information and communication K 64-66 Financial and insurance activities L 68 Real estate activities M 69-75 Professional, scientific and technical activities N 77-82 Administrative and support service activities O 84 Public administration and defence; compulsory social security P 85 Education Q 86-88 Human health and social work activities R-S 90-96 Arts, entertainment and recreation T 97-98 Activities of households as employers; undifferentiated goods- and services-producing activities of households for own use U 99 Activities of extraterritorial organizations and bodies 22