Policy Research Working Paper 9602 Trade Barriers in Government Procurement Alen Mulabdic Lorenzo Rotunno Development Economics, Development Research Group Macroeconomics, Trade and Investment Global Practice Governance Global Practice March 2021 Policy Research Working Paper 9602 Abstract This paper estimates trade barriers in government procure- Results also show that trade agreements with provisions ment, a market that accounts for 12 percent of world GDP. on government procurement increase cross-border flows Using data from inter-country input-output tables in a grav- of services, whereas the effect on goods is small and not ity model, the paper finds that home bias in government different from that in private markets. Provisions contain- procurement is significantly higher than in trade between ing transparency and procedural requirements drive the firms. However, this difference has been shrinking over time. liberalizing effect of trade agreements. This paper is a product of the Development Research Group, Development Economics, the Macroeconomics, Trade and Investment Global Practice, and the Governance 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 amulabdic@worldbank.org and lorenzo.ROTUNNO@univ-amu.fr. 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 Trade barriers in government procurement∗ Alen Mulabdic† Lorenzo Rotunno‡ Keywords : Government Procurement, Trade Agreements, Gravity Equation. JEL codes : F13, F14, F15, H57. ∗ We are grateful to Mario Larch and Federico Trionfetti for insightful exchanges and suggestions. We also thank Serena Cocciolo, Julien Gourdon, Asif Islam, Bill Maloney, Gustavo Piga, and participants to the World Bank Virtual Seminar Series on Deep Trade Agreements for useful comments. This paper is part of World Bank’s ongoing work on Deep Trade Agreements and a background paper for the flagship report on Global Public Procurement and Development Impact. Errors are our responsibility only. 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. † World Bank, 1818 H Street, Washington DC, USA. Email: amulabdic@worldbank.org. ‡ Aix-Marseille Univ., CNRS, EHESS, Centrale Marseille, IRD, AMSE, Marseille, France. Email: lorenzo.rotunno@univ-amu.fr 1 Introduction Government procurement is a major market, accounting for about 12% of world GDP in 2018 (Bosio et al., 2020). Given this important size, public authorities often prefer local to foreign providers in procurement contracts in order to achieve socioeconomic objectives (e.g., promoting “sustainable” local purchases, and the development of small and medium local enterprises).1 Buy-national provisions are prime examples of measures that explicitly exclude foreign firms from government contracts. The Global Trade Alert (GTA) initiative has collected data since 2009 on the adoption of protectionist measures in government procurement. The data show that 56 new discriminatory measures in government procurement were enacted on average each year between 2009 and 2018.2 In parallel with this persistent protectionism, governments have committed to greater market access in government procurement through the WTO Government Procurement Agreement (GPA) as well as targeted provisions within other free trade agreements (TAs). Membership to the WTO GPA expanded from the initial 34 signatory countries (including EU member states) in 1996 to a group of 48 as of 2020. Meanwhile, the number of TAs including enforceable provisions aimed at liberalizing procurement markets has been increasing over time. Figure 1 shows that provisions on government procurement have become more common since 2000. In 2017, a quarter of TAs in force had enforceable provisions on government procurement. 1 We use the terms “government”, “public sector” and “public authorities” interchangeably to indicate public institutions that are buyers in the public procurement market. 2 We include only policies that the GTA database classifies as “red” (i.e., that almost certainly discriminate against foreign firms) in the following policy areas: “Government Procurement: Domestic Price Preference”, “Government Procurement: Local Content Requirement”, “Government Procurement: Market Access Restrictions” and “Government Procurement: Tendering Process”. These protectionist measures involve around 2500 country pairs (one country implementing the policy measure and the other being among the targets) per year. 2 Figure 1: Number of TAs with and without enforceable provisions in government procurement 200 20 N. of TAs that enter into force 150 Cumulative number of TAs 15 100 10 50 5 0 0 19 8 19 0 19 2 19 4 19 6 19 8 19 0 19 2 19 4 19 6 19 8 19 0 19 2 19 4 19 6 19 8 19 0 19 2 19 4 19 6 20 8 20 0 20 2 20 4 20 6 20 8 20 0 20 2 20 4 16 5 6 6 6 6 6 7 7 7 7 7 8 8 8 8 8 9 9 9 9 9 0 0 0 0 0 1 1 1 19 With gov. procurement Without gov. procurement Note: Authors’ calculations on the basis of the Deep Trade Agreements data from Mattoo et al. (2020). We exclude “Partial Scope Agreements”. Against this seemingly contradicting policy landscape (with unilateral discriminatory measures being adopted together with liberalizing ones), this paper aims to assess the importance of trade barriers in government procurement and identify the impact of trade agreements on cross-border flows. We employ data from the Trade in Value-Added (TiVA) database of the OECD on 62 countries between 1995 and 2015. Our preferred definition of purchases by the public sector sums the “General Government expenditures” and the “Public Administration”, “Health” and “Education” output columns of the inter-country input-output (ICIO) tables. Bilateral trade values are distinguished by goods and services. Descriptive trends in the data show that the public sector spends considerably more on services than on goods compared to the private sector. What is more, the import share of expenditure in government procurement relative to the one in the rest of the economy is particularly low in services, although important heterogeneity emerges across countries. To move beyond descriptive evidence and estimate trade barriers in government procurement, we apply a canonical gravity framework relating bilateral cross-border procurement flows to different variables proxying for trade costs (or their inverse) and multilateral resistance terms. We also apply the same gravity framework to bilateral trade in the rest of the economy – what we refer to as the private sector or market – which we use as a benchmark as well as a way to observe possible spillovers of procurement-specific 3 policies. Gravity specifications without country-pair dummies allow us to identify the “border effect” introduced by McCallum (1995) – how much internal trade is larger than international trade – and subsequently estimated with different settings and techniques (Anderson and van Wincoop, 2003; Chen, 2004; de Sousa et al., 2012). This provides a first measure of trade barriers as it identifies the effect of crossing the border on domestic relative to international trade. Results show a large border effect in general, confirming the findings of the literature (de Sousa et al., 2012). Borders in government procurement are thicker than in private markets. The difference is significant and larger in services than in goods, confirming the descriptive evidence. While discrimination of foreign firms contributes to the large border effect in government procurement, other characteristics of procurement contracts unrelated to the nationality of the supplier (e.g., the existence of ‘set-aside’, differences in legal procedures across countries) might well inflate the value of domestic procurement contracts relative to that of contracts awarded to foreign firms. Our preferred specifications control for bilateral fixed effects and hence permit better identification of the effects of trade policies (Baier and Bergstrand, 2007). We focus on provisions aimed at liberalizing government procurement that are included in TAs, while controlling for other TAs (most notably in our sample, EU membership) and membership in the WTO GPA. We find that specific provisions included in TAs distinctively increase cross-border government procurement in services relative to trade in services in the private sector. Our estimates suggest that cross-border procurement in services is 77 percent higher when two countries are part of a TA with provisions on procurement. This effect is reduced when we isolate the influence of EU membership, which takes up almost half of our sample. The results point to an important effect of EU entry on government imports of both manufacturing and services, suggesting that EU directives aimed at opening up public procurement markets have been instrumental in increasing public purchases of goods and services from abroad. According to our findings, trade in public markets between two countries is 40 (for goods) to 60 percent (for services) higher when both are EU members. As for the WTO GPA, we also find robust effects but only for cross-border government procurement in services. Since more than half of the GPA members are EU countries and many engage in ‘deep’ TAs with procurement provisions, the effects of the GPA and those of TAs might be confounded (Anderson et al., 2017). In additional estimations, we find that the trade-creating effect of TAs is driven by ‘unilateral’ provisions – i.e., provisions for which it is difficult to exclude firms from 4 non-member countries. Two pieces of evidence support this finding. First, we estimate the effect of each provision in separate gravity models. While the clustering of provisions makes the interpretation of the findings purely suggestive, we find that measures aimed at fostering transparency and sharing of information (e.g., possibility of e-procurement, availability of statistics on government procurement) have the largest impact on bilateral trade. Second, we single out these provisions that are clearly unilateral (transparency and procedural matters) in a country-specific measure of ‘depth’ of TAs. Results from an extended gravity model (see, e.g., Heid et al. (2021); Beverelli et al. (2018)) show that the border effect in government procurement for services is significantly lower in countries with higher unilateral depth of TAs. Participation in TAs with unilateral provisions increases cross-border procurement flows relative to domestic ones. To examine further the tendency to buy national by public authorities, we estimate an index of “Constructed Home Bias” (CHB) in the public and private sectors (Anderson and Yotov, 2010b; Anderson et al., 2014). This measure is complementary to the border effect as it compares the amount of actual internal trade (as estimated by the gravity equation) relative to internal trade in a counterfactual frictionless trade scenario. Holding overall sales and expenditure fixed, trade barriers of different types around the world (and hence not only the national border) determine the size of the home bias index. Results confirm what anecdotal evidence suggests: home bias in government procurement is large and higher on average than in the private markets. The difference is less striking when we look at goods and services separately. This suggests that governments are more home-biased than the private sector because (i) they source more goods and services locally; and (ii) they spend more in sectors that are more home biased. Looking at variation over time, we find that the home bias in government procurement decreased for most countries in the sample over the 1995-2015 period, but important heterogeneity emerges. The developing countries in our sample including China, Vietnam and India experienced the largest declines in home bias, whereas government procurement has become more national in most developed economies, such as Japan, Germany and France. Developed countries are however the countries with the lowest levels of CHB. The paper contributes to the relatively scant academic literature on government procurement in the context of international trade. Baldwin (1970) was the first to formally analyze the role of government expenditures in a traditional factor proportions model of international trade. His findings that discrimination in public expenditure is inconsequential 5 for trade flows and specialization were revised and confirmed only partly in oligopolistic settings (Miyagiwa, 1991) and with imperfect information (McAfee and McMillan, 1989).3 Within general equilibrium models with increasing returns to scale, Brulhart and Trionfetti (2004) find that trade barriers in government expenditure can actually change the patterns of specialization, while Trionfetti (2001) identifies a significant impact of home-biased government procurement on agglomeration following trade liberalization. In all these papers, home bias in the public sector is treated as a parameter. Trionfetti (2000) provides a first attempt to quantify this home bias. He uses domestic input-output tables – which are part of the ICIO data – for seven European countries to compare import penetration ratios across public and private sectors. Rickard and Kono (2014) uses aggregate trade data in a gravity framework and find that countries with larger government procurement import less, thus suggesting public home bias. We extend this empirical strand of the literature by estimating trade barriers in government procurement directly, by using information from input-output tables, both in absolute terms and relative to the private sector. In doing so, we highlight the importance of the composition of government procurement – i.e., its bias towards services. In this paper, we infer government procurement purchases from inter-country input-output tables in order to estimate the effect of TAs and their provisions on cross-border purchases. Related work has exploited contract-level data mainly for the U.S. and the EU to assess the local bias in public purchases. Using EU data, Herz and Varela-Irimia (2020) find large border effects both across and within European countries adopting a gravity-style estimation approach, and Kutlina-Dimitrova and Lakatos (2014) provide evidence indicating that product market regulation and policies on Foreign Direct investments (FDIs) affect the probability of awarding a procurement contract to a foreign firm.4 Fronk (2014) estimates the effect of TAs in a gravity model using U.S. federal procurement data – as such, he has one buyer (the U.S.) from multiple suppliers. While measuring precisely procurement purchases (at least by certain public entities and for values above certain thresholds)5 , the contract-level data cover only a single or a few countries 3 Cole et al. (2017) extends the model of McAfee and McMillan (1989) to establish an equivalence between price preferences in procurement auctions and import tariffs. 4 Gourdon and Messent (2019) estimate the effect of TAs on the value of procurement contracts awarded by the EU on non-EU firms using both contract-level and aggregate data. Their set of TAs is thus limited to those signed by the EU. 5 The TED (Tenders Electronic Daily) database includes contracts awarded by public authorities (at the national and sub-national level) in countries of the European Economic Area and by EU institutions. Reporting is a requirement if the value of the contract is above a certain threshold (around 5 million euros 6 (like in the case of the EU) and hence do not permit to investigate the effect of TAs and their provisions. We overcome this limitation by using instead inter-country input-output tables to measure government procurement flows. The drawback of this data source is that we do not observe procurement purchases directly, but we employ constructed data based on national accounts. The use of the empirically successful gravity model aims at filtering out noise and identify central tendency in the data.6 Our analysis draws extensively from the large literature on the gravity model of trade (Head and Mayer, 2015; Anderson, 2011) to assess the policy determinants of trade flows. In doing so, we do not attempt to develop a fully-fledged theoretical model that explains, for instance, the allocation of public and private expenditures across sectors. Our focus is on the incidence of trade policy, taking aggregate sales and expenditure as given. Owing to the separability between allocation of resources within and across countries that is common to many models of trade, we infer trade costs in a “conditional general equilibrium” setting (Anderson and van Wincoop, 2004), i.e. taking as given the allocations of resources across type of goods and services in the public and private sectors. As Fronk (2014) shows, a gravity-style empirical model can be derived also from the auction framework of McAfee and McMillan (1989) combined with a standard comparative advantage model à la Eaton and Kortum (2002). We rely on this validity of the gravity framework for analyzing bilateral flows in government procurement and estimating measures of home bias. Further, our work expands the literature on the partial equilibrium effects of trade agreements (Baier and Bergstrand, 2007; Bergstrand et al., 2015) and its provisions (Dür et al., 2014; Kohl et al., 2016; Mattoo et al., 2017) by focusing on trade where the public sector is the buyer.7 When we focus on specific TA provisions on government procurement, we rely on recent work showing how to identify the effect of unilateral trade policy within a structural gravity model (Heid et al., 2021; Sellner, 2019; Beverelli et al., 2018). We isolate provisions, like the pubic sharing of information and statistics, whose application is non-excludable (i.e., conditional on compliance by the member countries, non-member countries benefit as well for construction, and 130 thousand euros for supplies and services). The U.S. Federal Procurement Data System collects contract award data for procurement contracts at the federal level only. 6 Fajgelbaum and Khandelwal (2016) use similar data from the World Input-Output Database (WIOD) to estimate the parameters of a non-homothetic gravity equation. 7 A theoretically-consistent estimate of the comparative statics effect of TAs requires to specify the full general equilibrium model because changes in trade costs generally affect the allocation of resources across sectors. Different assumptions on the underlying structure of the economy can lead to a common formulation of the comparative statics effect of a change in trade costs as reviewed by Costinot and Rodriguez-Clare (2015). Egger et al. (2011), for instance, estimate the full trade effect of TAs. 7 from the content of the provision), and find that they significantly increase cross-border relative to national government procurement of services. The rest of the paper proceeds as follows. In section 2, we briefly discuss the choice of the gravity equation as our empirical framework. In section 3, we describe the empirical specification and the data. Section 4 provides a descriptive analysis of trade data. Section 5 presents the results of the gravity estimations and the home bias indexes. In section 6 we conclude by discussing the policy implications of our results. 2 Theoretical framework In this section, we present our theoretical framework, justify its choice, and describe how we bring it to the data. We aim to define a simple framework that allows us to identify trade barriers in public procurement across countries. The gravity model can serve this purpose. It has been widely used to infer the determinants of bilateral trade and it is consistent with many general equilibrium models of trade (Head and Mayer, 2015). We argue that the gravity equation can be used also to explore the determinants of trade in government procurement. To show this, we work with the simplest theoretical framework that delivers a gravity equation: the one based on the national product differentiation assumption due to Armington (1969)8 , where each country is endowed with a differentiated variety of a type k (in our empirical applications, k corresponds to goods or services). As Anderson (1979) shows, this assumption coupled with CES preferences or technology delivers a gravity equation. To better capture the procurement of goods and service, we consider shipments of intermediate inputs. The private and public market s in each country j sources inputs of type k originated from country i. Crucially, varieties are differentiated by the type s ∈ {p, r} of buyer, where p stands for public and r denotes private market. One way to think about this assumption is that firms are specialized in either the public or private market.9 Let 8 The theoretical framework outlined here, being based on the gravity equation, can be derived from a number of assumptions on the demand and supply sides of the model (Head and Mayer, 2015). Ricardian comparative advantage models á la Eaton and Kortum (2002) and monopolistic competition models with Dixit-Stiglitz-type assumptions deliver a gravity equation. Larch and Lechthaler (2013), for instance, use a monopolistic competition framework to estimate the welfare maximizing share of domestic public procurement. 9 Note that with an endogenous characterization of the supply side (e.g., in a monopolistic competition or 8 k,s Xij denote the value of shipments of good or service k from country i to market (public or private) s of country j . Trade is subject to a variable cost factor tk,s ij > 1 of iceberg type. Given factory gate prices of pk,s k,s k,s k,s k,s i , destination prices are pij ≡ pi tij . Let Ej denote public or private expenditure on good type k in country j and Yik,s the income that suppliers in i derive from selling good k to market s. Governments choose their optimal demand for intermediate input k from country i in order to minimize costs subject to a CES technology, which, for simplicity, is assumed to be equal across public and private markets. The different varieties of intermediate inputs are thus assembled in a composite public good that is transferred to consumers. Consumers derive utility from this public good and a private good aggregate transferred by the private firms.10 Invoking the “trade separability” assumption (Anderson and van Wincoop, 2004), we require only that the allocation of resources across private and public goods can be separated from the allocation of income and expenditures within type (k, s) across countries.11 Under this assumption, the government’s problem can be partitioned solved in two steps. In a first step, the government chooses the level of aggregate expenditure and thus taxation that maximize household’s utility (Larch and Lechthaler, 2013). In a second step, it chooses the optimal mix of spending across type k and sourcing country i, taking as given optimal expenditure for each good type k and hence public expenditure and optimal taxation. Separability implies that only this last step determines directly bilateral trade flows. Crucially, taxation does not affect bilateral trade flows under the “conditional general equilibrium” (Anderson and van Wincoop, 2004), as long as it does not come from border tariffs, which we assume throughout. While this limits the scope of the theory, it enables us to focus on trade costs.12 Under this structure, the CES demand function for intermediate goods is: 1−σ k k,s pk,s k,s i tij k,s (1) Xij = Ej Pjk,s Eaton-Kortum model), labour can freely move across sectors and hence across productions for governments and for other firms. 10 Private firms provide the private good aggregate under perfect competition. 11 Cobb-Douglas preferences across public and private good aggregates satisfy this condition (e.g., see Larch and Lechthaler (2013)). 12 Changes in trade costs affect the optimal allocation of resources across sources of good k , without altering the overall expenditure (and hence income for the exporting countries) on good k . This result clearly hinges upon the type of analysis that we are after. In a full general equilibrium model, changes in the patterns of trade alter factory gate prices and hence income and expenditure. 9 k 1−σ k 1−σ where Pjk,s ≡ i pk,s k,s i tij is the market s price index of good type k – i.e. the unit cost that market s faces to buy a bundle k of intermediate input varieties. The term σk > 1 is the elasticity of substitution between varieties of intermediate good class k and is assumed to be equal across public and private market. Using market clearance on the supply 1−σ k side, Yik,s = j pk,s k,s i tij / k,s Pj k,s Ej to solve for the exogenous factory prices, we obtain the structural gravity model for each buyer s ∈ {p, r}: k,s k,s 1−σ k k,s Ej Yi tk,s ij (2) Xij = Y k,s Pjk,s Πk,s i 1−σ k 1−σ k tk,s ij Yik,s (3) Pjk,s = i Πk,s i Y k,s 1−σ k 1−σ k tk,s ij Ejk,s (4) Πk,s i = j Pjk,s Y k,s where Y k,s ≡ i Yik,s denotes world income generated from supplies of good k to buyer s. The Πi ’s terms are referred to as “sellers’ incidence” or “inward multilateral resistance”, while the price indexes Pj ’s are suitably re-interpreted as “buyers’ incidence” or “outward multilateral resistance” (Anderson and van Wincoop, 2003; Anderson and Yotov, 2010b). These terms summarize the average trade resistance between one country and the rest of the world. The system can be solved for the Pj ’s and Πi ’s terms (up to a scalar) given data on income and expenditure and estimates of the trade cost vector {tij }.13 The structural gravity model can be used to derive a theoretically-consistent index of home bias, defined as the amount of predicted internal trade given trade costs relative to the same internal flow that would arise in a frictionless benchmark. In absence of trade barriers k,s (tij = 1∀i, j ), trade flows are proportional to income and expenditures shares: Xi,i (tij = 1) = Yik,s Eik,s/Y k,s . The “Constructed Home Bias” (CHB) index (Anderson and Yotov, 2010b) 13 A gravity-type equation for bilateral cross-border procurement flows can be obtained also in the framework of Fronk (2014), where prices are determined in a first-price sealed-bid auction similar to McAfee and McMillan (1989). In his model, bilateral flows are still a function of importer-specific terms, exporter-specific terms, and bilateral factors, but the theoretical counterparts of some terms are different from those of eq (2) – e.g., the set of importer-specific terms include the expected (average) price of procurement contracts and a measure of competitiveness of the procurement market. 10 is thus: 1−σ k tk,s (5) CHBik,s ≡ ii Πk,s i Pj k,s This index summarizes how trade costs around the world inflates domestic shipments over international trade, holding aggregate sales and expenditure constant. It thus provide a specific measure of preference for local suppliers that can be computed for both the private and public market. As Anderson and Yotov (2010b) argue, the CHB index is comparable across type of goods and services, countries and over time, does not depend on normalization nor on estimates of σ k . Importantly, it can be estimated given the structure of the gravity equation. As such, the estimated index is meant to capture central tendency in the data and hence share the good empirical properties of the gravity equation.14 The CHB index is derived in eq (5) under a conditional general equilibrium analysis, which means that sales and expenditures (the Y ’s and the E ’s) do not change between the observed and the counterfactual (frictionless) scenario.15 As in Anderson and Yotov (2010b), this approach is consistent with a strict interpretation of CHB as a measure of the incidence of trade costs. 3 Empirical specification and data k,s Given data on the value of bilateral sectoral shipments Xij and proxies for the trade cost function tk,s ij , the parameters of the gravity equation in (2) can be consistently estimated. We follow common practice in the literature and use importer-year and exporter-year fixed effects in our panel regressions to control for the multilateral resistance and the sales and expenditure terms in eq (2). Adding a time subscript, the gravity model that we estimate is: k,s (6) Xij,t = exp mk,s k,s k,s k,s k,s j,t + ei,t + α Tij,t + εij,t 14 Another approach to measure trade cost is to solve the gravity equation in (2) for bilateral trade costs tij ’s (Novy, 2013). This measure however does not directly relate to the concept of home bias and, more importantly, is based on actual data. As we expect important measurement error to affect our data, we consider this approach in our application inferior to the CHB one. 15 Holding taxation fixed, sales and expenditure would vary between the baseline, observed scenario and a counterfactual one because factory gate prices (and, in a supply-side model, factor prices) would change in response to changes in the levels of trade barriers. 11 k,s k,s The term Tij,t is the matrix of possibly time-varying bilateral trade cost variables and αt φk,s (1−σk ) k,s k,s is the associated vector of coefficients: tks ij,t ≡ exp αt Tij,t ; where the empirical parameter φk,s measures the elasticity of ‘true’ trade costs with respect to the ‘observed’ ones, and is allowed to vary by type of good and buyer. The m and e terms denote importer-year and exporter-year fixed effects, each specific to a buyer and type of good. To avoid collinearity and consistently with the structural gravity model in eq (2) (see Anderson and Yotov, 2010b), we normalize exp(mk,s k,s U SA,t ) = 1 ⇒ PU SA,t = 1 in all our estimations. 16 In specifying the trade cost function, we follow two approaches. First, we include time-invariant determinants of trade barriers that have been extensively used in the literature in addition to time-variant and policy-driven variables – including measures that capture changes in trade barriers specific to the public procurement market. In this case, the trade cost function is specified as follows: φk,s (1−σk ) k,s k,s k,s k,s (7) tks ij,t ≡ exp β1 SM CT Yij + β2 DISTij + β3 CON T IGij + β4 COLON Yij k,s k,s k,s k,s (8) exp β5 LAN Gij + β6 LEGALij + β7 T AN OP ROCij,t + β8 T AP ROCij,t k,s k,s (9) exp β9 W T OGP Aij,t + β10 EUij,t where the SM CT Y indicator equals one if i = j , i.e. if the the trade flow is internal. The coefficient β1 thus identifies the (partial) border effect, i.e. how much trade within national borders is different from trade with other countries, controlling for other bilateral determinants of trade and for multilateral resistance terms. We control for a standard set of other time-invariant determinants of bilateral trade. The variable DIST is the the log of the population-weighted bilateral distance (Mayer and Zignago, 2011). CON T IG is a dummy equal to one if the two countries in the pair share a border, COLON Y equals one if the two countries share colonial history, LAN G equals one if the two countries share an official language, and LEGAL is a dummy for common legal origin. These variables are sourced from CEPII (Mayer and Zignago, 2011). The other determinants of trade costs are variables that measure participation in trade agreements, with a focus on trade policies related to government procurement. The T AN OP ROC indicator equals one if the two countries in the pair are part of a TA at time t 1−σk 16 k,s Given our structural interpretation of the model, the normalization implies that Pj,t = k,s Ej,t /EU k,s SA,t exp mk,s j,t . It follows that PU SA,t = 1. 12 without any provision on government procurement. The T AP ROC dummy captures instead country pairs that are involved in TAs that explicitly include a chapter on government procurement. In extensions, we also estimate the impact of single provisions on government procurement as collected in the Deep Trade Agreements (DTA) database (Mattoo et al., 2020). The variable W T OGP A equals one for country pairs where both countries are members of the WTO GPA.17 Among the 62 countries of our sample, 28 are EU members at some point in time. We isolate the distinctive role of the EU by adding a dummy for EU membership. The TA variables are mutually exclusive – i.e., the T AN OP ROC and T AP ROC dummies are turned to zero for EU countries when the EU indicator is equal to one. We first compare estimates of the trade cost function across public and private markets, for goods and services. We focus on the estimates of the coefficient on the SM CT Y dummy as it captures the border effect and hence can used as a first indicator of bias towards local purchases. We then assess whether policy efforts to liberalize government procurement markets have increased cross-border flows. Specifically, we test if the coefficients on the TA and GPA dummies are positive and significant and if they are higher for government than for private flows. The estimated coefficients on these trade policy variables in eq (7) are likely to be biased because the specification does not fully control for unobserved time-invariant heterogeneity that can drive both the propensity to increase cooperation through various agreements and trade flows (Baier and Bergstrand, 2007). To control for this unobserved heterogeneity, we estimate the following specification: (10) φk,s (1−σ k ) tks,F ij,t E k,s ≡ exp β1 k,s T AN OP ROCij,t + β2 k,s T AP ROCij,t + β3 k,s W T OGP Aij,t + β4 k,s EUij,t + γij where the γ ’s terms are bilateral fixed effects that capture unobserved and time-invariant determinants of trade costs.18 Our interest is in the (partial) effect of trade agreements.19 We isolate the effect of 17 While the agreement entered officially into force in 1996, it was firstly singed in 1994. We thus assume that the countries that entered the agreement in 1996 were already de facto members in 1995, the first year of our panel. 18 Collinearity requires further restrictions on the set of bilateral fixed effects. As in Agnosteva et al. (2019), we suppress the time-invariant internal trade cost dummies so that the estimates of international time-invariant trade costs are relative to a geometric mean of the two countries’ internal trade cost: k,s s,k s,k 1/2 exp(γij ) = ts,k ij / tii tij . 19 The effect of trade agreements directly implied by the estimated coefficients in eq (10) is ‘partial’ in 13 EU membership as almost half of the countries in our sample are members of the EU and the agreement is arguably the deepest form of economic integration among states, including specific directives on government procurement. Two policy variables should account for trade liberalization specific to government procurement. One is an indicator for the presence of provisions on government procurement in TAs. We draw from the Deep Trade Agreements database (Mattoo et al., 2020), which includes a section on the presence of around 100 provisions specific to government procurement in each TA. In our baseline specifications, we use a simple dummy, T AP ROC for the presence of provisions on government procurement in an agreement. The other variable is a dummy for membership in the WTO GPA. Started in 1996 and revised in 2014, the agreement aims to ensure national treatment to foreign firms in government procurement markets, although each member defines the areas of commitments (e.g. different public entities, goods vs. service) that can thus vary substantially across countries. Importantly and unlike most of the WTO agreement, the GPA is “plurilateral”, meaning that it binds only its signatories having de facto the same structure of a preferential TA. To take into account how trade barriers around the world create a preference for local purchases, we next estimate an index of home bias, the CHB. Differently from the border effect, the CHB measures how trade frictions shift up the observed internal trade relative to a frictionless benchmark, where international transactions are thus predicted to be much greater. To estimate the index, we manipulate the gravity equation in eq (2) as follows (see Anderson and Yotov, 2010b): φk,s (1−σ k ) k,s k,s Ytk,s Xij,t tk,s ii,t (11) CHB i,t = k,s k,s = 1−σ k Ej,t Yi,t k,s k,s Pj,t Πi,t where the φ term reflects the fact that we observe only an empirical estimates of trade costs (instead of the true ones). The estimated CHB is thus given by the predicted values of the gravity model rescaled by sectoral expenditures and incomes. The predictions are from the gravity specification in eq (10). We first obtain CHB for goods and services separately (and for each market s) and then aggregate those to the country level using the product of sales and expenditure shares as weights, similar to Anderson et al. (2014). contrast to the ‘full’ general-equilibrium effect that takes into account the influence on trade flows through multilateral resistance terms, sales and expenditure. 14 This approach gives consistent estimates of the CHB index if the gravity equation is correctly specified, i.e. if the country-specific fixed effects are consistent estimates of their theoretical counterparts. Fally (2015) shows that this automatically holds true when the Poisson pseudo-maximum-likelihood (PPML) estimator proposed by Silva and Tenreyro (2006) is employed and income and expenditure are consistent with bilateral trade flows k,s k,s k,s k,s (i.e., Yi,t = j Xij,t ; Ej,t = i Xij,t ). The peculiar properties of the estimator implies that the actual income and expenditure values equal the predicted ones, which should normally be used as both are endogenous in the general equilibrium gravity model. We thus employ the PPML estimator, which has the added advantages of controlling for heteroskedasticity in the data and statistically dealing with zero trade flows. The gravity equation in eq (6) is estimated separately for each market s and sector k , although covariances of the estimated coefficients are taken into account when testing significance of the difference in across public and private gravity. To bring this empirical artillery to the data, we need information on bilateral trade that involves the public sector as a buyer, on a large enough sample of countries. Other studies that investigate trade barriers in government procurement employ data from inter-country input-output tables (Riker, 2013; Messerlin and Mirodout, 2012) as these can split public expenditures from national accounts across type of goods and services purchased and country of origin. We thus follow this route and employ data from the TiVA initiative of the OECD. Similarly to other ICIO database (e.g., Timmer, 2012), the TiVA database harmonizes national IO tables and combines them with information from national accounts and bilateral trade statistics in goods and services to obtain an international input-output table (see OECD, 2013b for details). The estimation procedure allocates output from each country and sector to intermediate usage (by all sectors) or final demand across countries. While far from perfect and inevitably rife with measurement errors (especially compared with official trade statistics), this type of data is the only one that enables international comparison of public expenditures across countries and sectors.20 Data on procurement contracts that has been used in related work (Herz and Varela-Irimia (2020); Fronk (2014); see also footonote 5) are limited to one or to a group of countries (U.S. or the EU) and 20 An alternative approach that relies only on official trade statistics is used by Rickard and Kono (2014) (and adopted also by Gourdon and Messent (2019)). It indirectly identifies the effect of trade barriers on cross-border government procurement by allowing the effect of bilateral factors on trade as recorded by official statistics to vary with the size of the government procurement sector by country. An important limitation of this approach in our setting is that it departs from a structural gravity model: total government purchases, that enter overall expenditure in the gravity equation, are allowed to influence the direct effect of trade costs. 15 hence would make identification of the effect of different TAs problematic. We combine the 2018 and 2016 editions of the TiVA database to obtain data on 62 countries, for goods and services, between 1995 and 2015. The 2016 edition covers the 1995-2011 period, while the 2018 edition covers the 2005-2015 period. We employ data from the 2016 edition from 1995 to 2004, and data until 2015 from the latest edition. The sector classifications in the two editions are not fully compatible. The 2018 edition uses the ISIC Rev. 4 classification, whereas the 2016 edition is based on the ISIC Rev. 3 one. While harmonization of the two classifications at the sector-level can be problematic21 , the definitions of the goods and services aggregate sectors, and of the ‘buying’ sectors composing government entities did not change. We therefore conduct our analysis with the aggregates of goods and services. Table A1 in the Appendix lists the goods and services sectors that are included in the data. As for the country composition, Kazakhstan is included in the 2018 edition only, whereas Thailand never reports imports in public procurement (no imports are reported also in the Government Expenditures column). Therefore, we exclude both countries from the analysis. For simplicity, we also exclude the Rest-of-the-World aggregate that is part of the dataset. To measure government procurement flows, we have to define the perimeter of the public sector. The OECD defines public procurement as “intermediate consumption (goods and services purchased by governments for their own use, such as accounting or IT services), gross fixed capital formation (acquisition of capital excluding sales of fixed assets, such as building new roads) and social transfers in kind via market producers (goods and services produced by market producers, purchased by government and supplied to households)” (OECD, 2013a, p.130). We cannot measure gross fixed capital formation by the public sector because the TiVA database, like other ICIO data, does not provide a split between public and private gross fixed capital formation. We measure social transfers in kind with the entries under the “General Government Expenditure” component of final demand. To measure intermediate consumption, we take the column vectors from the input-output matrix that correspond to government entities. The European Commission (2017) proposes three ways to define these entities in an input-output matrix: (i) a “narrow” classification that includes only the “Public Administration” column; (ii) a “typical” definition that adds the “Health” and “Education” columns to the “Public Administration” one; and (iii) a 21 See this note: http://www.oecd.org/industry/ind/tiva-2018-differences-tiva-2016.pdf from the OECD on the subject. 16 “broad” classification that adds to the “typical” one the columns that pertain to utilities, half of the columns with postal and telecommunication services, and one third of the land transport column. The empirical evidence that we present in this and in the next sections relies on the “typical” definition, and we check the robustness of our main results to the use of the “narrow” definition (the results are shown in the Appendix). Because the TiVA data are not detailed enough (i.e., land transportation is included in a broader “transport and storage” sector), we cannot appropriately implement the “broad” definition. Once public procurement is defined, we identify a private market that is suitable for comparisons. The sum of the other columns in the ICIO table and the “Household expenditure” column in the final demand section is the most immediate and comparable definition of ‘private procurement’. This choice can nevertheless lead to an overlapping with government procurement to the extent to which public authorities operate outside the “typical” definition of government (i.e., outside the Public Administration, Health, and Education columns). Such an overlap between the public and private markets should work against finding significant differences in trade barriers between the two. 4 Descriptive trends Before turning to the empirical estimates of the gravity models, we investigate descriptive trends in the data. The objective here is twofold: (i) to identify patterns of expenditures across goods and services in public and private markets as these affect estimated home bias at the country-level; and (ii) to have a first look at trade barriers by looking at import penetration ratios. First, we compute the service expenditure share for each country in the ‘typical’ public and private markets as defined above. Figure 2 reports this share for 1995 and 2015, the first and the last years of our sample. One pattern stands out: government procurement is vastly about services. The average government in our sample devotes to services around 90 percent of total procurement purchases in 1995 and 2015. Importantly, the share of pubic purchases on various services is higher than the same share for the private market.22 22 This pattern is confirmed when we adopt the “narrow” definition of government procurement, as shown in Figure A1 in the Appendix. 17 Figure 2: Services share of purchases in public and private markets 1995 2015 SGP BRN ROU VNM TUR MYS PHL TUN KHM CHN TWN PHL CHN KOR BGR MEX IDN KHM HRV ARG ROW MAR PER JPN VNM MLT KOR ROW USA CHE MLT HUN MYS NOR SVK HRV LVA RUS CZE TWN CHE SVK ZAF IRL LTU CAN JPN CRI RUS SVN SVN CZE IND POL CRI DNK EST ISR POL SGP HUN FIN GBR ESP GRC TUR ISR AUT COL ROU CHL IND TUN USA MAR BGR CAN DEU FIN COL ISL ITA NLD PER AUS CHL ESP EST PRT NZL FRA BEL AUT ZAF DEU LTU BRN GRC NOR GBR SWE BRA IRL ISL CYP PRT BRA FRA MEX HKG BEL SWE NZL SAU ITA IDN LUX NLD DNK AUS SAU LVA ARG CYP HKG LUX 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 Public Private Note: Raw data are sourced from the TiVA database. Public procurement flows are computed using the “typical” definition (see the main text for details). The vertical solid line indicates the average service share in public procurement across countries. The dashed line reports the same average for private markets. We then turn to import penetration ratios defined as the value of imports divided by total expenditures. While purely descriptive, the measure has been used extensively to assess openness to trade, including in government procurement markets (Messerlin and Mirodout, 2012). We compute the import penetration ratios by country and type of purchasing entities (public or private). Figure 3 reports the ratio of public to private import penetration ratios for goods and services. A value greater than one suggests that public markets are more open than private ones. Public markets are less open than private ones in services for all countries, while for goods the picture is more nuanced. Between 1995 and 2015, relative openness of government procurement in services increases, whereas it slightly decreases on average for goods.23 23 When using the “narrow” definition of government procurement in Figure 2 in the Appendix, we find similar results: if anything, the relative government import penetration is lower especially in services. 18 Figure 3: Government import penetration/ Private import ratio by country 1995 2015 TUN VNM MAR MLT CHN ZAF CYP IND CRI TWN CHL MAR BEL LUX LTU NLD PER LVA HKG MEX ZAF CYP POL CHL MLT BGR KOR COL ITA TUR RUS SVK DNK TUN PRT HKG COL SVN PHL ITA HUN CRI ISR CZE SAU EST TWN PER MYS MYS EST LTU IDN HUN JPN CHN NZL KOR ARG POL FRA BRA AUT KHM DEU ROU BRN ROW CAN PRT BRA AUS KHM CAN BGR ESP SWE IDN IRL PHL ROW RUS LUX USA ISL NZL SVN HRV NOR GRC SGP SWE MEX ISR IND DEU ESP FIN LVA SGP VNM JPN GBR DNK FIN GBR CZE BRN USA AUT ROU BEL NLD FRA SVK NOR GRC ISL CHE CHE HRV ARG AUS IRL TUR SAU 0 .5 1 1.5 2 2.5 0 1 2 3 Goods Services Note: Raw data are sourced from the TiVA database. Public procurement flows are computed using the “typical” definition (see the main text for details). While purely illustrative, this descriptive analysis delivers some messages that are relevant to the subsequent econometric analysis. Government procurement is mainly about services, which are generally less traded than goods. These two observations alone mechanically increase home bias in the public sector at the country level as services are weighted more in the public expenditure basket than in the private one. The sector-specific gravity model in eq (6) controls for this influence as it allows comparisons of estimates across public and private markets within the goods or services aggregate. The analysis of import penetration ratios indeed suggests that already within services, public markets are less open than private ones, with important heterogeneity across countries and over time. The ensuing empirical analysis aims to investigate this variation further. 5 Empirical results In this section, we discuss the estimates from the gravity equation (6) and the estimated CHB. The objective is to estimate the border effect in government procurement and the effect of TAs (relative to private markets) by applying the empirical framework described in section 3. 19 To make the analysis clearer and in line with the descriptive evidence, we sum up bilateral trade values over supplying sectors in a goods (including also primary sectors) and a services aggregates. To allow for adjustments over time in trade flows (Piermartini and Yotov, 2016), we use data from six years in four-year intervals (1995, 1999, 2003, 2007, 2011, 2015). We report robustness checks by (i) leaving the dependent variable at the sector level (see Table A1 in the Appendix) and allowing for sector-specific multilateral resistances; and (ii) using the full 1995-2015 yearly panel of country pairs. For each specification and supplying sector, we report the estimates for the private market next to the ones for the government one to ease comparison. Coefficients in bold are significantly different (at the 10% level) across the two markets. 5.1 Gravity results Tables 1 and 2 report the PPML gravity estimates for goods and services, respectively. In columns (1) and (2), we use the ‘pooled’ specification – controlling for exporter-year and importer-year fixed effects (see eq (7)). The large and significant coefficient on the ‘same country’ dummy (SM CT Y variable) gives a strong indication of home bias, especially in government procurement. All other trade cost variables (except for distance) are switched off for same-country pairs, so that the omitted category for the ‘same country’ variable includes country-pairs that have, for instance, no contiguous border, a different official language and are no EU members. For trade in goods (Table 1), the estimates in column (1) imply that, relative to the omitted category, government procurement from local suppliers is around 30 times higher. In private markets, local sales are 23 times higher than international sales. As expected, the border effect is much larger in services, and significantly higher in government procurement than in private markets (Table 2). The estimates in column (1) suggest that government purchases of local services are almost a thousand times the purchases of services from abroad. The disproportionately large border effect in government procurement is only an indication of protectionist trade policies (e.g., ‘buy-national’ policies). Other characteristics that are specific to government contracts and that are not protectionist might still end up favouring local over foreign firms. In the U.S. for instance, set-aside policies that provide preferences for certain categories of firms (e.g., small and medium sized firms, veteran- or Native-owned firms) can indirectly exclude foreign bidders from competition. 20 The common language and legal origin dummy variables in the gravity equation can control, at least partially, for a local bias in government procurement due to cross-country differences in the legal language and procedure. The estimates indeed suggest that speaking the same official language increases significantly cross-border procurement, but the effect is not significantly different than the one on trade in private markets. Sharing the same type of legal system does not seem to affect trade flows. Coefficients on the other time-invariant determinants of trade costs have the expected sign and most of them are statistically significant. Distance has the usual depressing role on bilateral trade, regardless of whether the purchaser is a private or public entity. The negative effect is rather on the lower end of the range of distance effects found in the literature (Disdier and Head, 2008) and, interestingly, it is significantly lower (in absolute terms) in goods and stronger in services for public compared to private purchases. Time-varying and policy-driven determinants of trade are included in columns (1) and (2) of Tables 1 and 2 merely as controls. Omitted variable bias is likely to plague their partial effects on trade. Columns (3) to (10) report the estimates of gravity models with country-pair fixed effects, which absorb the influence of all time-invariant determinants of trade flows (e.g., see the specification in eq (10)). As Baier and Bergstrand (2007) argue, this specification attenuates the endogeneity concerns related to the coefficients on time-varying variables measuring changes in trade policy. The implied effects of trade agreements are ‘partial’ as the multilateral resistance terms, sales, and expenditure are taken as given. Since many countries are part of multiple types of agreements relevant to government procurement (e.g., countries sign deep TAs with procurement provisions, while being part of the WTO GPA and – relevant to our sample of countries – of the EU), we assess the contribution of each type of agreement by adding them progressively to our specification. The PPML estimates in Table 1 point to weak effects of initiatives specific to government procurement (i.e., TAs with provisions on government procurement and WTO GPA) on trade in goods. The estimates in column (5) suggest that cross-border government procurement in goods goes up by 20 percent when two countries join an TA with procurement provisions, but the effect is actually lower than that of joining an TA without those provisions. The effect of the WTO GPA is instead null. Furthermore, we find that the effect of these policies that are meant to liberalize government procurement is even slightly lower than in private markets. These findings suggest that TAs have not been particularly instrumental in boosting cross-border procurement of goods. When it 21 comes to goods trade, policies specific to procurement markets included in TAs might partly be proxying for the effect of provisions in other areas (e.g., non-tariff measures, investments). Trade agreements have instead significant and important effects on cross-border government procurement in services. As Table 2 shows, most of the average trade effect of TAs is driven by those that have specific provisions on government procurement. The estimates in column (5) imply that entering an TAs with provisions on government procurement boosts trade in public markets by 77 percent – the relative effect on private markets being only a 15 percent increase – compared to the 37 percent increase brought about by TAs without procurement provisions. Columns (7) and (9) show that part of this distinctive effect of procurement provisions is driven by other policy initiatives: WTO GPA and the EU. In particular, membership to the WTO GPA increases significantly cross-border public procurement of services – an effect that is even larger than that of TAs with procurement provisions. Entering the EU single market has promoted the most cross-border purchases of services by public authorities, as shown in column (9). Looking at the estimates for private markets, the positive effects of TAs with procurement provisions and of the WTO GPA disappear when we control for the EU dummy (column (10)): as expected, it is EU membership that boosts trade in services between firms, and not trade policies specific to government procurement. Going from column (3) to column (9), the effect of deep TAs with procurement provisions is almost halved, suggesting that the trade creating effects of these TAs partly overlap with those of the WTO GPA and of the EU single market. The confounding effects of deep TAs and the WTO GPA is not surprising. Anderson et al. (2017) find that the legal text of chapters on government procurement in preferential trade agreements is often similar to the one of the WTO GPA (especially of the revised GPA that entered into force in 2014). In our sample of 62 countries, the likelihood that a country is part of a preferential trade agreement with provisions on government procurement with at least another country in a given year equals 95 percent for GPA members, and only 34 percent for non-GPA members. In other words, almost all GPA signatories participate also in a TA with procurement provisions in our sample. The disproportionate presence of EU countries in our sample and the depth of trade agreements signed by the EU with other countries makes identification of separate effects of EU and TA memberships also problematic. Our gravity estimates for services thus indicate that each of these trade policies (preferential trade agreements, WTO GPA, and the EU) have a distinctive trade-creating effect when it comes to government 22 procurement. The evidence on the border effect and on the trade effects of TAs and their provisions is confirmed in three sets of robustness checks, whose results are reported in the Appendix. Tables A2 and A3 show that the estimates of the gravity models are similar when we use an alternative “narrow” definition of government procurement, which excludes purchases recorded in the “Health” and “Education” columns in the TiVA input-output tables. Results are confirmed also if we estimate the gravity models using the full yearly panel (Tables A4 and A5).24 Finally, we confirm our baseline findings when the dependent variable is further disaggregated by the components of the goods and services sectors listed in Table A1. In Tables A6 and A7, the regressions control for country-industry-year fixed effects, consistent with an industry-level gravity model, while we maintain country-pair fixed effects, which reflect the assumption that trade costs vary across services and goods but not within each of the two aggregates. The estimates are close to the baseline ones in Tables 1 and 2. One difference is that the results for goods are even weaker than in the baseline – e.g., WTO GPA membership is now negatively associated with cross-border procurement. Table 1: PPML Gravity estimates, Goods (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Same country 3.439*** 3.147*** (0.176) (0.187) Contiguity 0.239** 0.280** (0.121) (0.126) Common language 0.352*** 0.277*** (0.108) (0.101) Colony 0.290* 0.385 (0.159) (0.240) Ln Distance -0.651*** -0.812*** (0.049) (0.051) Common legal origin -0.002 0.041 (0.083) (0.082) TA -0.086 0.038 0.214*** 0.294*** 0.384** 0.318*** 0.385** 0.318*** 0.381** 0.314*** (0.120) (0.135) (0.042) (0.037) (0.152) (0.098) (0.152) (0.098) (0.152) (0.098) TA w/ procurement 0.168 0.269** 0.191*** 0.288*** 0.212*** 0.274*** 0.163*** 0.218*** (0.106) (0.121) (0.043) (0.038) (0.045) (0.041) (0.048) (0.039) WTO GPA 0.596*** 0.362*** -0.071 0.044 -0.171*** -0.088 (0.119) (0.124) (0.055) (0.054) (0.052) (0.063) EU 0.064 0.126 0.499*** 0.575*** (0.118) (0.105) (0.090) (0.069) Obs 20,886 20,886 20,886 20,886 20,886 20,886 20,886 20,886 20,886 20,886 24 Egger et al. (2020) argue for the use of annual data for the estimation of the dynamic trade effects of TAs. Their results nonetheless confirm that using annual or time-interval data does not affect substantially the contemporaneous effects, which are the focus of our paper. Future work might investigate the anticipated and lagged effects of trade agreements on cross-border government procurement. 23 Note: All regressions include importer-year and exporter-year fixed effects. Columns (3) to (10) include also country-pair fixed effects. In columns (3) and (4), the T A dummy equals one if the two countries in the pair belong to the same TA (with or without provisions on government procurement). Data are for the years 1995, 1999, 2003, 2007, 2011 and 2015. Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. Table 2: PPML Gravity estimates, Services (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Same country 6.992*** 5.802*** (0.183) (0.196) Contiguity 0.186 0.394*** (0.114) (0.128) Common language 0.647*** 0.608*** (0.106) (0.102) Colony 0.674*** 0.890*** (0.184) (0.180) Ln Distance -0.446*** -0.380*** (0.053) (0.057) Common legal origin 0.075 0.067 (0.071) (0.081) TA 0.353** 0.401*** 0.543*** 0.087** 0.318*** -0.207** 0.317*** -0.206** 0.314*** -0.208** (0.158) (0.144) (0.057) (0.041) (0.086) (0.100) (0.085) (0.099) (0.085) (0.099) TA w/ procurement -0.104 -0.071 0.572*** 0.148*** 0.390*** 0.094** 0.326*** 0.034 (0.118) (0.121) (0.064) (0.041) (0.056) (0.043) (0.056) (0.044) WTO GPA 1.001*** 0.736*** 0.427*** 0.132*** 0.302*** 0.041 (0.113) (0.105) (0.058) (0.049) (0.055) (0.052) EU 0.032 0.110 0.669*** 0.337*** (0.128) (0.116) (0.085) (0.065) Obs 20,886 20,886 20,886 20,886 20,886 20,886 20,886 20,886 20,886 20,886 Note: All regressions include importer-year and exporter-year fixed effects. Columns (3) to (10) include also country-pair fixed effects. In columns (3) and (4), the T A dummy equals one if the two countries in the pair belong to the same TA (with or without provisions on government procurement). Data are for the years 1995, 1999, 2003, 2007, 2011 and 2015. Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. The baseline results rely on the use of dummy variables to identify the presence of provisions on government procurement in TAs. The DTA database of Mattoo et al. (2020) however provides also detailed information on the inclusion of specific provisions in each TA. We thus exploit this information to gain a better understanding of the type of measures that can drive the average trade effect of covering government procurement in a TA. The database includes one hundred questions about the treatment of government procurement in the legal text of TAs. We consider the ‘horizontal’ depth of the government procurement chapters (if any) and exclude information about the treatment of certain provisions that are difficult to classify as moves towards opening up the market.25 After this cleaning of the data, we end up with 39 distinct provisions on government procurement. 25 Specifically, we drop questions that measure ‘vertical’ depth – i.e., those about the content of phasing in provisions for developing countries, that compare the content of the provision with the corresponding article of the (revised) WTO GPA, and combine some questions that are mutually exclusive (e.g., whether the chapter covers only goods or goods and services). 24 We perform two distinct exercises to assess the role of different types of provisions. First, we rely on the gravity model with bilateral fixed effects (see columns (3) to (10) in Tables 1 and 2) to assess the trade effect of each provision on trade flows. Specifically, we run separate regressions of bilateral flows on an indicator variable for the presence of a provision, controlling for TAs with other government provisions, TAs without government procurement, WTO GPA and EU membership (the effect of the single provision is thus relative to a country pair that is not part of a TA at a given time). The results of this exercise are purely suggestive of the effect of each provision, because in reality many provisions come in clusters – this would create collinearity between the provision dummy and the dummy for the presence of other provisions on government procurement. In general, the presence of one provision can simply proxy for a broader liberalizing stance in government procurement (as captured by the general dummy variable used in the baseline results). With this caveat in mind, we plot in figure A3 of the Appendix the statistically significant estimated coefficients when we consider government procurement in goods. The provision with the largest trade-creating effect concerns the disclosure of statistics and quantitative information of the country’s procurement market. Provisions about the possibility to further expand coverage in the future (“Expansion coverage” and “Reduction discrimination”) are among the most effective ones. Other provisions that should potentially be more relevant (e.g., whether the provisions in the procurement chapter are enforceable, or whether the chapter covers both goods and services) seem to have a lower effect, which might however be biased because these provisions are likely to trigger the adoption of other ones. Only the six provisions with largest coefficients have a qualitatively larger effect than in the baseline estimates, where all government provisions are bunched together (see Table 1). Figure A4 shows the same results for services. Almost all provisions taken individually contribute significantly to cross-border government procurement of services. As for goods, provisions on the disclosure of information, the availability of electronic auctions and the absence of phase ins for developing countries are among those with the largest impact. The rest of provisions have however similar effects, thus suggesting that it is difficult to disentangle the contribution of each provision. We then adopt a more disciplined approach where we group provisions sharing similar characteristics. The results of the provision-specific regressions suggest that initiatives 25 aimed at enhancing transparency and making procedures more accessible have the highest trade impact. A common trait of these provisions is that, while included in preferential trade agreements, they do not discriminate against non-members – they have a public good component and hence non-member countries cannot be excluded from its use. For instance, firms from non-member countries can access new and more detailed information and statistics about procurement contracts that governments make available as a result of provisions in TAs. To identify the distinctive impact of these provisions, we construct a variable that captures the extent of importers’ “unilateral” liberalization – i.e., liberalization in areas likely to benefit both trade agreement member countries as well as non-members. This variable is equal to the importer’s share of provisions in government procurement areas related to transparency and procedural disciplines ever included in any of its trade agreements.26 The variable is unilateral as it only varies across importers and over time. To estimate its impact in a gravity framework, we follow recent work by Heid et al. (2021); Piermartini and Yotov (2016); Beverelli et al. (2018) and add to our baseline specification (eq (10)) an interaction between the unilateral provision variable and the same-country dummy.27 As noted by these papers, crucial assumptions for identification are that the trade policy measure (i) does not discriminate across trading partners; and (ii) it does not affect domestic trade. Assumption (ii) might be violated in our setting – domestic firms might well benefit from, e.g., the availability of information on upcoming contracts or more transparent procedures –, thus affecting the significance and the sign of the coefficient on the interaction between the unilateral provision variable and the same-country dummy. We interpret this coefficient as indicating the extent to which deep commitments in government procurement affect domestic relative to international trade (see also Anderson et al. (2018) for a similar interpretation in a comparable exercise). Tables 3 and 4 present the results for goods and services, respectively. For goods, the estimates suggest that adopting non-discriminatory measures has a null or even positive effect on domestic vs. cross-border government procurement. In particular, the coefficient 26 These areas include requirements that information on the procurement laws and regulations is published and publicly available and the promotion of electronic auctions. The questions that form the DTA indicator variables on the content of TAs in government procurement cover eight areas: overview, non-discrimination, coverage, procedural disciplines, transparency, dispute settlement, and new issues. 27 Sellner (2019) finds that this approach to identify the effects of non-discriminatory trade policies outperforms other methods such as a two-step estimator where estimates of the importer-year fixed effects in the baseline gravity equation would be regressed on the unilateral variable. 26 on the unilateral depth variable interacted with the same-country dummy turns positive in column (3) of Table 3, where we control for the influence of GDP and GDP per capita of the importer (as proxies for country size and economic development) on the border effect. This seemingly surprising result is confirmed in column (5), where we include interactions with a measure of the quality of institutions (the average across the six categories of the World Bank’s World Governance Indicators (WGI) database, which is found to decrease the border effect, consistent with the evidence from Beverelli et al. (2018)), and with other TA variables that can have non-discriminatory effect – the EU and GPA dummies equal one if the importer is a member of these TAs. The positive and significant interaction effect found also for trade in private markets, where unilateral depth in procurement provisions should not matter, casts doubts on the reliability of these findings. All in all, the results indicate that domestic firms might take advantage of more transparent procedures and open data on government procurement in goods, a the expenses of foreign firms. Different than for goods, results in services, where the border effect is found to be significantly larger in public than private markets (see columns (1) and (2) of Table 2), suggest that countries with deep provisions on government procurement are more open to international trade. The interaction effect in Table 4 is consistently negative and significant across specifications for public markets, and the effect becomes weaker as we control for size and economic development. As an indication that the unilateral depth measures captures policies relevant to government procurement in services, the interaction effect is not significant for private markets. We thus find that deep non-discriminatory provisions on government procurement in TAs have been instrumental in promoting cross-border procurement flows in services, where governments have been buying significantly more locally than firms. An important result from Tables 3 and 4 is that the bilateral trade policy effects become much lower and lose significance as we control for the unilateral components of TAs. One explanation for this intriguing result is that part of the bilateral component of the TA variables is actually unilateral, and hence not identified. Another way to read the finding is that most of the trade effect documented in the baseline specifications is actually non-discriminatory. Importantly, this is the case also for the EU dummy, which was the trade policy variable with the strongest effect in Tables 1 and 2. The negative coefficient on the interaction between the same-country dummy and the EU indicator suggests that a substantial part of this effect comes from trade with both EU and non-EU countries. 27 Table 3: PPML Gravity estimates: unilateral depth in trade agreements. Goods (1) (2) (3) (4) (5) (6) Gov. Priv. Gov. Priv. Gov. Priv. TA 0.358** 0.242** 0.091 -0.015 0.079 0.009 (0.156) (0.106) (0.118) (0.078) (0.125) (0.076) TA w/ procurement 0.133** 0.119** 0.082 0.037 0.067 0.020 (0.053) (0.046) (0.057) (0.061) (0.057) (0.062) WTO GPA -0.223*** -0.250*** -0.120* -0.118* 0.022 0.010 (0.068) (0.070) (0.061) (0.071) (0.069) (0.090) EU 0.472*** 0.503*** 0.314*** 0.358*** -0.161 0.082 (0.092) (0.059) (0.099) (0.073) (0.106) (0.089) Same country × Unilateral Procurement Depth -0.158 -0.461*** 0.537*** 0.192** 0.627*** 0.207** (0.136) (0.117) (0.095) (0.092) (0.116) (0.086) Same country × Ln(GDP) 0.774* 0.569** 0.094 0.704** (0.440) (0.279) (0.456) (0.305) Same country × Ln(GPDpc) -1.464*** -1.129*** -0.707 -1.198*** (0.488) (0.290) (0.504) (0.318) Same country × Institutions -0.368*** 0.006 (0.130) (0.100) Same country × EU (importer) -1.112*** -0.590*** (0.162) (0.178) Same country × GPA (importer) 0.468*** 0.414*** (0.138) (0.160) Obs 23,064 23,064 23,064 23,064 19,220 19,220 Note: All regressions include importer-year, exporter-year and country-pair fixed effects. Data are for the years 1995, 1999, 2003, 2007, 2011 and 2015. Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. Table 4: PPML Gravity estimates: unilateral depth in trade agreements. Services (1) (2) (3) (4) (5) (6) Gov. Priv. Gov. Priv. Gov. Priv. TA 0.026 -0.214** -0.365*** -0.300*** -0.319*** -0.261*** (0.105) (0.101) (0.105) (0.092) (0.101) (0.084) TA w/ procurement -0.001 0.026 -0.083 -0.007 -0.107* -0.023 (0.062) (0.051) (0.068) (0.054) (0.063) (0.050) WTO GPA -0.257*** 0.031 -0.082 0.048 0.364*** 0.266*** (0.087) (0.066) (0.067) (0.068) (0.090) (0.076) EU 0.431*** 0.332*** 0.489*** 0.434*** 0.127 0.078 (0.095) (0.068) (0.090) (0.070) (0.106) (0.086) Same country × Unilateral Procurement Depth -1.492*** -0.028 -0.277** 0.236* -0.295*** 0.150 (0.188) (0.115) (0.128) (0.127) (0.106) (0.119) Same country × Ln(GDP) -1.395*** -1.102*** -1.664*** -1.197*** (0.404) (0.283) (0.426) (0.302) Same country × Ln(GDPpc) 0.789* 1.085*** 1.102** 1.229*** (0.410) (0.309) (0.434) (0.330) Same country × Institutions 0.064 -0.168 (0.136) (0.115) Same country × EU (importer) -0.991*** -0.933*** (0.202) (0.166) Same country × GPA (importer) 1.141*** 0.779*** (0.159) (0.120) Obs 23,064 23,064 23,064 23,064 19,220 19,220 Note: All regressions include importer-year, exporter-year and country-pair fixed effects. Data are for the years 1995, 1999, 2003, 2007, 2011 and 2015. Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. 28 5.2 Constructed Home Bias As the estimates of the border effect in Tables 1 and 2 suggest, government procurement is mostly local in spite of the liberalizing effects of trade agreements, especially in services. The border effect however provides only a partial measure of home bias since it does not take into account the effects of (changes in) trade barriers in other countries – in a gravity model, these influences play out through the multilateral resistance terms (see eq (2)). To overcome this drawback, we estimate the Constructed Home Bias (CHB) index proposed by Anderson and Yotov (2010b) for government procurement and for private markets. The CHB measures how much trade frictions around the world interact in shifting up domestic trade relative to what would be observed in a counterfactual world without trade barriers, holding constant overall sales and expenditure. As such, it aims to capture the general equilibrium interactions of trade barriers (conditional on aggregate sales and expenditure) and hence goes beyond the ‘partial equilibrium’ view of local bias from the perspective of a single country. The index encompasses all types of trade frictions (both ‘man-made’ trade policies and structural barriers) and hence it cannot be interpreted as a measure of protectionism. Estimated CHB indexes in public and private markets are constructed as in eq (11) separately for goods and services. The fixed-effects specification of the trade cost function in eq (10) (whose estimates are reported in columns (9) and (10) of Tables 1 and 2) is used in the estimation as it controls for all time-invariant factors that can affect bilateral trade.28 We first address the question of whether home bias in government procurement is higher than in private markets. Perhaps not surprisingly, the answer is a strong (but not resounding) “Yes”. Figure 4 shows scatter plots of the estimated CHB index in government procurement against the same index for private markets, for goods and services. We take logs of both variables in order to smooth out the visual effect of some extreme outlier. Governments are generally more home biased than firms in their purchasing strategies as most of the observations lie above the 45 degree lines. Government home bias is higher than the private one for 67 percent of the country-year observations in goods, and a similar 62 percent in services. 28 Note that under that specification tk,s k,s ii,t = tjj,t ∀i, j , i.e. the border effect is equal across countries. Estimated trade costs worldwide affect the CHB via the multilateral resistance terms. 29 Figure 4: CHB indexes by sector Goods Services 10 10 8 8 Government CHB (in logs) Government CHB (in logs) 6 6 4 4 2 2 0 0 2 4 6 8 10 0 2 4 6 8 10 Private CHB (in logs) Private CHB (in logs) 45 line To better appreciate differences across public and private markets as well as heterogeneity across countries, Table A2 in the Appendix reports the values of the estimated CHB indexes for goods. Home bias in government procurement is on average 35% higher than home bias in private markets throughout the period. Looking at the average CHBs throughout the period, some differences emerge across countries. Similarly to what Anderson and Yotov (2010a) find for total trade, CHB is massive for small countries like Cambodia, Cyprus, and Brunei. This is because these countries naturally trade a lot with other countries and thus a high share of their income goes through bilateral trade barriers around the world, driving up the multilateral resistance terms. At the other end of the spectrum, CHBs are the lowest for large countries such as the U.S., Japan and China. While governments are more home biased than firms in their purchases of goods, their CHBs have declined more strongly. Only 22 out of 62 countries in our sample experienced an increase in government home bias, whereas CHBs in private markets went up for 32 countries. On average, home bias in government procurement of goods is 10 percent lower in 2015 relative to 1995. In private markets, the average home bias went up on average by 15 percent. Table A3 in the Appendix reports estimated CHBs for services. As for goods, government procurement in services is more home biased than services purchases by private firms – on average, government CHBs are 18 percent higher than private CHBs. Between 1995 and 30 2015, home bias in government procurement of services went up in only 19 countries out of 62 in our sample – a share similar to that for private markets. The average CHB in government procurement is 11 percent lower in 2015 relative to 1995, a decrease that follows the one observed in goods. Unlike for goods, home bias in services went down on average also for private markets. The estimated CHB indexes are then aggregated at the country level as weighted sum of the sectoral CHBs, where the weights are equal to their expenditure-times-sales shares (see Anderson et al. (2014)). Figure 5 plots the country-level government CHBs against the private ones. At the country level, government procurement markets are even more home biased than at the sectoral level. Government home bias is 70 percent higher than private home bias on average (government CHB is higher than private CHB in 71 percent of the country-year observations). This finding suggests that governments spend more (relative to private firms) on aggregate sectors that are more home biased. This composition effect contributes to the already important difference in home bias across government and private markets at the sector level. Figure 5: CHB indexes at the country level 10 Government CHB (in logs) 4 2 06 8 0 2 4 6 8 10 Private CHB (in logs) 45 line Table A10 in the Appendix reports the country-level CHB indexes for government 31 procurement and private firms. The last two columns give the percent change in the CHB for each country in the sample. These numbers are showed graphically in Figure 6. At the aggregate level like at the sector level, we find that home bias in government procurement has gone down over time – the average change equals a 11 percent decrease (remarkably similar to the changes observed in the goods and in the services sectors), and more so than in private markets. Large variation however emerges across countries. The largest percent increases in government home bias are observed among developed economies – e.g., Japan, Germany and France. These countries have however also low levels of CHBs. Developing countries such as China, Vietnam and India report the largest drops in government home bias. Interestingly given the large trade effect of the EU single market (see Tables 1 and 2), most old EU member states experience an increase in home bias in government procurement during the sample period, whereas home bias went down in countries that entered the EU during the period (Croatia being the exception). Figure 5: Relative change in CHB indexes at the country level between 2015 and 1995 CHN VNM IND KHM CRI ROU PER SAU EST CHL LTU IDN MLT BGR MEX LVA SVK RUS KOR AUS SGP COL POL TUR LUX TUN CZE MYS PHL NZL HKG ARG IRL ISL ZAF MAR USA NOR CYP GBR CHE ISR CAN HUN ESP SVN NLD FIN BRA TWN BRN SWE BEL HRV PRT GRC ITA DNK AUT FRA DEU JPN -1 0 1 2 Government Private Overall, the analysis using the CHB indexes reveals that, while government procurement is significantly more home biased than private purchases, it has gone down faster. Important 32 heterogeneity however emerges across countries. While the gravity results suggest that TAs have partly raised cross-border government procurement, the CHB values remind us that government procurement remains vastly home biased, even if the trends show a slow opening up to trade. 6 Concluding remarks This paper estimates trade barriers in government procurement. In doing so, it analyses the role of trade agreements to see if and how much they have contributed to reduce those barriers. Using new Inter-Country Input-Output (ICIO) tables from the TiVA database for 62 countries between 1995 and 2015, we obtain estimates of government purchases across sectors and countries of origin and use those in a standard gravity model. The estimates suggest that governments are significantly more local in their purchasing decisions than private firms. The border effect is large and significantly higher for public markets, especially in services. Yet, we find that trade agreements and the EU single market in particular have contributed to the opening up of government procurement, with the effect being robust in services. Non-discriminatory provisions specific to government procurement are driving the trade-creating effect of TAs in services, whereas the effects are overall weak in goods. The estimates from the gravity model are then used to estimate indexes of home bias in government procurement and private markets. These measures capture the overall tendency to trade locally rather than internationally, and hence include also factors other than protectionist trade policies. Home bias is larger in government procurement than in private markets. Two mechanisms can account for this difference: (i) governments being on average more home biased than firms in purchases of both services and goods; and (ii) governments spending relatively more on sectors (goods and services) that are more home biased. We find that this difference is however shrinking over time: home bias in government procurement is declining faster than home bias in private markets. Our results have implications for trade negotiations that touch upon government procurement issues. A strong result of our analysis is that national borders are relatively thicker in services, which absorb the lion share of public spending. Furthermore, we find that policy initiatives targeted to government procurement have been able to reduce trade 33 barriers in services. These findings alone highlight potential complementarities between trade negotiations on services and government procurement. Serious and substantial service trade liberalization can have important liberalizing knock-on effects on government procurement procurement. 34 References Agnosteva, D.E., Anderson, J.E., Yotov, Y.V., 2019. Intra-national trade costs: Assaying regional frictions. European Economic Review 112, 32 – 50. Anderson, J.E., 1979. A Theoretical Foundation for the Gravity Equation. American Economic Review 69(1), 106–16. Anderson, J.E., 2011. The Gravity Model. Annual Review of Economics 3(1), 133–160. Anderson, J.E., Borchert, I., Mattoo, A., Yotov, Y.V., 2018. Dark costs, missing data: Shedding some light on services trade. European Economic Review 105, 193–214. Anderson, J.E., Milot, C.A., Yotov, Y.V., 2014. How Much Does Geography Deflect Services Trade? Canadian Answers. International Economic Review 55, 791–818. Anderson, J.E., van Wincoop, E., 2003. Gravity with Gravitas: A Solution to the Border Puzzle. American Economic Review 93(1), 170–192. Anderson, J.E., van Wincoop, E., 2004. Trade Costs. Journal of Economic Literature 42(3), 691–751. Anderson, J.E., Yotov, Y.V., 2010a. Specialization: Pro- and Anti-globalizing, 1990-2002. NBER Working Papers 16301. Anderson, J.E., Yotov, Y.V., 2010b. The Changing Incidence of Geography. American Economic Review 100(5), 2157–86. Anderson, R.D., Muller, A.C., Pelletier, P., 2017. Regional Trade Agreements and Procurement Rules. Facilitators or Hindrances?, in: Georgopulos, A.C., Hoekman, B., Mavroidis, P.C. (Eds.), The Internazionalization of Government Procurement Regulation. Armington, P.S., 1969. A theory of demand for products distinguished by place of production. Staff Papers - International Monetary Fund 16(1), pp. 159–178. Baier, S.L., Bergstrand, J.H., 2007. Do free trade agreements actually increase members’ international trade? Journal of International Economics 71(1), 72–95. Baldwin, R.E., 1970. Nontariff distortions of international trade. Brookings Institution, Washington DC. 35 Bergstrand, J.H., Larch, M., Yotov, Y.V., 2015. Economic integration agreements, border effects, and distance elasticities in the gravity equation. European Economic Review 78, 307 – 327. Beverelli, C., Keck, A., Larch, M., Yotov, Y.V., 2018. Institutions, Trade and Development: A Quantitative Analysis. CESifo Working Paper Series 6920. CESifo. Bosio, E., Djankov, S., Glaeser, E.L., Shleifer, A., 2020. Public Procurement in Law and Practice. Working Paper 27188. National Bureau of Economic Research. Brulhart, M., Trionfetti, F., 2004. Public expenditure, international specialisation and agglomeration. European Economic Review 48(4), 851–881. Chen, N., 2004. Intra-national versus international trade in the European Union: why do national borders matter? Journal of International Economics 63(1), 93–118. Cole, M.T., Davies, R.B., Kaplan, T., 2017. Protection in government procurement auctions. Journal of International Economics 106(C), 134–142. Costinot, A., Rodriguez-Clare, A., 2015. Chapter 4 - trade theory with numbers: Quantifying the consequences of globalization, Elsevier. volume 4 of Handbook of International Economics, pp. 197 – 261. Disdier, A.C., Head, K., 2008. The Puzzling Persistence of the Distance Effect on Bilateral Trade. The Review of Economics and Statistics 90(1), 37–48. Dür, A., Baccini, L., Elsig, M., 2014. The design of international trade agreements: Introducing a new dataset. The Review of International Organizations 9(3), 353–375. Eaton, J., Kortum, S., 2002. Technology, Geography, and Trade. Econometrica 70(5), 1741–1779. Egger, P., Larch, M., Staub, K.E., Winkelmann, R., 2011. The Trade Effects of Endogenous Preferential Trade Agreements. American Economic Journal: Economic Policy 3(3), 113–43. Egger, P.H., Larch, M., Yotov, Y.V., 2020. Gravity-Model Estimation with Time-Interval Data: Revisiting the Impact of Free Trade Agreements. CESifo Working Paper Series 8553. 36 Fajgelbaum, P., Khandelwal, A., 2016. Measuring the Unequal Gains from Trade. The Quarterly Journal of Economics 131(3), 1113–1180. Fally, T., 2015. Structural gravity and fixed effects. Journal of International Economics 97(1), 76 – 85. Fronk, J., 2014. Do Procurement Agreements Increase Partners’ Trade? Working Paper. Gourdon, J., Messent, J., 2019. How Government Procurement Measures Can Affect Trade. Journal of World Trade 53(5), 679–757. Head, K., Mayer, T., 2015. Chapter 3 - gravity equations: Workhorse,toolkit, and cookbook, in: Elhanan Helpman, K.R., Gopinath, G. (Eds.), Handbook of International Economics. Elsevier. volume 4, pp. 131 – 195. Heid, B., Larch, M., Yotov, Y., 2021. Estimating the Effects of Non-discriminatory Trade Policies within Structural Gravity Models. Canadian Journal of Economics Forthcoming. Herz, B., Varela-Irimia, X.L., 2020. Border effects in European public procurement. Journal of Economic Geography 20(6), 1359–1405. Kohl, T., Brakman, S., Garretsen, H., 2016. Do Trade Agreements Stimulate International Trade Differently? Evidence from 296 Trade Agreements. The World Economy 39(1), 97–131. Kutlina-Dimitrova, Z., Lakatos, C., 2014. Determinants of direct cross-border public procurement in EU member states. Review of World Economics 152. Larch, M., Lechthaler, W., 2013. Buy National or Buy International? The optimal design of government spending in an open economy. International Review of Economics & Finance 26, 87–108. Mattoo, A., Mulabdic, A., Ruta, M., 2017. Trade Creation and Trade Diversion in Deep Agreements. The World Bank. Mattoo, A., Rocha, N., Ruta, M., 2020. Handbook of Deep Trade Agreements. Number 34055 in World Bank Publications, The World Bank. Mayer, T., Zignago, S., 2011. Notes on CEPII’s distances measures: The GeoDist database. Working Papers 2011-25. CEPII research center. 37 McAfee, R.P., McMillan, J., 1989. Government procurement and international trade. Journal of International Economics 26(3-4), 291–308. McCallum, J., 1995. National Borders Matter: Canada-U.S. Regional Trade Patterns. American Economic Review 85(3), 615–23. Messerlin, P., Mirodout, S., 2012. EU public procurement markets: How open are they. Policy brief. Groupe d’Economie Mondiale. Miyagiwa, K., 1991. Oligopoly and Discriminatory Government Procurement Policy. The American Economic Review 81(5), pp. 1320–1328. Novy, D., 2013. Gravity Redux: Measuring International Trade Costs With Panel Data. Economic Inquiry 51(1), 101–121. OECD, 2013a. Government at glance 2013. Report - Directorate for Public Governance and Territorial Development, Organisation for Economic Cooperation and Development. OECD, 2013b. Measuring trade in value added: An OECD-WTO joint initiative. TiVA database. Organisation for Economic Cooperation and Development (OECD). Piermartini, R., Yotov, Y.V., 2016. Estimating trade policy effects with structural gravity. WTO Staff Working Papers ERSD-2016-10. World Trade Organization (WTO), Economic Research and Statistics Division. Rickard, S., Kono, D., 2014. Think globally, buy locally: International agreements and government procurement. The Review of International Organizations 9(3), 333–352. Riker, D., 2013. Government Consumption and Global Value Chains. Technical Report 2013-04A. Office of Economics Working Papers. U.S. International Trade Commission. Sellner, R., 2019. Non-discriminatory Trade Policies in Panel Structural Gravity Models: Evidence from Monte Carlo Simulations. Review of International Economics 27(3), 854–887. Silva, J.M.C.S., Tenreyro, S., 2006. The Log of Gravity. The Review of Economics and Statistics 88(4), 641–658. de Sousa, J., Mayer, T., Zignago, S., 2012. Market access in global and regional trade. Regional Science and Urban Economics 42(6), 1037–1052. 38 The European Commission, 2017. Measurement of impact of cross-border penetration in public procurement. Report. Timmer, M.P., 2012. The World Input-Output Database (WIOD): Contents, Sources and Methods. Technical Report. WIOD Working Paper Number 10. Trionfetti, F., 2000. Discriminatory Public Procurement and International Trade. The World Economy 23(1), 57–76. Trionfetti, F., 2001. Public Procurement, Market Integration, and Income Inequalities. Review of International Economics 9(1), 29–41. 39 Appendix Table A1: List of sectors. Aggregate sector Sector ISIC Rev 3 (1995 - 2004) ISIC Rev 4 (2005 - 2015) Agriculture, Forestry and Fishing 01, 02, 03, 04, 05 01, 02, 03 Mining 10, 11, 12 , 13, 14 05, 06, 07, 08, 09 Food and Beverages 15, 16 10, 11, 12 Textile and Apparel 17, 18, 19 13, 14, 15 Wood, paper products and printing 20, 21, 22 16, 17, 18 Goods Chemical products 23, 24, 25, 26 19, 20, 21, 22, 23 Metals and metal products 27, 28 24, 25 Machinery and equipment 29, 30, 31, 32, 33 26, 27, 28 Transport equipment 34, 35 29, 30 Other manufacturing 36, 37 31, 32, 33 Retail and hotel services 50, 51, 52, 55 45, 46, 47, 55, 56 60, 61, 62, 63, 64 Transport and telecommunication services 49, 50, 51, 52, 53, 58, 59, 60, 61 Finance and insurance 65, 66, 67 64, 65, 66 Services Real estate 70, 71 68 Public administration, health and education 75, 80, 85, 86, 87, 84, 85, 86, 87, 88, 90, 88 91, 92, 93, 94, 95, 96 Other services 72, 73, 74 62, 63, 69, 70, 71, 72, 73, 74, 75, 77, 78, 79, 80, 81, 82 Note: The “Sector” column reports the sectoral aggregations that are used in the sector-level regressions (see Tables A6 and A7). The third and fourth columns reported the corresponding two-digit chapters of the ISIC Rev. 3 and ISIC Rev. 4 industrial classifications. Figure A1: Services share of purchases in public and private markets. “Narrow” definition 1995 2015 SGP BRN TUR MYS KHM PHL TWN CHN HRV MLT ROU MEX PHL ARG BGR MAR ZAF JPN USA KHM MLT NOR LTU TUN KOR KOR SVK ROW CHN CAN CAN HUN LVA DNK VNM SVK ISR CZE GRC IRL EST FIN RUS RUS TUN CRI HUN HRV PER SVN CZE AUT NLD VNM POL ESP GBR CHE SVN POL IDN DEU AUS NZL MYS TWN MAR ITA ROW BGR JPN SGP FIN FRA SWE GRC BRN ISR CRI GBR FRA USA CHL BRA NOR COL COL SWE ESP EST BRA ROU DEU BEL AUT LTU IRL NLD CYP IND SAU PER ISL ISL MEX PRT ITA SAU DNK AUS CHE CHL PRT ZAF IND TUR LUX LVA HKG HKG BEL IDN NZL LUX ARG CYP 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 Public Private Note: Raw data are source from the TiVA database. Public procurement flows are computed using the “narrow” definition (see the main text for details). The vertical solid line indicates the average service share in public procurement across countries. The dashed line reports the same average for private markets. 40 Figure A2: Government import penetration/ Private import ratio by country. “Narrow” definition. 1995 2015 MAR VNM TUN MLT CRI ZAF CHN TUR BEL IND PER TWN CHL CHL HKG LVA COL BRA CYP MAR PRT LUX POL COL DNK CYP RUS MEX MYS NLD PHL BGR KOR HKG MLT SVN ZAF TUN ITA SVK NZL CZE LTU CRI JPN USA AUT ITA CHE KOR IDN EST ROW PER HUN PRT IND POL SVN ROU ARG HUN DEU MYS BRA CHN FRA AUS ISR RUS TWN LTU LVA KHM ESP ESP GBR PHL KHM ROW EST CAN MEX IDN NOR HRV IRL GRC BGR NZL BRN ISL SAU ISR CZE SWE FIN SGP USA DEU CAN GBR VNM FIN GRC JPN SWE BEL ROU CHE LUX AUT SGP FRA ISL BRN NLD NOR SVK DNK AUS ARG HRV IRL TUR SAU 0 1 2 3 0 1 2 3 Goods Services Note: Raw data are sourced from the TiVA database. Public procurement flows are computed using the “narrow” definition (see the main text for details). 41 Table A2: PPML Gravity estimates, Goods, “narrow” definition. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Same country 3.439*** 3.147*** (0.176) (0.187) Contiguity 0.239** 0.280** (0.121) (0.126) Common language 0.352*** 0.277*** (0.108) (0.101) Colony 0.290* 0.385 (0.159) (0.240) Ln Distance -0.651*** -0.812*** (0.049) (0.051) Common legal origin -0.002 0.041 (0.083) (0.082) TA -0.086 0.038 0.214*** 0.294*** 0.384** 0.318*** 0.385** 0.318*** 0.381** 0.314*** (0.120) (0.135) (0.042) (0.037) (0.152) (0.098) (0.152) (0.098) (0.152) (0.098) TA w/ procurement 0.168 0.269** 0.191*** 0.288*** 0.212*** 0.274*** 0.163*** 0.218*** (0.106) (0.121) (0.043) (0.038) (0.045) (0.041) (0.048) (0.039) WTO GPA 0.596*** 0.362*** -0.071 0.044 -0.171*** -0.088 (0.119) (0.124) (0.055) (0.054) (0.052) (0.063) EU 0.064 0.126 0.499*** 0.575*** (0.118) (0.105) (0.090) (0.069) Obs 23,064 23,064 23,064 23,064 23,064 23,064 23,064 23,064 23,064 23,064 Note: All regressions include importer-year and exporter-year fixed effects. Columns (3) to (10) include also country-pair fixed effects. In columns (3) and (4), the T A dummy equals one if the two countries in the pair belong to the same TA (with or without provisions on government procurement). Data are for the years 1995, 1999, 2003, 2007, 2011 and 2015. Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. Table A3: PPML Gravity estimates, Services, “narrow” definition. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Same country 7.341*** 5.805*** (0.173) (0.195) Contiguity 0.233** 0.390*** (0.103) (0.128) Common language 0.612*** 0.609*** (0.096) (0.102) Colony 0.706*** 0.890*** (0.182) (0.180) Ln Distance -0.420*** -0.382*** (0.051) (0.057) Common legal origin 0.065 0.068 (0.067) (0.080) TA 0.435*** 0.399*** 0.684*** 0.082** 0.482*** -0.208** 0.481*** -0.208** 0.478*** -0.210** (0.156) (0.144) (0.069) (0.040) (0.085) (0.099) (0.083) (0.099) (0.083) (0.099) TA w/ procurement 0.001 -0.078 0.710*** 0.142*** 0.465*** 0.091** 0.402*** 0.030 (0.106) (0.121) (0.079) (0.040) (0.067) (0.043) (0.070) (0.044) WTO GPA 0.972*** 0.738*** 0.560*** 0.127** 0.427*** 0.035 (0.101) (0.106) (0.062) (0.049) (0.058) (0.052) EU 0.170 0.103 0.751*** 0.335*** (0.117) (0.116) (0.096) (0.065) Obs 23,064 23,064 23,064 23,064 23,064 23,064 23,064 23,064 23,064 23,064 Note: All regressions include importer-year and exporter-year fixed effects. Columns (3) to (10) include also country-pair fixed effects. In columns (3) and (4), the T A dummy equals one if the two countries in the pair belong to the same TA (with or without provisions on government procurement). Data are for the years 1995, 1999, 2003, 2007, 2011 and 2015. Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. 42 Table A4: PPML Gravity estimates, Goods, full sample. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Same country 3.460*** 3.128*** (0.166) (0.187) Contiguity 0.238** 0.290** (0.120) (0.127) Common language 0.360*** 0.287*** (0.110) (0.101) Colony 0.323** 0.396* (0.161) (0.236) Ln Distance -0.646*** -0.819*** (0.048) (0.052) Common legal origin -0.003 0.038 (0.081) (0.081) TA -0.095 0.037 0.167*** 0.268*** 0.329** 0.275*** 0.330** 0.275*** 0.326** 0.272*** (0.118) (0.136) (0.042) (0.037) (0.130) (0.088) (0.130) (0.088) (0.130) (0.088) TA w/ procurement 0.163 0.227* 0.142*** 0.266*** 0.156*** 0.258*** 0.102** 0.190*** (0.107) (0.126) (0.043) (0.039) (0.047) (0.046) (0.049) (0.044) WTO GPA 0.602*** 0.378*** -0.051 0.024 -0.169*** -0.118* (0.116) (0.123) (0.062) (0.062) (0.058) (0.068) EU 0.079 0.083 0.467*** 0.565*** (0.118) (0.105) (0.089) (0.073) Obs 80,724 80,724 80,724 80,724 80,724 80,724 80,724 80,724 80,724 80,724 Note: All regressions include importer-year and exporter-year fixed effects. Columns (3) to (10) include also country-pair fixed effects. In columns (3) and (4), the T A dummy equals one if the two countries in the pair belong to the same TA (with or without provisions on government procurement). Data are for all years between 1995 and 2015. Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. Table A5: PPML Gravity estimates, Services, full sample. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Same country 6.977*** 5.781*** (0.181) (0.195) Contiguity 0.190* 0.390*** (0.112) (0.128) Common language 0.636*** 0.612*** (0.103) (0.102) Colony 0.676*** 0.888*** (0.180) (0.176) Ln Distance -0.455*** -0.391*** (0.054) (0.057) Common legal origin 0.077 0.066 (0.069) (0.080) TA 0.368** 0.426*** 0.479*** 0.069* 0.277*** -0.199** 0.275*** -0.199** 0.273*** -0.201** (0.165) (0.151) (0.049) (0.037) (0.074) (0.083) (0.074) (0.083) (0.074) (0.083) TA w/ procurement -0.116 -0.092 0.509*** 0.134*** 0.368*** 0.095** 0.317*** 0.039 (0.117) (0.123) (0.057) (0.038) (0.053) (0.043) (0.052) (0.045) WTO GPA 0.996*** 0.753*** 0.342*** 0.098* 0.233*** 0.006 (0.110) (0.105) (0.061) (0.051) (0.063) (0.055) EU 0.007 0.071 0.598*** 0.314*** (0.126) (0.114) (0.086) (0.065) Obs 80,724 80,724 80,724 80,724 80,724 80,724 80,724 80,724 80,724 80,724 Note: All regressions include importer-year and exporter-year fixed effects. Columns (3) to (10) include also country-pair fixed effects. In columns (3) and (4), the T A dummy equals one if the two countries in the pair belong to the same TA (with or without provisions on government procurement). Data are for all years between 1995 and 2015. Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. 43 Table A6: PPML Gravity estimates, Goods, sector-level trade data. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Same country 3.457*** 3.218*** (0.153) (0.178) Contiguity 0.138 0.228* (0.120) (0.125) Common language 0.412*** 0.306*** (0.106) (0.105) Colony 0.370** 0.427* (0.162) (0.222) Ln Distance -0.751*** -0.895*** (0.046) (0.051) Common legal origin 0.016 0.105 (0.069) (0.078) TA -0.091 0.052 0.147*** 0.265*** 0.263 0.270*** 0.265* 0.270*** 0.255 0.262*** (0.106) (0.126) (0.044) (0.037) (0.160) (0.092) (0.160) (0.092) (0.160) (0.092) TA w/ procurement 0.128 0.281** 0.132*** 0.264*** 0.178*** 0.251*** 0.105** 0.176*** (0.099) (0.122) (0.044) (0.039) (0.050) (0.043) (0.050) (0.038) WTO GPA 0.698*** 0.417*** -0.157** 0.035 -0.308*** -0.129* (0.116) (0.126) (0.074) (0.062) (0.082) (0.078) EU 0.090 0.211** 0.582*** 0.617*** (0.124) (0.096) (0.102) (0.082) Obs 230,330 230,454 230,330 230,454 230,330 230,454 230,330 230,454 230,330 230,454 Note: All regressions include importer-sector-year and exporter-sector-year fixed effects. Columns (3) to (10) include also country-pair fixed effects. In columns (3) and (4), the T A dummy equals one if the two countries in the pair belong to the same TA (with or without provisions on government procurement). Data are for the years 1995, 1999, 2003, 2007, 2011 and 2015. The dependent variable is at the sector level (Table A1 reports the list of sectors). Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. 44 Table A7: PPML Gravity estimates, Services, sector-level trade data. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Gov. Priv. Same country 7.056*** 5.802*** (0.173) (0.192) Contiguity 0.147 0.386*** (0.108) (0.127) Common language 0.639*** 0.584*** (0.097) (0.103) Colony 0.487*** 0.741*** (0.166) (0.167) Ln Distance -0.506*** -0.403*** (0.051) (0.056) Common legal origin 0.094 0.077 (0.066) (0.079) FTA w/out procurement 0.244* 0.357** 0.508*** 0.095** 0.276*** -0.199** 0.274*** -0.199** 0.270*** -0.202** (0.145) (0.142) (0.047) (0.038) (0.070) (0.080) (0.069) (0.080) (0.069) (0.080) FTA w/ procurement -0.126 -0.062 0.539*** 0.157*** 0.372*** 0.099** 0.320*** 0.042 (0.109) (0.120) (0.053) (0.040) (0.043) (0.041) (0.042) (0.042) WTO GPA 1.060*** 0.742*** 0.424*** 0.145*** 0.322*** 0.057 (0.107) (0.106) (0.050) (0.044) (0.049) (0.045) EU 0.020 0.119 0.611*** 0.334*** (0.123) (0.113) (0.076) (0.060) Obs 184,512 184,512 184,512 184,512 184,512 184,512 184,512 184,512 184,512 184,512 Note: All regressions include importer-secotr-year and exporter-sector-year fixed effects. Columns (3) to (10) include also country-pair fixed effects. In columns (3) and (4), the T A dummy equals one if the two countries in the pair belong to the same TA (with or without provisions on government procurement). Data are for the years 1995, 1999, 2003, 2007, 2011 and 2015. The dependent variable is at the sector level (Table A1 reports the list of sectors). Coefficients in bold are statistically different between the ‘Gov.’ and ‘Priv.’ regressions. Robust standard errors clustered at the country-pair level are in parentheses. Significant at: *10%, **5%, ***1% level. Figure A3: Trade effect of single provisions on government procurement – Goods Dispute settlement Limited tendering Tender documentation Offsets Domestic review Info to bidders Detailed coverage Info on planned procurement Technical specifications Exceptions from coverage IP protection Modification of coverage Goods and services Deadlines Treatment of tenders and awards Conditions of participation Conditions on previous awards Enforceable Qualification of suppliers Transparency Info to third parties Coverage entities Publication of info Phase in for developing cty Negotiations Expansion coverage E-procurement Reduction discrimination Electronic auctions Statistics 0 .1 .2 .3 .4 coefficient on gov. procurement provision Note: The bars represent estimated coefficient on a dummy for the presence of the indicated provision. The dependent variable of the gravity model is bilateral flows in government procurement. Controls include a dummy for TAs without government procurement, a dummy for TAs with provisions other than the one being ‘tested’, a dummy for membership in the WTO GPA, and an EU dummy. All regressions include exporter-year, importer-year and country-pair fixed effects. 45 Figure A4: Trade effect of single provisions on government procurement – Services Safety standards Conflict of interests Future accession Cooperation SME participation Reduction discrimination Rules of origin Dispute settlement National treatment Limited tendering Tender documentation Offsets Selective tendering Publication of info Enforceable Conditions on previous awards Domestic review Info to bidders Info on planned procurement Detailed coverage Technical specifications Exceptions from coverage IP protection Info to third parties Treatment of tenders and awards Conditions of participation Modification of coverage Goods and services Coverage entities Qualification of suppliers Deadlines Transparency Expansion coverage Phase in for developing cty E-procurement Negotiations Statistics Electronic auctions 0 .1 .2 .3 .4 .5 coefficient on gov. procurement provision Note: The bars represent estimated coefficient on a dummy for the presence of the indicated provision. The dependent variable of the gravity model is bilateral flows in government procurement. Controls include a dummy for TAs without government procurement, a dummy for TAs with provisions other than the one being ‘tested’, a dummy for membership in the WTO GPA, and an EU dummy. All regressions include exporter-year, importer-year and country-pair fixed effects. 46 Table A8: CHB Indexes – Goods 1995 1999 2003 2007 2011 2015 Avg. 15-95 % Cty Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv ARG 315 108 293 104 387 165 204 127 124 94 81 84 234 114 -74% -22% AUS 83 65 83 66 85 65 70 55 59 49 73 64 76 61 -12% -1% AUT 105 92 118 101 120 95 101 96 116 110 114 136 112 105 9% 48% BEL 86 57 92 63 84 62 63 66 72 79 79 104 79 72 -8% 82% BGR 1089 783 1151 1017 953 792 738 535 995 617 1044 734 995 746 -4% -6% BRA 44 32 66 44 75 44 52 32 31 24 43 37 52 35 -2% 15% BRN 5687 3954 8597 4233 10690 3063 2508 2026 2862 1752 1940 2540 5381 2928 -66% -36% CAN 26 32 24 27 19 26 20 27 24 31 23 34 23 29 -13% 6% CHE 37 65 43 73 42 73 30 79 31 82 29 93 35 78 -22% 43% CHL 622 215 594 258 693 258 434 175 422 172 372 201 523 213 -40% -7% CHN 23 12 13 9 11 7 8 6 6 4 4 3 11 7 -84% -77% COL 322 214 218 242 336 269 245 199 204 169 230 212 259 218 -28% -1% CRI 1932 1388 1363 1138 1660 1271 1430 1305 1268 1311 1077 1287 1455 1283 -44% -7% CYP 5553 3039 3936 3336 3102 3054 3184 2834 4429 3812 5999 6395 4367 3745 8% 110% CZE 243 198 295 189 235 147 167 110 172 123 193 152 218 153 -20% -23% DEU 11 10 13 11 14 11 11 11 13 14 14 17 13 12 24% 74% DNK 150 137 160 150 159 147 101 157 111 201 102 260 131 175 -32% 89% ESP 48 33 53 33 45 29 30 28 38 40 44 51 43 35 -8% 55% EST 2630 2686 2458 2426 2314 1880 1701 1261 1800 1445 1746 1667 2108 1894 -34% -38% FIN 138 135 137 137 141 134 95 131 161 167 158 230 138 156 14% 70% FRA 14 16 18 17 17 17 17 21 19 26 21 35 18 22 50% 121% GBR 15 20 14 19 15 20 16 23 19 33 19 35 16 25 25% 75% GRC 220 192 260 214 167 201 145 212 241 294 333 410 228 254 52% 113% HKG 799 403 761 546 1113 772 468 275 520 271 417 240 680 418 -48% -40% HRV 556 826 786 926 871 808 713 753 874 1033 914 1309 786 942 64% 58% HUN 278 242 277 224 220 176 159 138 206 168 209 203 225 192 -25% -16% IDN 152 64 153 76 133 70 185 58 156 39 162 44 157 58 6% -31% IND 116 41 83 37 89 33 65 24 50 21 49 21 75 30 -58% -48% IRL 89 142 71 114 61 103 50 128 59 185 46 163 63 139 -48% 15% ISL 2373 2704 2280 2407 2063 2378 1893 2239 2556 3127 2508 3445 2279 2717 6% 27% ISR 156 263 128 256 134 266 133 256 136 259 137 271 138 262 -12% 3% ITA 34 18 34 18 32 17 22 19 32 25 31 33 31 22 -10% 80% JPN 5 5 6 6 7 8 9 10 7 10 10 15 7 9 99% 185% KHM 13947 5332 12677 3470 6894 2761 7429 2805 6273 2283 4640 1799 8643 3075 -67% -66% KOR 43 27 44 29 38 26 28 23 29 22 26 23 35 25 -38% -14% LTU 1886 1909 1756 1642 1468 1183 1247 809 1223 844 1395 1052 1496 1240 -26% -45% LUX 1370 775 1494 817 1328 775 1227 823 1608 1102 2015 1410 1507 950 47% 82% LVA 2553 3026 2955 2930 2057 2163 2072 1403 2588 1793 2794 2095 2503 2235 9% -31% MAR 745 377 1522 388 1801 383 497 420 464 407 417 466 908 407 -44% 23% MEX 97 50 66 31 64 32 38 30 43 36 38 39 58 37 -60% -22% MLT 2212 3247 2497 3082 2726 3032 1750 3008 1916 3642 2101 4369 2200 3397 -5% 35% MYS 145 114 143 96 108 77 115 76 100 68 94 67 118 83 -35% -42% NLD 44 47 49 52 45 50 46 52 57 61 64 79 51 57 45% 68% NOR 195 132 204 126 194 115 89 104 87 118 107 161 146 126 -45% 22% NZL 621 320 551 360 513 317 349 302 355 329 375 373 461 333 -40% 17% PER 577 394 556 407 706 386 653 321 570 254 497 259 593 337 -14% -34% PHL 195 159 203 176 307 201 311 183 288 161 186 129 248 168 -5% -19% POL 143 113 205 112 188 101 96 70 108 76 106 92 141 94 -26% -18% PRT 227 146 234 150 220 148 198 167 252 218 310 285 240 186 36% 96% ROU 431 329 409 412 347 327 318 191 365 231 455 281 388 295 5% -15% RUS 63 60 116 94 75 58 35 33 34 32 41 39 61 53 -35% -35% SAU 210 130 183 121 154 104 121 82 106 72 90 93 144 100 -57% -29% SGP 110 127 118 135 116 134 135 130 121 122 120 124 120 129 9% -2% SVK 530 448 659 467 585 345 333 217 402 239 386 268 482 331 -27% -40% SVN 759 650 856 688 828 616 594 502 680 638 673 787 732 647 -11% 21% SWE 72 78 73 77 75 76 59 80 66 99 83 133 72 90 15% 71% TUN 1656 686 1431 626 1780 642 758 774 628 850 606 909 1143 748 -63% 33% TUR 107 81 103 85 137 64 174 60 194 64 160 65 146 70 49% -20% TWN 48 46 59 48 74 48 83 50 78 48 79 54 70 49 65% 17% USA 3 4 2 4 2 4 3 5 3 6 4 6 3 5 48% 31% VNM 748 394 653 295 558 229 374 169 267 123 172 83 462 215 -77% -79% ZAF 147 129 188 140 145 119 156 115 150 119 212 171 166 132 44% 33% Note: Constructed Home Bias indexes estimated from the specifications in columns (7) and (8), Table 3. 47 Table A9: CHB Indexes – Services 1995 1999 2003 2007 2011 2015 Avg. 15-95 % Cty Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv ARG 143 108 134 108 431 320 272 244 171 167 125 148 213 182 -12% 37% AUS 69 55 68 55 67 51 56 47 42 35 45 39 58 47 -35% -29% AUT 105 118 125 139 136 130 130 122 142 126 156 149 132 131 49% 26% BEL 82 74 94 84 92 83 89 82 95 84 104 102 92 85 27% 38% BGR 2618 1759 2273 2246 1636 1625 1216 893 1323 872 1429 996 1749 1399 -45% -43% BRA 27 46 40 59 57 78 34 42 25 29 33 40 36 49 21% -13% BRN 3780 8678 3833 14555 4158 15485 3549 11235 3932 9421 4881 14954 4022 12388 29% 72% CAN 36 50 36 47 35 43 31 37 29 37 35 43 34 43 -1% -14% CHE 112 69 142 86 139 83 135 82 112 68 108 71 125 77 -3% 2% CHL 689 367 574 390 712 422 486 323 371 247 350 255 531 334 -49% -31% CHN 57 59 29 34 27 26 22 22 13 14 8 8 26 27 -85% -87% COL 310 268 228 324 407 415 282 305 219 233 210 262 276 301 -32% -2% CRI 3192 2513 2710 2322 2675 2213 3174 2150 1853 1577 1327 1249 2488 2004 -58% -50% CYP 3808 2953 2937 3044 2488 2623 2102 1520 2086 1542 3503 1996 2821 2280 -8% -32% CZE 443 367 427 377 320 278 253 212 263 205 341 278 341 286 -23% -24% DEU 11 11 13 14 15 15 16 15 17 16 18 19 15 15 70% 67% DNK 110 146 117 163 119 155 124 138 135 151 158 177 127 155 44% 21% ESP 48 43 51 47 45 37 38 30 41 38 53 49 46 41 11% 14% EST 4935 6030 3963 3968 3472 2858 2325 1871 2536 2134 2454 2326 3281 3198 -50% -61% FIN 170 229 189 249 183 222 179 214 176 221 200 264 183 233 18% 16% FRA 14 18 16 21 16 19 17 18 19 20 22 24 17 20 55% 31% GBR 20 20 17 16 15 15 14 14 20 20 19 19 18 17 -5% -9% GRC 237 206 235 222 192 195 153 163 216 213 322 318 226 220 36% 55% HKG 393 126 281 121 330 141 380 156 411 167 319 170 352 147 -19% 34% HRV 882 1403 1021 1558 1102 1174 872 907 981 1076 1189 1389 1008 1251 35% -1% HUN 495 602 530 607 372 415 339 346 426 415 506 526 445 485 2% -13% IDN 283 164 457 243 325 186 234 161 158 105 148 108 267 161 -48% -35% IND 137 103 98 89 114 77 75 56 60 46 51 38 89 68 -63% -63% IRL 340 246 295 194 211 148 189 108 236 151 305 142 263 165 -10% -42% ISL 2804 4284 2249 3525 1937 3017 1600 2081 2840 3753 2495 3437 2321 3350 -11% -20% ISR 175 309 160 292 183 309 228 318 196 262 170 241 185 289 -3% -22% ITA 25 23 25 23 23 21 23 20 28 24 35 30 27 23 39% 30% JPN 6 5 7 7 8 8 12 11 10 10 14 14 10 9 131% 182% KHM 30961 9879 28426 10579 17122 9184 22783 9029 17959 7517 12991 5652 21707 8640 -58% -43% KOR 85 56 95 66 79 57 57 49 66 52 55 49 73 55 -36% -13% LTU 2991 3985 2092 3116 1867 2081 1339 1348 1480 1512 1564 1585 1889 2271 -48% -60% LUX 1317 376 1274 317 1147 291 917 204 932 208 913 188 1084 264 -31% -50% LVA 3833 4461 3028 3105 2616 2314 1677 1271 2145 1513 2235 1728 2589 2398 -42% -61% MAR 706 993 705 1161 739 1159 689 941 665 912 649 968 692 1022 -8% -3% MEX 160 97 97 68 84 60 98 61 98 66 95 69 105 70 -41% -29% MLT 6891 4935 6910 4513 6399 4334 4961 2279 4026 2000 3729 1961 5486 3337 -46% -60% MYS 397 271 509 315 396 279 385 266 318 213 317 214 387 260 -20% -21% NLD 50 57 56 59 50 55 45 55 49 59 59 71 52 59 16% 26% NOR 157 192 154 189 132 171 134 149 127 147 143 182 141 171 -9% -5% NZL 429 351 439 392 421 344 365 324 355 321 346 311 392 340 -19% -11% PER 916 575 872 648 955 652 820 639 591 445 448 406 767 561 -51% -29% PHL 506 411 478 375 670 432 682 399 557 316 393 253 548 364 -22% -39% POL 207 199 190 160 183 139 125 104 130 103 141 120 163 137 -32% -40% PRT 242 224 225 225 206 207 199 196 243 227 330 301 241 230 36% 35% ROU 1173 804 938 817 656 616 354 319 447 302 472 315 673 529 -60% -61% RUS 79 108 164 180 80 94 41 45 35 36 48 46 74 85 -39% -58% SAU 138 271 121 282 124 300 124 331 94 225 69 183 112 265 -50% -32% SGP 415 152 429 159 442 164 336 144 311 114 266 107 366 140 -36% -30% SVK 1218 1140 1284 1088 1026 816 753 581 670 491 720 567 945 781 -41% -50% SVN 1283 1274 1308 1321 1244 1205 1098 983 1119 1083 1461 1370 1252 1206 14% 8% SWE 71 105 73 110 77 110 80 105 84 105 90 123 79 110 26% 18% TUN 1705 2065 1486 1727 1564 1651 1124 1571 1292 1565 1309 1771 1413 1725 -23% -14% TUR 271 128 169 118 174 126 176 76 191 75 182 75 194 100 -33% -42% TWN 125 98 121 96 155 110 149 126 155 125 151 128 143 114 21% 31% USA 4 3 3 3 3 3 3 3 4 4 3 3 3 3 -10% -5% VNM 2840 1353 2600 1187 2632 1123 2007 826 1392 569 871 371 2057 905 -69% -73% ZAF 184 190 207 212 194 187 157 170 130 147 166 195 173 184 -9% 3% Note: Constructed Home Bias indexes estimated from the specifications in columns (7) and (8), Table 4. 48 Table A10: Aggregated CHB Indexes 1995 1999 2003 2007 2011 2015 Avg. 15-95 % Cty Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv Gov Priv ARG 143 108 134 107 431 235 270 179 169 125 124 111 212 144 -13% 3% AUS 70 57 69 57 67 53 56 48 42 37 45 41 58 49 -35% -28% AUT 104 110 125 130 136 122 129 117 142 123 156 147 132 125 50% 34% BEL 80 68 94 79 92 78 89 79 95 83 104 103 92 82 29% 50% BGR 2573 1150 2253 1575 1626 1218 1208 769 1320 791 1425 918 1734 1070 -45% -20% BRA 27 39 40 53 57 61 34 38 25 27 33 39 36 43 21% -1% BRN 3782 5980 3836 7142 4161 5231 3524 3136 3909 2717 4682 3602 3982 4635 24% -40% CAN 36 43 36 40 34 37 31 34 29 35 35 41 33 38 -1% -5% CHE 108 68 141 84 138 82 130 82 109 70 105 73 122 76 -3% 8% CHL 689 294 575 344 712 365 485 260 372 222 350 240 530 288 -49% -18% CHN 56 16 29 13 26 11 21 9 13 6 8 4 25 10 -86% -76% COL 310 248 228 295 407 359 282 262 219 207 210 245 276 269 -32% -1% CRI 3179 1930 2689 1698 2666 1820 3126 1831 1844 1502 1324 1256 2471 1673 -58% -35% CYP 3809 2977 2936 3108 2487 2705 2106 1620 2090 1661 3510 2131 2823 2367 -8% -28% CZE 440 289 425 290 319 221 252 173 262 176 339 229 340 230 -23% -21% DEU 11 11 13 13 15 14 16 14 17 15 18 18 15 14 71% 72% DNK 109 143 117 161 119 154 124 141 135 157 158 185 127 157 45% 30% ESP 48 39 51 42 45 35 38 29 41 39 53 49 46 39 11% 27% EST 4909 4180 3948 3432 3462 2555 2322 1720 2532 1932 2449 2144 3270 2660 -50% -49% FIN 168 183 189 199 183 190 176 183 176 205 200 257 182 203 19% 40% FRA 14 17 16 20 16 19 17 18 19 20 22 25 17 20 56% 44% GBR 20 20 17 16 15 16 14 15 20 21 19 20 18 18 -4% -3% GRC 235 201 236 221 192 196 153 169 216 225 322 333 226 224 37% 66% HKG 392 136 281 127 331 145 381 167 412 180 319 181 353 156 -18% 33% HRV 876 1082 1018 1248 1100 1036 869 860 979 1063 1185 1369 1004 1110 35% 26% HUN 492 398 526 383 370 295 335 254 422 298 498 366 441 332 1% -8% IDN 279 88 437 111 316 105 233 87 158 57 148 63 262 85 -47% -29% IND 136 56 98 52 114 48 75 35 60 29 51 28 89 41 -63% -49% IRL 321 190 291 159 209 134 183 111 230 156 279 146 252 149 -13% -23% ISL 2791 3595 2241 3120 1939 2874 1603 2109 2842 3609 2498 3449 2319 3126 -10% -4% ISR 173 292 159 283 183 299 226 302 195 262 170 247 184 281 -2% -16% ITA 25 21 25 21 23 20 23 20 28 24 35 31 27 23 39% 45% JPN 6 5 7 7 8 8 12 11 10 10 14 15 10 9 130% 183% KHM 30812 7136 28153 5025 16829 4074 21967 4252 17403 3315 12585 2629 21291 4405 -59% -63% KOR 84 38 93 44 78 39 56 34 64 33 54 34 72 37 -36% -9% LTU 2974 2820 2089 2417 1862 1701 1338 1155 1478 1205 1562 1393 1884 1782 -47% -51% LUX 1231 411 1284 347 1156 315 923 214 939 217 919 192 1075 283 -25% -53% LVA 3815 3866 3026 3062 2610 2280 1679 1290 2147 1553 2238 1779 2586 2305 -41% -54% MAR 706 527 706 571 740 588 685 632 660 585 643 659 690 594 -9% 25% MEX 160 74 97 48 84 48 95 46 96 51 92 55 104 54 -43% -25% MLT 6779 4058 6838 3869 6359 3806 4890 2350 3992 2082 3714 2026 5429 3032 -45% -50% MYS 390 173 496 151 381 122 370 122 304 106 302 104 374 130 -23% -40% NLD 50 53 57 57 50 54 45 55 49 59 59 72 52 59 18% 36% NOR 155 171 154 171 133 156 133 137 126 140 142 178 141 159 -8% 4% NZL 429 341 439 383 421 337 364 318 355 323 347 322 393 337 -19% -5% PER 909 494 867 544 952 535 817 471 591 348 449 339 764 455 -51% -31% PHL 491 219 468 250 660 293 668 267 548 223 384 178 536 238 -22% -19% POL 207 153 190 142 183 126 124 93 130 93 140 111 162 120 -32% -28% PRT 240 190 226 200 206 190 199 190 243 225 330 298 241 216 37% 57% ROU 1131 462 918 590 648 468 354 262 446 276 472 305 661 394 -58% -34% RUS 79 83 163 141 80 80 41 41 35 35 48 44 74 70 -39% -47% SAU 138 210 121 211 125 199 124 157 94 130 69 143 112 175 -50% -32% SGP 390 142 409 153 420 157 329 142 304 116 262 109 352 136 -33% -23% SVK 1205 775 1277 826 1022 609 744 422 667 402 716 451 938 581 -41% -42% SVN 1277 977 1303 1060 1240 969 1091 833 1114 960 1443 1205 1245 1001 13% 23% SWE 70 95 73 100 77 101 79 99 84 104 90 125 79 104 28% 31% TUN 1705 997 1486 984 1565 1034 1116 1179 1263 1209 1274 1303 1401 1118 -25% 31% TUR 262 104 168 106 174 93 176 71 191 71 182 72 192 86 -31% -31% TWN 121 67 120 72 153 76 147 81 153 76 150 87 141 76 23% 29% USA 4 4 3 3 3 3 3 4 4 4 3 4 3 3 -9% -3% VNM 2808 628 2556 475 2557 357 1858 261 1288 179 787 119 1975 337 -72% -81% ZAF 183 162 207 182 193 160 157 149 131 138 167 188 173 163 -9% 16% Note: Constructed Home Bias indexes estimated from the specifications in columns (7) and (8), Tables 3 and 4. Aggregated CHBs equal a weighted average of sector-level CHBs from Tables A2 and A3, where the weights equal the product of expenditure and sales (as shares). 49