Regionalism in Standards: Good or Bad for Trade? Maggie Xiaoyang Cheny University of Colorado at Boulder Aaditya Mattooz World Bank October 2004 Abstract Regional agreements on standards have been largely ignored by economists and unconditionally blessed by multilateral trade rules. We ...nd, theoretically and empir- ically, that such agreements increase trade between participating countries but not necessarily with the rest of the world. Adopting a common standard in a region, i.e. harmonization, boosts exports of excluded developed countries to the region but re- duces exports of excluded developing countries - possibly because developing country ...rms are hurt more by an increase in the stringency of standards and bene...t less from economies of scale in integrated markets. Mutual Recognition Agreements (MRAs) are more uniformly trade promoting unless they contain restrictive rules of origin, in which case intra-regional trade increases at the expense of trade with other, especially developing, countries. We propose a modi...cation of international trade rules to strike a better balance between the interests of integrating and excluded countries. Key words: standard, harmonization, MRA, rules of origin, scale economy JEL codes: F12, F13, F15 The authors have bene...ted from discussions with Paul Brenton, Carsten Fink, Kishore Gawande, Bernard Hoekman, Murat Iyigun, Jim Markusen, Keith Maskus, Robert McNown, Kamal Saggi, and participants of seminars at the World Bank, US International Trade Commission (ITC), and University of Colorado at Boulder. The views expressed here are those of the authors and should not be attributed to the World Bank. y Department of Economics, University of Colorado at Boulder, Boulder, CO 80309; Phone: (303) 859-1853; Email: xiaoyang.chen@colorado.edu. z Development Research Group, World Bank, 1818 H Street, NW, Washington, DC 20433; Phone: (202) 458-7611; Email: amattoo@worldbank.org. 1 1 Introduction In their recently launched trade talks, the European Union and the Association of South- east Asian Nations (ASEAN) agreed to focus not on tari¤s and quotas but on what Pascal Lamy, the EU trade commissioner, called "the real 21st century trade issues": harmo- nizing standards.1 This is the latest episode in a process of deep integration that is most advanced within the European Union but also underway in many other regions. Two fac- tors explain the shift in regional negotiating emphasis away from conventional barriers and toward standards. First, multilateral negotiations have achieved remarkable reduc- tions in tari¤s and quotas but done relatively little to reduce the trade restrictive impact of technical barriers. Second, while multilateral trade rules governing regional agreements on tari¤s seek at least in principle to balance the interests of integrating countries and the rights of excluded countries,2 the rules treat regional agreements on standards as always benign and worthy of encouragement.3 Are regional agreements on technical barriers indeed an unambiguous blessing for global trade? The voluminous research on regionalism with its almost exclusive focus on tari¤s and quotas provides no adequate answer. This paper is a ...rst step in the theoretical and empirical analysis of regional initiatives on technical barriers to trade. Using a simple analytical model, it asks: How must the conventional analysis of free trade areas and customs unions be modi...ed to apply to agreements on standards? And what elements of international agreements on standards are relevant to predicting their impact on patterns of trade? Then employing a specially constructed panel dataset that directly identi...es di¤erent types of policy initiatives in each manufacturing industry, the paper addresses two empirical questions: Do regional agreements on standards lead to signi...cant increase 1The two groups of countries have ageed to begin by harmonizing sanitary standards in the agricultural and ...sheries sector, as well as technical standards for electronics and wood-based industries. Pascal Lamy, the EU trade commissioner said that the choice was driven by the belief that "the real 21st century trade issues were standards and rules in areas such as safety, health or consumer protection", rather than tarri¤s or quotas ("EU and Asean to pave way for trade pact talks", Financial Times, 7 September 2004.) 2These rules are in Article XXIV of GATT 1994. 3Article 2.7 of the WTO's Agreement on Technical Barriers to Trade encourages members to "give positive consideration to accepting as equivalent technical regulations of other members, provided they are satis...ed that these regulations adequately ful...ll the objectives of their own regulations." This provision would seem to allow a country to selectively recognize standards of other countries, without violating the fundamental obligation not to discriminate between its trading partners. There is no mention of the rights of, or obligations vis-à-vis, countries that happen not to receive "positive consideration." 2 in trade between participating countries? And what happens to trade with those that are left out? Agreements on standards raise issues that are both politically and analytically chal- lenging. Unlike tari¤s, standards cannot be simply negotiated away because the original reason for their existence is not trade protection but the enhancement of welfare by rem- edying market failure - arising, for example, from invisible safety attributes of products, negative environmental externalities, or product incompatibility due to the producers' failure to coordinate. Agreements on standards must therefore secure the gains from inte- grated markets without unduly compromising the role of standards as remedies for market failure. Not only are the motives for standards ostensibly honorable, so in principle is their implementation: unlike tari¤s, the same standards are imposed on both foreign and do- mestic ...rms. However, in spite of the supposed symmetry of treatment, the impact on trade may turn out highly asymmetric because the costs of compliance are likely to di¤er across countries. There are in fact three main types of agreements dealing with technical barriers to trade. The simplest, and potentially most powerful is the mutual recognition of existing standards, whereby a country grants unrestricted access of its market to products that meet any participating country's standards. This was the approach taken in principle by the European Union, with the spur of the Cassis de Dijon judgement of the European Court of Justice. Mutual recognition agreements (MRAs) are, however, not likely to be an option if there is a signi...cant di¤erence in the initial standards of the countries, as became evident in the context of the European Union.4 In such cases, a certain degree of harmonization is a precondition for countries to al- low products of other countries to access their markets. The most important example of such harmonization is the New Approach of the European Union, which resulted in a set of directives from the European Commission setting out essential health and safety re- 4The central problem in the EU mutual recognition approach is the overarching exemption contained in Article 36 of the EC treaty. This provision preserves the member countries'rights to restrict or prohibit imports on grounds of health and safety and other policy objectives, as long as this is not "a means of arbitrary discrimination or a disguised restriction on trade". This provision substantially dilutes the e¤ects of implementing mutual recognition because it allows a country with stringent regulations not to recognize as equivalent the regulations of other countries with lower stringency. 3 quirements for most regulated products.5 Available evidence suggests that harmonization within the EU tended toward the high range of initial standards. Vogel (1995) points out that the role of the Union's richest and most powerful members, which imposed the most stringent standards, has been critical in setting the EU standards agenda; their political and economic importance has served to make EU standards progressively stricter.6 In many other cases, neither mutual recognition nor harmonization of substantive standards may be deemed feasible or desirable. Countries may nevertheless choose at least to mutually recognize each other's conformity assessment requirements, i.e. country A trusts country B to certify that the products made by country B conform to country A's standards. Examples of such initiatives are the intra-EU MRAs on some unharmonized industries and the EU's agreements with a number of other countries. A key element of these agreements is the rule of origin.7 The MRAs between the EU and USA and the EU and Canada specify that conformity assessment done in one of the MRA countries, in which products are manufactured or through which they are imported, is accepted throughout the entire agreement region. Other agreements, such as the MRAs the EU has concluded with Australia and New Zealand, impose restrictive rules of origin which require that third country products continue to meet the conformity assessment of each country in the region. As a prelude to our empirical investigation of the e¤ects of these agreements on trade patterns, a simple analytical model is used to generate certain testable hypotheses. The main analytical results can be explained by drawing a partial analogy between standards harmonization and mutual recognition, on the one hand, and a customs union and a free trade area, on the other. As in the case of a customs union, the economic impact of stan- 5In practice, most of the broad manufacturing product areas where technical regulations are important have now been harmonized, particularly product areas where the mutual recognition approach was seen to be failing. 6EC (1998) Single Market Review concludes that the harmonized standards in most reviewed industries have been set higher than initial levels in most member countries. The history of EU automobile emission, chemical, and packaging standards also demonstrates that these standards have frequently been harmonized at a level slightly lower than that preferred by the Union's most stringent states, including Germany, Denmark, and Netherlands, but higher than favored by less strict members such as Italy, UK, and Spain (Vogel, 1995). 7"Rules of origin" are de...ned by the WTO as "the criteria used to de...ne where a product was made. They are an essential part of trade rules because a number of policies (such as preferential trade agreements) discriminate between exporting countries." 4 dards harmonization depends on the level at which the harmonized standard is set. Unlike the case of a customs union, standards harmonization has a market integration e¤ect that creates scale economy bene...ts for the ...rms of not just participating but also third coun- tries. The impact on the ...rms of a speci...c country depends on how the costs of meeting the new harmonized level of the standard compare with the bene...ts from economies of scale in integrated markets. We ...nd that if ...rms from one set of countries ("developing") have a higher variable cost of meeting a standard and reap fewer scale economy bene...ts in integrated markets than ...rms from another set of countries ("developed"), then the former can su¤er a decline in exports to the integrated market when harmonization raises some destination countries'standards. As in the case of a free trade area, the economic impact of a standards MRA depends critically on the choice of rules of origin. For the participating countries, an MRA is in e¤ect a downward harmonization of standards since ...rms are now free to meet the least costly of the initial standards: trade is stimulated not only by market integration but also by the reduced stringency of the standard. The analytical implications for imports from third countries di¤er dramatically with rules of origin. If ...rms of third countries are denied the bene...ts of the MRA and must continue to meet the original standard in each market, then they will face unchanged absolute conditions but su¤er a decline in relative competitiveness - and hence a decline in exports to the region. In contrast, if the ...rms of non-participating countries are also entitled to access the entire region by conforming to the least costly standard, then they too reap bene...ts. In this case, it is shown that ...rms from the set of countries ("developing") where it is more costly to meet a standard may bene...t more, because there is a greater reduction in their compliance costs thanks to the decline in the stringency of the standard. In order to test the empirical validity of these propositions, we construct the ...rst dataset that directly identi...es policy initiatives of di¤erent types on standards for manu- facturing industries in 42 countries over the period of 1986-2001.8 These include all OECD countries and 14 developing countries that are the largest exporters of manufactured goods outside the OECD and account for over 80% of non-OECD manufactured exports. The 8In contrast, the existing Perinorm dataset describes standards and technical regulations, but does not allow an unambiguous measurement of harmonization or mutual recognition. 5 policy measures include each harmonization directive and MRA concluded between the countries in the set. We concord the policy measures, which often pertain to a speci...c attribute (e.g. safety) of a variety of products, with trade data at the SITC (revision 2) 3-digit industry level. We then estimate the signi...cance of the impacts of these measures on bilateral trade across countries and over time, controlling for other inuences. Our evidence broadly con...rms the conclusions drawn from the model. Regional har- monization signi...cantly increases intra-regional trade in a¤ected industries. Exports to the region of excluded developed countries also increase, but exports of excluded devel- oping countries decline. These asymmetric e¤ects may arise because developing country ...rms are hurt more by an increase in the stringency of standards and bene...t less from economies of scale in integrated markets. Mutual Recognition Agreements (MRAs) pro- mote trade both within the region and with the rest of the world. But when they contain restrictive rules of origin, then intra-regional trade increases at the expense of trade with other, especially developing, countries. To place our contribution in the context of the existing literature,9 the analytical section builds on the work of Baldwin (2000) and Ganslandt and Markusen (2001). In particular, Baldwin (2000) anticipated some of the results of this paper on MRAs, but assumed identical countries with identical ...xed costs of complying with standards. So the implications of harmonization and asymmetric e¤ects on excluded countries were beyond the scope of that model. As far as we know, only two previous studies have empirically explored the impact of shared standards on trade. Swann et al (1996) regressed British net exports, exports and imports over the period 1985-1991 on counts of voluntary na- tional ("idiosyncratic") and international standards recognized by the United Kingdom and Germany. They discovered that international standards to which Britain was a party had little impact on imports but a signi...cantly positive e¤ect on exports while British national standards tended to raise both imports and exports. Moenius (1999) regressed bilateral trade volumes in 4-digit SITC sectors on counts of shared standards in a sample of 14 countries over 16 years, and found a positive and highly signi...cant relationship. Our paper di¤ers from these empirical contributions in a number of aspects. First, instead 9See Maskus and Wilson (2001) for a review of the literature. 6 of relying on approximate measures of shared standards, we directly identify harmoniza- tion directives and mutual recognition initiatives in speci...c industries across countries, and also distinguish between the impacts of these two types of measures. Second, we examine not only the e¤ect on trade between participating countries, but also on trade with excluded countries. Finally, we allow for di¤ering impacts of harmonization across destination markets, depending on whether they previously had more or less stringent standards, and across source countries, depending on the level of development. The rest of this paper is organized as follows. In Section 2, we present the analytical model and identify the main implications for trade of di¤erent types of initiatives. We discuss the data in Section 3, and present the empirical evidence in Section 4. In Section 5, we examine the robustness of our estimates. Section 6 concludes the paper and draws out the implications for the design of international trade rules. 2 Model We construct a model that enables us to capture the essence of regional initiatives on standards while allowing a¤ected countries to be heterogeneous. Each country imposes a mandatory standard sj, which a ...rm must meet in order to sell its goods to the country's consumers.10 To keep the model fairly general, we do not specify a particular rationale for the standard. However, to motivate the analysis, it is convenient to think of a safety standard which pertains to a product attribute (e.g. inammability) that cannot be independently observed by consumers.11 But the assurance that a particular product meets a higher standard has a positive impact on consumer demand for the product and thus a ...rm's revenue.12 Compliance with the standard is assumed to a¤ect both the marginal and ...xed costs 10We take sj as given in this paper, even though the level of sj in each country could be treated as endogenously determined, based on factors such as preferences, market size, income, and technology. Allowing standards to be optimally set, prior to and after harmonization, would signi...cantly complicate the analysis without providing much additional insight into the issues addressed here. 11Most EU harmonization directives have set out essential requirements that are targeted at the safety of the product user and protecting the user's health (from infection, toxins, explosion, cuts or wounds from mechnical parts, electric shock, burning, falls, etc.). 12One exception in the EU New Approach is the "Packaging and Packaging Waste Directive". This directive aims at reducing a negative consumption externality, such as pollution, and thus does not cause shift in consumer demand. However, our ...ndings also apply to this type of standard. 7 of ...rms. We assume that the marginal cost of production is identical for all ...rms in a particular country i and proportional to the level of the standard in the destination market j, sj, i.e. ci(sj) cisj. Furthermore, a ...rm (denoted by a) in country i must incur a ...xed cost of production, denoted by Fi , to meet each distinct standard in the a destination markets to which it sells. There is a continuum of n potential ...rms in each country i with their ...xed cost, Fi , uniformly distributed between Fi and Fi + a F, i.e. Fia UNIF [Fi;Fi + F].13 Fi is assumed to be independent of the level of the standard a in the destination market, sj. First, we consider the ...rms' behavior. We assume that ...rms treat markets with di¤erent standards as segmented, i.e. a ...rm will not ...nd it worthwhile to supply multiple markets by complying with the most stringent standard.14 In contrast, ...rms treat markets with the same standard as a single market when making entry decisions and subsequently competing in quantities. Solving backwards, a representative ...rm (...rm a) that is located in country i and sells in all markets with the same standard at sj, chooses its output qij to solve the following pro...t-maximization problem: max qij X a = a (1) ij j XRij j sj;Pqzj;qij z Xcisjqij Fi ; j where j represents any market that sets the standard at sj, and Rij denotes the revenue that is a function of si. Note that ...rms only need to incur a single ...xed cost to serve all the markets with the standard set at sj. The ...rst-order condition is: @Rij sj;Pqzj;qij z cisj = 0; (2) @qij which yields the pro...t-maximizing reaction function, i.e. qij = qij (qzj) where z 6= i. By solving all reaction functions simultaneously, we ...nd qij for each i as a function of nzj (8z), 13Alternatively, we could have assumed that ...rms in each country i must incur identical ...xed costs. The assumption in the text simpli...es the analysis with little loss of generality. 14In other words, the di¤erence in total variable costs of complying with any two standards is larger than any saving in ...xed costs from meeting just the stricter one. Our assumption also holds when the issue is not vertical di¤erentiation of products on some quality dimension, but the incompatibility of products that conform to di¤erent standards (e.g. two-prong versus three-prong plugs). 8 the number of ...rms from country z selling in market j. Denote Vij = Rij sj;Pqzj;qij z cisjqij as the revenue less the variable cost realized in market j at the pro...t maximum. Suppose Nj number of markets set the same standard as country j.15 A ...rm would choose to enter these Nj markets simultaneously only if the nonnegative total pro...t con- dition holds: Pj a a ij = NjVij Fi > 0. Next consider importing country markets. The number of ...rms from country i that supply each of these Nj markets, denoted by nij, can be found by solving the following equation: NjVij Fi nij = n ; (3) F because, in each country i, there is a continuum of n ...rms with their ...xed cost uniformly distributed between Fi and Fi + F. Rewriting the above equation, we get nij as a reaction function of the number of rival ...rms from every other country z (z 6= i), i.e. nij = nij(nzj). Solving the reaction functions simultaneously for all j, we obtain the equilibrium number of ...rms in any market j from each source country, i.e. nij. Hence, the total imports of market j from country i is given by Qij = nijqij. The goal of our model is to analyze the impact on bilateral trade of a policy initiative that deals with standards. For this purpose, we ...rst characterize the impact on the imports of market j from country i by totally di¤erentiating Qij: @qij dQij = nijdqij + qijdnij + X nij dnzj : (4) @nzj z On the right hand side of the above equation, the ...rst argument shows the direct e¤ect on Qij of a change in the individual ...rm's output; the second argument represents the direct e¤ect on Qij of a change in the number of ...rms from country i; the third argument reects the indirect e¤ect of a change in the number of ...rms from each source country through its inuence on an individual ...rm's output. Let's take a closer look at the elements of equation (4). First of all, a change in the 15In this model, we assume that destinations markets only di¤er in the level of the standard. 9 level of the standard has a direct impact on an individual ...rm's output, indicated by: dqij = @qij=@sj dsj: (5) Furthermore, based on equation (3), both Nj and sj determine the number of ...rms from each source country supplying market j. Thus, at equilibrium, dnij can be de...ned as: @nij @nij 1 @Vzj dnij = X dnzj = Vzj dNj + Nj dsj ; (6) @nzj z X @nzj z zj @sj where ij Ni(@Vij=@nij) + F=n > 0. Note that in this paper we assume ci > maxf@Rij=@sj;¯cg, which indicates @Vij=@sj, @Qij=@sj < 0, i.e. ...rms would not voluntarily produce a quality that exceeds the level of the standard when the standard pertains to some unobserved attribute of a product.16 Taking into account equations (5) and (6), equation (4) can be rewritten as: 1 @nzj dQij = ijX" X ij @nmj z m Vmj + #ijVmj#dNj + (7) "Nij " j + ijX @nzj @Vmj + #ij @nmj @sj m @Vzj # @qij nij#dsj; @sj @sj where ij nij(@qij=@niz) and #ij qij(@nij=@niz). The impact on imports of country j from i of any policy initiative that deals with standards is made up of two elements: the e¤ect of market integration (dNj) and the e¤ect of change in the standard (dsj). The ...rst argument on the right-hand-side of equation (7) captures the aggregate impact of market integration on imports through a change in the number of supplying ...rms. When dNj > 0, there is an increase in the number of destination markets served by incurring a single ...xed cost, which promotes economies of scale and thus spurs market entry. The second argument reects the direct and indirect 16c is de...ned in Appendix B.1. This assumption is in alignment with Akerlof (1970), who ...nds that, in ¯ the presence of information asymmetry in qualities, all ...rms would produce the lowest quality possible at equilibrium. When a positive level of standard is adopted to partially correct the market failure, ...rms are forced to meet the standard. 10 e¤ects of a change in the level of the standard, respectively, on the number of ...rms and ...rms'output. Assessing the relative strengths of these two e¤ects helps us determine the overall impact on imports of any regional initiatives that deal with standards. We are particularly interested in exploring the possible asymmetric e¤ects of regional agreements on standards in a heterogeneous world. We assume that the world consists of two types of countries: type K and type L, which di¤er in terms of their ...rms'costs of complying with standards, i.e. ci and Fi. Two broad cases are possible: (i) cK < cL and FK > FL or (ii) cK < cL and FK 6 FL. Our analysis focuses on case (i) for several reasons. First of all, case (i) is more analytically challenging, and once the implications in this case are established, those in case (ii) can be worked out quite straightforwardly. More importantly, case (i) may more accurately depict situations where ...rms from both rich and poor countries produce the same product. It is likely that ...rms from richer capital-abundant countries (type K) tend to adopt a more capital-intensive technology that requires a relatively large Fi, while ...rms from poorer labor-abundant countries (type L) tend to adopt a more labor-intensive technology that requires a relatively large ci. In fact, the World Bank's (2004) Technical Barriers to Trade Survey reveals that a ...rm's percentage of compliance cost spent on product re-design for each export market (cor- responding to Fi in our paper) is positively correlated with the GDP per capita of the country in which it is located, whereas that spent on hiring additional labor for production and testing (corresponding to ci in our paper) is negatively correlated with the GDP per capita of the country. This limited evidence does not, however, provide the basis for an a priori assertion that type K countries correspond to developed countries and type L to developing countries. Rather, we treat this correspondence as a hypothesis to be tested in the empirical section. In the rest of this section, we examine the impact on both intra-regional trade and trade with excluded countries of two types of regional initiatives: upward harmonization and downward harmonization (with or without rules of origin). 11 2.1 Upward harmonization We begin by investigating the impact on trade when a subset of type K countries, denoted by H fhg, harmonize their standards at the level of the initially most stringent stan- dards, maxfsj : j 2 Hg. The objective of such an approach, reecting roughly the EU's directives in the 1990s, would be to integrate markets without compromising the role of the standard as a remedy for market failures in any country. First, we look at the set of countries within H for which dsj = 0; i.e. the countries in the harmonizing region with the most stringent initial standards. The impact on ...rms selling in these markets is solely due to the market integration e¤ect (i.e. the increase in Nj). All exporters to these markets reap economies of scale since they now serve the entire region incurring one single ...xed cost. But such market integration bene...ts are greater for ...rms of type K countries, since Fk > Fl and they enjoy a larger reduction in the per-market ...xed cost. So we shall certainly see increased entry of ...rms and imports from type K countries. However, for ...rms of type L countries, the direct bene...ts from increased economies of scale are counteracted by the more intense competition from ...rms of type K countries. Nevertheless, imports from type L countries may still increase if the former e¤ect is not completely o¤set by the latter. Next we look at the set of countries within H for which dsj > 0; i.e. the countries with initial standards less stringent than the eventual harmonized level. Intuitively, the bene...t to ...rms of increased scale economies may be partially or even completely eroded by the increased cost of complying with stricter standards. Provided the initial distance between standards, and hence the upward adjustment of standards, is below a certain threshold value, the imports of harmonizing markets from type K countries will increase. However, imports from type L countries are more likely to decrease because the direct bene...ts of economies of scale are now counteracted by two factors: type L ...rms realize smaller economies of scale (because they have lower ...xed costs than type K ...rms ), and are hurt more by the increase in the level of the standard (because they have higher variable costs of meeting a standard). Whether the imports from type L countries actually decline depends on the extent of their cost disadvantage. The overall impacts on imports of harmonizing countries are summarized below: 12 Proposition 1 When region H harmonizes standards at maxfsj : j 2 Hg, (i) imports of harmonizing markets with dsj = 0 from type K countries increase, whereas those from type L countries increase if cl < g(ck); (ii) imports of harmonizing markets with dsj > 0 from type K countries increase if dsj < s, whereas those from type L countries decrease if cl > g(ck). ¯ Proof. See Appendix B.1. Moreover, we compare the impacts of upward harmonization across destination markets and ...nd: Proposition 2 The increase in imports of harmonizing market j, i.e. dQij, from any exporter i is a strictly decreasing function of dsj. Proof. See Appendix B.2. The intuition is obvious: the higher dsj, the more the scale economy boost to imports is diluted by the higher costs of compliance with standards. 2.2 Downward harmonization (mutual recognition) Consider now the impact on trade when the countries of region H decide to mutually recognize (MR) one another's standards. In other words, products that comply with a standard set by any participating country can be freely sold in the entire region. It is straightforward to establish that such mutual recognition is equivalent in e¤ect to the downward harmonizing of standards at the level of minfsj : j 2 Hg, since ...rms are free to comply with the least strict standard in the region. The impact on trade with third countries turns out to depend on whether the bene...ts of MR are extended to third country ...rms. When the bene...ts of mutual recognition are restricted to ...rms within region H, the markets of individual countries in this region remain segmented, with the same initial standards, to ...rms outside the region. Although the absolute conditions of access remain unchanged, ...rms of excluded countries face a decline in relative competitiveness because ...rms of participating countries not only realize 13 greater economies of scale but also may enjoy reduced stringency of standards. The implications of such mutual recognition di¤er sharply for ...rms within the region from those in excluded countries, as summarized below: Proposition 3 When region H harmonizes standards at minfsj : j 2 Hg and imposes restrictive rules of origin, then (i) intra-regional trade increases at the expense of imports from the rest of world; (ii) in the rest of the world, type L countries see a larger decline in their exports to region H if cl > g1(ck). Proof. See Appendix B.3. In contrast, when the bene...ts of mutual recognition are open to all ...rms, the results in certain respects mirror those presented in Proposition 1. As in the case of upward harmonization, all ...rms reap the bene...ts of increased economies of scale due to market integration. But there is now an additional bene...t arising from the downward adjustment of the standard in certain countries. Although both of these e¤ects may encourage entry and boost total trade, they are asymmetrically distributed between exporter countries. Type K ...rms with larger ...xed costs tend to bene...t more from scale economies whereas type L ...rms with a higher variable cost of compliance derive greater bene...ts from reduced stringency of the standard. When there is a signi...cant downward adjustment of the standard, the rise in exports of countries of type L may be greater than that of type K. Proposition 4 If region H harmonizes standards at minfsj : j 2 Hg and does not impose restrictive rules of origin, then (i) imports of region H from type K countries ­both within the region and in the rest of the world ­increase, whereas those from type L countries increase if cl < g(ck); (ii) the increase in the total imports of region H from excluded type L countries is greater than that of excluded type K countries if cl > g2(ck) and PH dsj < s. ¯ Proof. See Appendix B.4. 14 3 Data We employ a balanced dataset, from COMTRADE, covering the trade of 42 countries at the SITC rev. 2 3-digit level of manufacturing industries from 1986 to 2001. The sample consists of 28 OECD countries and 14 developing countries that are the largest manufacturing exporters outside the OECD (and have complete sectoral import data).17 Inevitably, there exist some null observations since not all countries trade in each industry every year, and so it is appropriate to use the Tobit model for estimation. We identify the industries, countries and the time period that are a¤ected by each directive or MRA drawing upon the relevant o¢ cial documents, and then construct the variables displayed in Table 1. The harmonization directives and MRAs are not directly related to speci...c products but to product attributes.18 Thus a single industry (e.g. elec- tronic products) may be a¤ected by multiple directives (e.g. those pertaining, respectively, to low voltage equipment and electromagnetic compatibility). Di¤erent approaches can be taken to quantify the harmonization measures, depending on how these measures are believed to a¤ect trade. It is simplest, and seems reasonable, to assume that the impact is linearly related to the number of directives in each industry, i.e. each additional directive in any industry has an identical incremental a¤ect on trade. Alternative assumptions are possible, e.g. that additional directives have a diminishing e¤ect on trade, or that the nature of the impact di¤ers in each industry. We established that the results are not sensitive to the choice of approach. HARijrt counts the number of directives applicable to industry r between exporter i and importer j in year t. HAR_Mijrt counts the number of directives applicable to in- dustry r between importer j and any country other than exporter i in year t. For instance, HAR_Mijrt = 1 for imports of United Kingdom from China in the sector of machinery tools since 1995, because the United Kingdom has implemented a single directive applica- ble to machinery tools along with other EU members at the end of 1994. HAR_Eijrt measures the number of directives implemented in industry r between exporter i and any country other than importer j in year t. 17Czech Republic and Slovak Republic are excluded because of the lack of sectoral trade data in 1993 while Belgium and Luxembourg are considered as one unit throughout the period. 18The lists of EU directives and MRAs are provided Table A.1 and Table A.2 in Appendix A. 15 Table 1: Notations in estimation Regressand ln(importijrt) the natural logarithm of the imports of country j from country i in industry r and year t Fixed e¤ects irt exporter-industry-year jrt importer-industry-year ijr exporter-importer-industry Regressors HARijrt the number of harmonization directives between i and j in industry r and year t HAR_Mijrt the number of harmonization directives between j and any country other than i in industry r and year t HAR_Eijrt the number of harmonization directives between i and any country other than j in industry r and year t MRA_ROijrt 1 if an MRA with rules of origin exists between i and j in industry r and year t, and 0 otherwise MRA_RO_Mijrt 1 if an MRA with rules of origin exists between j and any country other than i in industry r and year t, and 0 otherwise MRA_RO_Eijrt 1 if an MRA with rules of origin exists between i and any country other than j in industry r and year t, and 0 otherwise MRA_NROijrt 1 if an MRA without rules of origin exists between i and j in industry r and year t, and 0 otherwise MRA_NRO_Mijrt 1 if an MRA without rules of origin exists between j and any country other than i in industry r and year t, and 0 otherwise MRA_NRO_Eijrt 1 if an MRA without rules of origin exists between i and any country other than i in industry r and year t, and 0 otherwise RTAijt 1 if an RTA exists between i and j, and 0 otherwise RTA_Mijt 1 if an RTA exists between j and any country other than i, and 0 otherwise RTA_Eijt 1 if an RTA exists between i and any country other than j, and 0 otherwise 16 MRA_ROijrt and MRA_NROijrt are dummy variables that reect the existence of an MRA, respectively, with or without the rules of origin between exporter i and importer j in industry r in year t. The cases where importer j reaches an MRA with or without the rules of origin with any country other than exporter i are respectively represented by MRA_RO_Mijrt and MRA_NRO_Mijrt. The same de...nitions apply to MRA_RO_Eijrt and MRA_NRO_Eijrt except that the party involved in an MRA is the exporter. The rest of the regressors are dummy variables constructed in a similar fashion to capture the existence of Regional Trade Agreements (RTAs).19 4 Empirical evidence In this section, we examine empirically the impact of the di¤erent approaches that deal with technical barriers to trade. The main equation we estimate takes the following form: 8 > > +1HARijrt++ > + > irt jrt ijr > > > > ln(importijrt) = > + > > 2HAR_Mijrt+ 3HAR_Eijrt > > > < + 4 MRA_ROijrt+ 5MRA_RO_Mijrt+ 6MRA_RO_Eijrt > > > + > > 7 MRA_NROijrt+ 8 MRA_NRO_Mijrt+ 9MRA_NRO_Eijrt > > > > > > > + RTAijt+ RTA_Mijt+ RTA_Eijt > 10 11 12 > : +"ijrt which can be denoted as y = X + " with y representing ln(importijrt), X the vector of explanatory variables, the vector of coe¢ cients and " the error term. Three types of ...xed e¤ects are included: the nested exporter-industry-year ...xed e¤ect captures factors such as sectoral output in the exporting country at a particular time, the nested importer-industry-year ...xed e¤ect controls for factors such as sectoral demand and domestic competition in the importing country at a particular time, while the nested importer-exporter-industry ...xed e¤ect includes any time-invariant bilateral sectoral ele- 19We include RTAs e¤ective in our sample countries over the period of 1986-2001, including the EC, EFTA, CEFTA, NAFTA, and the CER (the Closer Economic Relations agreement between Australia and New Zealand). 17 ments such as distance. The use of these extensive ...xed e¤ects enables us to isolate the role of agreements on technical regulations in explaining the changes in the pattern of trade over time. 4.1 The estimated e¤ects of harmonization and MRAs Table 2 reports the estimation results using the Tobit model. Coe¢ cients on all variables are statistically signi...cant at the 1% level and exhibit the signs predicted in Section 2. Column I reveals that the harmonization directives unambiguously stimulate intra- regional trade, as well as trade with excluded countries. In fact, the magnitudes of the estimated impacts are quite large. It would appear that a directive implemented in an industry between two countries on average raises their imports from each other by 32% (e0:2749 = 1:32), and imports from a country outside the harmonizing region by nearly 10% (e0:0950 = 1:096). The boost to trade is attributable to the positive impact of increased scale economies which outweighs, on average, the possible negative e¤ect on trade of increased stringency in some countries'standards. The impact of an MRA, as predicted in Section 2, turns out to be sensitive to whether it includes restrictive rules of origin. MRAs with rules of origin provide a powerful stim- ulus to intra-regional trade but at the expense of imports from countries outside the region.20 The negative coe¢ cient on MRA_RO_M, -0.4768, implies that imports from an excluded country su¤er a 38% decline in a¤ected industries. However, when an MRA does not include restrictive rules of origin, imports from both member countries and third countries increase, indicated by the positive coe¢ cients on both MRA_NRO and MRA_NRO_M. Finally, regional trade agreements are found on average to increase trade not only between participating countries but also with the rest of the world.21 20 The substantial magnitude of the coe¢ cient on MRA_RO may be explained by the trivial amount of initial trade in the a¤ected industries before the implementation of MRAs. 21 The literature on the impact of RTA on extra-bloc trade is mixed. Frankel, Stein, and Wei (1997) examine intra- and extra-bloc trade for eight RTAs-EC, EFTA, NAFTA, MERCOSUR, and the Andean Pact, AFTA, the CER, and the East Asian Economic Caucus for the period 1970-1992. Their study suggests that RTAs had positive impacts on all trade ows except for the EC's intra-bloc trade and EFTA's and NAFTA's trade with nonmembers. The authors conclude that there is little trade diversion. Soloaga and Winters (2001) ...nd that RTAs between developing countries increase extra-bloc trade. In particular, ASEAN since 1980 increases intra-bloc trade, extra-bloc imports, and extra-bloc exports. Our estimates suggest that the average e¤ect of these agreements is positive. 18 Table 2: Estimated e¤ects of harmonization and MRAs Regressand: ln(import) I II Harmonization on intra-regional trade 0.2749*** 0.2562*** (HAR) (0.010) (0.011) importers with stricter initial standards 0.0989*** (HAR strictness) (0.022) Harmonization on imports from the ROW 0.0950*** 0.0310*** (HAR_M) (0.005) (0.005) importers with stricter initial standards 0.2477*** (HAR_M strictness) (0.011) Harmonization on exports to the ROW 0.6438*** 0.6393*** (HAR_E) (0.005) (0.005) MRAs with rules of origin on intra-regional trade 2.3540*** 2.3589*** (MRA_RO) (0.031) (0.031) MRAs with rules of origin on imports from the ROW -0.4768*** -0.4799*** (MRA_RO_M) (0.037) (0.037) MRAs with rules of origin on exports to the ROW 0.3956*** 0.3963*** (MRA_RO_E) (0.036) (0.036) MRAs without rules of origin on intra-regional trade 0.6362*** 0.6390*** (MRA_NRO) (0.074) (0.074) MRAs without rules of origin on imports from the ROW 0.7794*** 0.7950*** (MRA_NRO_M) (0.037) (0.037) MRAs without rules of origin on exports to the ROW 1.6235*** 1.6154*** (MRA_NRO_E) (0.037) (0.037) RTA on intra-regional trade 1.7225*** 1.7266*** (RTA) (0.011) (0.011) RTA on imports from the ROW 0.0458*** 0.0404*** (RTA_M) (0.005) (0.005) RTA on exports to the ROW 0.0309** 0.0292*** (RTA_E) (0.005) (0.005) Number of observations 4160352 4160352 Log likelihood -7840111 -7839841 Standard errors are reported in parentheses. Exporter/Importer-industry-year, pair-industry ...xed e¤ects are controlled. ***, **, and * represent 1%, 5%, and 10% signi...cance levels, respectively. 19 In Section 2, Proposition 2 predicted that trade stimulus is negatively correlated with the extent by which the importing country raises its standard - because an increase in the stringency of the standard may partially or completely o¤set the bene...t from market integration and dampens imports. As noted in the introduction, the European Union's richest members generally imposed the most stringent standards, and used their inuence to ensure that the EU's harmonized standards were set close to their own levels. The avail- able evidence suggests that the core set of countries with relatively strict initial standards consists of Germany, Denmark, and the Netherlands (Vogel, 1995). We also considered alternative de...nitions, e.g. the top third of EC and EFTA countries ranked by GDP per capita in 2001, and the results were similar. An additional interactive term, strictness, is generated to test Proposition 2. For the harmonizing countries which can be presumed to have stricter initial standards, strictness is equal to 1; 0 for the rest of harmonizing importers. In column II of Table 2, the signi...cant coe¢ cient on the variable "HAR strictness" con...rms that the increase in imports of these two groups of countries di¤ers signi...cantly in magnitude. Intra-regional imports of countries that were likely to have raised their standards during harmonization increase by 29%, considerably less than the 43% increase for countries with initially stricter standards. This sharp di¤erence is also evident in imports from outside the harmonizing area. Imports of countries with less strict initial standards from a third country increase by 3%, while imports of countries with stricter initial standards increase by 32%.22 The rest of the coe¢ cients are similar to those in column I. 4.2 The divergent impact on imports from third countries Do regional agreements in standards have a similar impact on exporters everywhere in the rest of the world? As discussed in Section 2, ...rms that incur a larger ...xed cost of complying with country-speci...c standards reap larger bene...ts from scale economies, while ...rms that incur a higher variable cost of compliance will su¤er (bene...t) more from any increase (decline) in the stringency of the standard. Any asymmetry in costs can thus be expected 22As we gradually expand the set of countries deemed to have stricter initial standards (on the basis of GDP per capita), the magnitude of the coe¢ cients on the interacting terms becomes smaller. This tendency indicates that the three countries considered here are appropriate elements of the subset with the most stringent standards. 20 to lead to quite di¤erent e¤ects on exporters in di¤erent origins. As we suggested in Section 2, the distinction between countries according the level of development may correspond to the analytical distinction we make between countries on the basis of their costs of meeting standards. We generate a dummy variable, i.e. developing, which is equal to 1 for a developing country exporter and 0 for a developed country exporter. We consider all OECD countries as developed and the rest of the sample countries as developing. This dummy variable is used to interact with the three variables which capture the impact on imports from third countries: HAR_M, MRA_RO_M, and MRA_NRO_M. The regression results are reported in column I of Table 3. While regional harmonization promotes imports from developed third countries by 34%, there is actually a 16% reduction in imports from the excluded developing world (the coe¢ cient on HAR_M for developing country exporters is -0.1675 and e 0:1675= 0:84). These results are consistent with the assumption, which drives the predictions in Proposition 1 in Section 2, that developing countries tend to ...nd it more costly to meet stricter standards, and reap smaller bene...ts from economies of scale generated by integrated markets (i.e. ck < cl and Fk > Fl). The impact of MRAs on developing and developed countries turns out to depend on whether they include rules of origin. If they do, they hurt developing country exports (39% decline) more than they hurt developed country exports (28% decline). But MRAs without rules of origin enhance developing country exports even more than they enhance developed country exports, revealed by the positive coe¢ cient on the interacting term MRA_NRO_M developing. A possible explanation is o¤ered by the intuition behind Proposition 4 in Section 2: the greater reduction in compliance costs for developing country ...rms may o¤set the smaller bene...ts they derive from economies of scale. Finally, we directly test for the hypotheses in Proposition 1 by comparing respectively the four cases: the importer had more/less stringent initial standards while the exporter is a developed/developing country. Column II of Table 3 shows that with harmoniza- tion, exports to countries with stricter initial standards of both developed and developing countries increase, but the increase in the former (57%) is much greater than the increase in the latter (5%). Developed third countries'exports to countries that raise their stan- 21 Table 3: The divergent impact on imports from third countries Regressand: ln(import) I II Harmonization on intra-regional trade 0.3002*** 0.2825*** (HAR) (0.010) (0.011) importers with stricter initial standards 0.0951*** (HAR strictness) (0.023) Harmonization on imports from the ROW 0.2912*** (HAR_M) (0.007) from developing countries -0.4587*** (HAR_M developing) (0.012) from developed countries to importers with stricter initial standards 0.4491*** (HAR_M with developing = 0 & strictness = 1) (0.014) from developing countries to importers with stricter initial standards 0.0458*** (HAR_M with developing = 1 & strictness = 1) (0.017) from developed countries to importers with less strict initial standards 0.2369*** (HAR_M with developing = 0 & strictness = 0) (0.008) from developing countries to importers with less strict initial standards -0.2421*** (HAR_M with developing = 1 & strictness = 0) (0.010) Harmonization on exports to the ROW 0.6357*** 0.6315*** (HAR_E) (0.005) (0.005) MRAs with rules of origin on intra-regional trade 2.3185*** 2.3233*** (MRA_RO) (0.031) (0.031) MRAs with rules of origin on imports from the ROW -0.3294*** -0.3334*** (MRA_RO_M) (0.041) (0.041) from developing countries -0.1792*** -0.1763*** (MRA_RO_M Developing) (0.044) (0.044) MRAs with rules of origin on exports to the ROW 0.3204*** 0.3213*** (MRA_RO_E) (0.036) (0.036) MRAs without rules of origin on intra-regional trade 0.6612*** 0.6641*** (MRA_NRO) (0.075) (0.075) MRAs without rules of origin on imports from the ROW 0.8237*** 0.8379*** (MRA_NRO_M) (0.040) (0.040) from developing countries 0.2267*** 0.2296*** (MRA_NRO_M Developing) (0.041) (0.041) MRAs without rules of origin on exports to the ROW 1.5031*** 1.4952*** (MRA_NRO_E) (0.037) (0.037) RTA on intra-regional trade 1.7331*** 1.7371*** RTA on imports from the ROW 0.0347*** 0.0295*** RTA on exports to the ROW 0.004 0.0027 22 dards also rise (by 27%), but developing countries'exports decline by 22%. These results are consistent with the predictions in Proposition 1, and the assumption that developing countries bene...t less from economies of scale, and thus see a smaller increase in exports to the market that does not increase the stringency of its standard and are hurt more by an increase in the stringency of the standards in other markets to which their exports decline. These ...ndings suggest that harmonization of standards is generally advantageous to participating and excluded developed countries that have similar cost structures, but could hurt the exports of developing countries. In the case of mutual recognition agreements, excluded developed and developing countries have greater commonness of cause: absent rules of origin both gain, with rules of origin both lose, with a larger impact on developing countries in each case. 5 Robustness analysis Our econometric results have been obtained with a range of controls designed to eliminate any correlation between the endogenous variables and the error term. However, we cannot rule out econometric problems arising for two reasons: omission of unobserved e¤ects and endogeneity of regressors. First of all, initiatives on standards may not be the only measures that have drawn the countries of the European Union to trade closer together. For example, it could be that liberalization of transport inside the EU has reduced the costs of transport inside the Union faster than the costs outside the Union. Secondly, it may be that the initiatives on standards have been taken in precisely those industries in which trade between members was growing, so the initiatives are at least in part the results rather than the cause of trade growth. In this section we address these concerns and test the robustness of our results. 5.1 Unobserved e¤ects We ...rst consider the possible omission of unobserved e¤ects, which are not already embod- ied in the multiple nested ...xed e¤ects included in Section 4. Such e¤ects must therefore 23 consist of time-variant bilateral factors such as the preferential political or economic re- lations between two countries that might be correlated with the explanatory variables of interest. Following the approach suggested in Baltagi (2001) and originally due to Mundlak (1978) on individual e¤ects, we attempt to test and capture this time-variant bilateral e¤ect, denoted by , by assuming is a linear function of the averages of all the ijt ijt existing explanatory variables (measures of regional initiatives) across industries: (8) ijt = Xijt: + vijt; where Xijt: is a 1 R vector of observations on the explanatory variables averaged across industries. This e¤ect is uncorrelated with the explanatory variables if and only if = 0. As Mundlak (1978) assumed, without loss of generality, the X are deviations from their sample mean. The main equation to be estimated becomes: y = X + PX + ; (9) where P = IN IN IT JR, and the new error term has zero mean. The estimation results with the control of unobserved time-variant bilateral e¤ect are reported in Table 4. The coe¢ cient on PX is statistically signi...cant and positive, rejecting the null of zero correlation between the unobserved e¤ect and explanatory variables. It suggests that over time a stronger bilateral relationship leads to a larger amount of sectoral trade. Furthermore, note that the magnitude of most estimates has fallen except for the coe¢ cients on HAR and MRA_RO, compared to column I in Table 2. This result shows that consideration of the unobserved e¤ect reduces the explanatory power of most of the regressors but not of intra-EU harmonization and MRA with rules of origin.23 While this test cannot be regarded as conclusive, at least the inclusion of a measure of unobserved e¤ects does not alter our qualitative conclusions. 23The considerable decrease in the coe¢ cients of RTA variables with the inclusion of is not surprising, ijt since RTA variables also measure time-variant bilateral relations. 24 Table 4: Robustness analysis: unobserved e¤ects Regressand: ln(import) I Harmonization on intra-regional trade 0.4561*** (HAR) (0.010) Harmonization on imports from the ROW 0.0515*** (HAR_M) (0.005) Harmonization on exports to the ROW 0.6046*** (HAR_E) (0.005) MRAs with rules of origin on intra-regional trade 2.4154*** (MRA_RO) (0.031) MRAs with rules of origin on imports from the ROW -0.7087*** (MRA_RO_M) (0.037) MRAs with rules of origin on exports to the ROW 0.1738*** (MRA_RO_E) (0.036) MRAs without rules of origin on intra-regional trade 0.3197*** (MRA_NRO) (0.074) MRAs without rules of origin on imports from the ROW 0.2512*** (MRA_NRO_M) (0.037) MRAs without rules of origin on exports to the ROW 1.1430*** (MRA_NRO_E) (0.037) RTA on intra-regional trade 0.0539*** (RTA) (0.019) RTA on imports from the ROW -1.2349*** (RTA_M) (0.012) RTA on exports to the ROW -1.2540** (RTA_E) (0.012) Time-variant bilateral e¤ect (PX) 11.2570*** Number of observations 4160352 Log likelihood -7834125 25 5.2 Endogenous regressors The problem of endogenous regressors would lead to an overestimation of the trade- enhancing e¤ect of initiatives on standards if the initiatives were implemented where trade was already growing rapidly. Formally, this concern can be expressed as x = 1 if x > 0 (10) x = 0 otherwise, where x refers to the endogenous regressor in the vector of explanatory variables X, and x = 0Z + v where Z represents a vector of exogenous/predetermined variables. To correct for the endogeneity, we ...rst use the Instrumental Variable (IV) approach. If prior growth in trade led to the policy initiatives, then it would be appropriate to use the lagged three-year average trade growth, i.e. growthijr(t 1)( Z), as an instrument. To avoid the problem of identi...cation, we construct a comprehensive indicator, denoted by INI, that considers all the policy initiatives, i.e. harmonization, MRA with rules of origin, and MRA without rules origin: INI = 1 if there exist any of the above initiatives between the exporter and the importer at an industry in a particular year and 0 otherwise. Column II of Table 5 reports the two-stage IV estimation results, while column I reports the benchmark results where the potential endogeneity of regressors is not corrected. In the ...rst stage, the e¤ect of lagged average trade growth on the initiatives, estimated in a probit model, is signi...cantly negative. This result suggests that two countries are more likely to take a standards-related initiative in situations of relatively slow growth of bilateral trade. In the second stage, we ...nd that a 10% increase in the likelihood of an initiative is associated with a 76% increase in trade. Put di¤erently, the estimated e¤ect of INI after correcting for possible endogeneity is 1.5544, which is signi...cantly larger than the estimated e¤ect reported in column I. These results suggest that if there were an endogeneity problem, failure to correct for it would lead to an underestimation of the impact of the initiatives. As an alternative to the IV approach, we adopt the two-step estimator proposed by 26 Table 5: Robustness analysis: endogenous regressors I II III Stage 1: Initiative on lagged three-year average trade growth ­ -0.1583*** -0.1583*** Stage 2: Regressand: ln(imports) Initiative on intra-regional trade / 1.4821*** 1.5544*** / 1.5010*** Probability of the initiative on intra-regional trade 5.6690*** (INI / INI_PROB) (0.021) (0.021) (0.020) Initiative on imports from the ROW -0.0193* 0.3020*** 0.0618*** (INI_M) (0.010) (0.010) (0.010) Initiative on exports to the ROW 1.1464*** 1.3365*** 1.07*** (INI_E) (0.010) (0.009) (0.010) RTA on intra-regional trade 1.8908*** 2.1206*** 1.6474*** (RTA) (0.013) (0.012) (0.013) RTA on imports from the ROW 0.0911*** -0.0450*** -0.0471*** (RTA_M) (0.006) (0.006) (0.006) RTA on exports to the ROW -0.0091 -0.0919*** -0.0787*** (RTA_E) (0.006) (0.006) (0.006) Generalized residual ­ ­ 6.4712*** Number of observations 1804781 1804781 1804781 Log likelihood -4598521 -4566748 -4536913 27 Vella (1993) which deals with censored endogenous regressors.24 As in the previous IV approach, we ...rst estimate the e¤ect of the lagged three-year average trade growth on standards-related initiatives. Then we compute the generalized residual from the ...rst stage, i.e. E (^jx), to be included as an additional regressor in our original estimating v equation in column I. The original equation can be rewritten in terms of its conditional expectation as follows: E (yjx) = X + E (^jx): v (11) The two-step estimators are reported in column III of Table 5. The estimates obtained in the second stage again con...rm that correcting for endogeneity associated with the prior trade growth rates leads to a slight upward revision of the impact of initiatives. 6 Conclusion This paper has analyzed the implications for trade of various regional initiatives that deal with technical barriers. Both analytically and empirically, we found that harmonization and mutual recognition can have a signi...cant positive impact on trade within the region and with third countries. But there is a quali...cation in each case. If the harmonized standard is stricter than the initial standard in some countries, then the bene...ts from market integration in terms of economies of scale can be o¤set by the increased production cost due to a stricter standard. Countries with stricter initial standards witness a larger increase in imports relative to those with less strict initial standards. Exports of developing countries to the harmonizing region di¤er sharply across destinations: their exports to countries that raise their standards fall, whereas exports to the harmonizing countries with stricter initial standards rise moderately. When mutual recognition agreements contain restrictive rules of origin, then their bene...ts are con...ned to countries within the region at the expense of imports from the rest 24Vella (1993) uni...es the literature on censored endogenous regressors and self-selectivity bias, and employs generalized residuals to adjust for the inconsistency caused by the endogeneity of the censored regressors. The estimator proposed by this procedure (with binary censored endogenous variable), in the case where X does not contain an intercept and only values of y corresponding to speci...c values of x are observed, is essentially the two-step estimator in Heckman (1979) that corrects the self-selectivity bias. Hausman (1978) also argues that inconsistency due to the endogeneity of regressors can be adjusted by the inclusion of the residuals in place of the predicted values of the endogenous variable. 28 of the world, in particular, developing countries. When MRAs are open to ...rms regardless of origins, both intra-regional trade and trade with the rest of the world, especially with developing countries, rise substantially. We noted in the introduction that multilateral rules on goods trade have taken a per- missive approach to regional agreement on standards. It is neither feasible nor desirable to restrict the freedom of countries to harmonize or mutually recognize their standards. However, more could be done to protect the rights of countries that are excluded from such agreements. This is particularly important because few of the agreements on standards include developing countries, and the big di¤erences in social preferences over issues such as safety and the environment suggest that few developing countries are likely to be party to such agreements with industrial countries in the foreseeable future. Multilateral rules should make it more di¢ cult to conclude agreements of the type that impose an unnec- essarily high price of exclusion, especially on developing countries. Thus, the imposition of restrictive rules of origin, which deny the bene...ts of mutual recognition agreements to third countries for no legitimate reason, should be outlawed. It is harder to prevent the upward harmonization of standards. But it may be possible to build on the presumption that international trade law already creates in favor of the use of international standards by individual countries (in Articles 2.4 and 2.5 of the WTO's Agreement on Technical Barriers to Trade) and oblige countries that harmonize their standards to demonstrate why international standards or the less strict of the original standards are not adequate to meet their regulatory goals. This paper should be seen as the beginning of a research program, and there remains much scope for deepening the analysis. In particular, two types of industry-level data would help: ...rst, on how the level of harmonized standards compares with the standards that countries originally imposed; second, the implications of complying with standards for the cost of ...rms, across industries and countries. Such data would make it feasible to carry out the industry-level analysis that would ideally be the next step in this research program. 29 Appendix A Table A.1: The New Approach Directives Directives Reference Low voltage equipment 73/23/EEC Simple pressure vessels 97/23/EC Toys 88/378/EEC Construction products 89/106/EEC Electromagnetic compatibility 89/336/EEC Machinery 98/37/EC Personal protective equipment 89/686/EEC Non-automatic weighing instruments 90/384/EEC Active implantable medical devices 90/385/EEC Gas appliances 90/396/EEC Hot water boilers 92/42/EEC Civil explosives 93/15/EEC Medical devices 93/42/EEC Potentially explosive atmospheres 94/9/EEC Recreational craft 94/25/EC Lifts 95/16/EC Refrigeration appliances 96/57/EC Pressure equipment 97/23/EC In vitro diagnostic medical devices 98/79/EC Radio and telecommunications terminal equipment 99/5/EC Cable installation designed to carry person 00/9/EC Packaging and packaging waste 94/62/EC High speed rail systems 96/48/EC Marine equipment 96/98/EC Table A.2: The MRAs of Conformity Assessment MRA of Conformity Assessment Rules of Origin EU and Australia Yes EU and New Zealand Yes EFTA and Australia Yes EFTA and New Zealand Yes INTRA EU Yes EU and USA No EU and Canada No Australia and New Zealand No Canada and Korea No Canada and Swiss No 30 Appendix B.1: Proof for Proposition 1. Proof. (i) In markets with strictest initial standards, dsj = 0 and dNj > 0, and thus equation (7) becomes 1 @nzj dQij = (a.1) ij X" X ij @nmj z m Vmj + #ijVzj#dNj; where ij nij@qij=@niz and #ij qij@nij=@niz. Equation (a.1) can be further simpli...ed as dQij @nij dQij = Vzj dNj; (a.2) dnij X @nzj z ij where dQij=dnij = @Qij=@nij + Pz(Provided @nij=@nzj)(Vzj= ). Pz (@Qij=@nzj) (@nzj=@nij). Denote 'ij (dQij=dnij) ij that @Pj =@Qij > SOCij=nij (SOCij 2(@Pj=@Qij)nij denotes the second- order condition to maximize ij), qij=nij < @qij=@nij = qij (@Pj=@Qij)=SOC < 0 and thus @Qij=@nij = qij + nij@qij=@nij > 0. Furthermore, we ...nd @ Pj Qzj @qzj @Vzj=@nij = (a.3) @nij czsj @nij @Pj @Qij @Pj @qzj = + czsj @Qij @nij Qzj @Qzj + Pj nzj @nij < 0; because @Pj =@Qmj < 0 8m (the negative slope of the demand function), @Qij @qij @Pj nij 1 = qij > 0; (a.4) @nij = qij + nij @nij = qij 1 @Qij SOCij 2 (@Pj =@Qzj)Qzj + Pj = czsj at pro...t maximization, and @qzj=@nij < 0. Thus, (nNj= F) @Vzj=@nij @nzj=@nij = < 0: (a.5) (nNj= F) @Vzj=@nzj 1 31 In addition, we know @Qij=@nzj = qij@nij=@nzj+nij@qij=@nzj < 0. Hence, dQij=dnij > 0. Moreover, because @Pj =@Qkj = @Pj =@Qlj (products that meet the same standard are identical to consumers regardless of the production location) and qkj > qlj, we ...nd @qzj=@nkj < @qzj=@nlj (where @qzj=@nij = qij (@Pj=@Qij)=SOCzj), as well as (@Pj =@Qkj) (@Qkj=@nkj) < (@Pj =@Qlj) (@Qlj=@nlj) (where @Qij=@nij is de...ned in (a.5)). Therefore, @Vzj=@nkj < @Vzj=@nlj < 0, which unambiguously leads to @nkj=@nlj > 1. Because the pro...t-maximizing reaction function, i.e. qij = f qzjjz6=i , is a strictly decreasing function of ci, given by @qij sj = < 0 (a.6) @ci @2Rij=@qij2 where @2Rij=@qij < 0. At equilibrium, @qij=@ci < 0: We ...nd that Vij is also strictly 2 decreasing when ci rises, because @Vij @Rij @qij @qij = sj qij ci < 0 (a.7) @ci @qij @ci @ci given @Rij=@qij > 0 and @qij=@ci < 0. Thus, Vkj > Vlj. Because Vkj > Vlj > (@nkj=@nlj) Vlj, 'kj > 0 and dQkj > 0. As for dQlj, when dsj = 0, dQlj > 0 (i.e. 'lj > 0) if and only if Vlj=Vkj > @nlj=@nkj, where Vij (Pj cisjqij) for i = k;l. The above condition holds when ((Pj clsj)qlj=(Pj cksj)qkj =)(Pj clsj)2=(Pj cksj)2 > @nkj=@nlj, where Pj > cisj, @nkj=@nlj is de...ned in (a.3) with i = k, and the ...rst equality follows, at pro...t maximum, Pj @Pj=@qij cisj = 0. When cl < g(ck) 1=sj[Pj + b(Pj cksj)] where b [ Nj= lj(@Vlj=@nkj)]1=2, (Pj clsj)2=(Pj cksj)2 > @nkj=@nlj, 'lj > 0 and thus dQlj > 0; when cl > g(ck), dQlj < 0. (ii) In contrast, in markets with less strict initial standards, dsj > 0 and dNj > 0. Equation (7) becomes dQij @nij @qij Nj dQij @nij @Vzj dQij = Vzj dNj + nij + (a.8) dnij X @nzj @sj dnij @nzj @sj z ij " ij X z #dsj: 32 Denote ij nij(@qij=@sj) + Nj= ij Because (dQij=dnij) Pz(@nij=@nzj)(@Vzj=@sj). @nij @Vij = Nj < 0; (a.9) @ci ij @ci we ...nd nkj < nlj. Furthermore, @Pj @qij ci @sj = < 0; (a.10) @sj SOC because ci > @Pj=@sj, and SOC < 0. Given ck < cl, we obtain @qlj=@sj < @qkj=@sj < 0. Thus nlj@qlj=@sj < nkj@qkj=@sj < 0, where nij@qij=@sj denotes the direct e¤ect of an increase in sj that appears in equation (a.8). The second multiplier of dsj in equation (a.8) denotes the indirect e¤ect of an increase in sj. We consider ci > c @Pj =@sj + ¯ (Nj= ij net e¤ect of an increase in sj, i.e. )(dQij=dnij) Pz [(@nij=@nzj)ij(@Vzj=@sj)] SOC=nij, which indicates the negative < 0. Denote the solution of dsj that sets (a.8) to zero as s for i = k, which yields s = ('kj= kj)dNj > 0. We conclude dQkj > 0 when ¯ ¯ dsj < s while dQkj < 0 otherwise. ¯ When dsj > 0, given another additional negative impact on dQlj arising from dsj > 0, we know dQlj < 0 when cl > g(ck). When cl < g(ck), dQlj < 0 if and only if Vlj=Vkj < < 0. The above condition holds lj(dsj=dNj)1=[(dQij=dnij)=Vkj] @nlj=@nkj where ij when cl > g(ck) [ (dsj=dNj)1=[(dQij=dnij)=Vkj] ¯ 1=sj[Pj+¯b(Pj cksj)] where¯b Nj= lj (@Vlj=@nkj)]1=2. Because b < b, g(ck) < g(ck). Thus, we conclude that, when ¯ ¯ dsj > 0 and cl >g(ck), dQlj < 0. ¯ Appendix B.2: Proof for Proposition 2. Proof. As suggested in Appendix B.1, when ci > c~, the net e¤ect of an increase in sj is negative. Comparing the harmonizing markets of dsj = 0 with those of dsj > 0, we ...nd that, in the latter markets, the positive market integration e¤ect may be partially or even completely o¤set by the negative e¤ect of dsj > 0. Thus, the increase in imports of harmonizing market j, i.e. dQij, from any exporter i is a strictly decreasing function of 33 dsj. Appendix B.3: Proof for Proposition 3. Proof. (i) In the case of mutual recognition with restrictive rules of origin, dnij > 0 for i;j 2 H because dNj > 0 and dsj 6 0, and dnzj = 0 for z 2 H in equation (6). Thus = dnij > 0 and dnzj < 0. As a result, for i;j 2 H, given dsj 6 0, dnij > 0, dnzj < 0, and @nij=@nzj < 0, dQij > 0 in equation (4). For i 2 H;z 2 H, and j 2 H, = dQzj @Qzj dQzj @nzj = + < 0; (a.11) dnij @nij dnzj @nij and thus given dsj = 0 and dnij > 0, dQzj < 0. (ii) For k;l 2 H, j 2 H, = d(Qlj Qkj) d(nljqlj nkjqkj) = (a.12) dnij dnij dqlj dqkj @nlj = nlj dnij nkj dnij + qlj @nij qkj@nkj : @nij Note nlj=nkj > dqkj=dqlj, because @Pj =@Qj (@Qj=@qlj) dqkj=dqlj = @Pj =@Qj (@Qj=@qkj) + 1 @Pj =@Qj nlj nlj = < : (a.13) @Pj =@Qj nkj + 1 nkj Therefore, nljdqlj=dnij < nkjdqkj=dnij. When cl > g1(ck) 1=sj[Pj+¯b0(Pj cksj)] where b0 =(Nj@Vlj=@nkj), @nlj=@nkj < qkj=qlj, and qlj@nlj=@nij < qkj@nkj=@nij given ¯ lj @nkj=@nij = 1 with i 2 H;k 2 H. Therefore, d(Qlj = Qkj)=dnij de...ned in equation (a.12) is negative. Given dnij > 0, d(Qlj Qkj) < 0 and thus dQlj < dQkj < 0. Appendix B.4: Proof for Proposition 4. Proof. (i) In the case of mutual recognition without rules of origin, dNj > 0 and dsj 6 0 34 for exporters of all origins. When the exporter countries are type K (either within the region or in the rest of the world) or type L with cl < g(ck), the e¤ects of both dNj > 0 and dsj 6 0 are positive as shown in Appendix B.1. Thus, imports of any harmonizing market from type K countries ­both within the region and in the rest of the world ­and type L countries with cl < g(ck) increase, i.e. dQij > 0, dQkj > 0, dQlj > 0 for i; j 2 H, k;l 2 H. = (ii) As mentioned in Appendix B.1, the negative direct e¤ect of dsj > 0, i.e. nij@qij=@sj, is stronger for i = l. Next we consider and compare the indirect e¤ect of dsj, i.e. Nj= ij (a.5) for i = k;l and (dQij=dnij) Pz (@nij=@nzj) (@Vzj=@sj), where @nij=@nzj is de...ned in equation @Vzj @Pj @qzj = cz qzj + (Pj cisj) (a.14) @sj @sj @sj = qzj "@Pj cz + @sj @qzj @sj @Pj @qzj #: When cl > g2(Fk) @Pj =@sj+ lj b [(@Pj =@sj) ck+ kj] with zj (@Pj =@qzj)(@qzj=@sj) and b (qkj=qlj) maxf kj =(Nj@Vkj=@nlj); Nj= lj (@Vlj=@nkj)g, we then obtain that Pz negative net e¤ect of dsj > 0 is stronger forPz= l, i.e. (@nkj=@nzj) (@Vzj=@sj) > 0 as well as (@nlj=@nzj) (@Vzj=@sj) < 0. Overall, the i < < 0 where is de...ned lj kj ij in Appendix B.1, which in turn suggests the positive impact of dsj < 0 is larger for i = l. Now consider = minf lj PjgdQij given j 2 H, where dQij is de...ned in equation (a.8). Denote kj < 0 for j 2 H. Then XdQlj XdQkj = 'kj)dNj + (a.15) lj kj dsj j j X('lj j X j > X('lj 'kj)dNj + j Xdsj: j We ...nd that Pj dQij > 0 if Pj dsj < s (1= 'kj)dNj. ¯ ) Pj('lj 35 References [1] Akerlof, George. (1970) "The Market for 'Lemons': Quality Uncertainty and the Market Mechanism". Quarterly Journal of Economics 84: 488-500. [2] Baldwin, Richard. (2000) Regulatory Protectionism, Developing Nations and A Two- Tier World Trade System. CEPR Discussion Paper No. 2574. [3] Baltagi, Badi. (2001) Econometric Analysis of Panel Data, 2nd Edition. John Wiley & Sons, Chichester, England. [4] Frankel, J., Ernesto Stein, and Shang-Jin Wei. (1997) Regional Trading Blocs in the World Economic System. Institute for International Economics. Washington, DC. [5] Ganslandt, Mattias, and James Markusen. (2001) "Standards and Related Regula- tions in International Trade: A Modeling Approach". NBER Working Paper No. w8346. [6] Hausman, J. (1978) "Speci...cation Tests in Econometrics". Econometrica 46: 1251- 1271. [7] Heckman, J. (1979) "Sample Selection Bias as a Speci...cation Error". Econometrica 47: 153-161. [8] Maskus, Keith E. and John S. Wilson. (2001) "A Review of Past Attempts and the New Policy Context", in Quantifying the Impact of Technical Barriers to Trade: Can it be done? edited by Keith E. Maskus and John S. Wilson. University of Michigan Press, Ann Arbor, MI. [9] Moenius, Johannes. (2000) Three Essays on Trade Barriers and Trade Volumes. Ph.D. Dissertation. University of California, San Diego. [10] Mundlak, Yair. (1978) "On The Pooling of Time Series and Cross-Section Data". Econometrica 46: 69-85. 36 [11] Soloaga, Isidro, and L. Alan Winters. (2001) "Regionalism in the Nineties: What Ef- fect on Trade?" CEPR Discussion Paper 2183. Center for Economic Policy Research, London. [12] Swann, Peter, Paul Temple, and Mark Shurmer. (1996) "Standards and Trade Per- formance: The UK Experience". Economic Journal 106: 1297-1313. [13] European Commission. (1998) Single Market Review. Subseries III: Dismantling of Barriers. Volume 1: Technical Barriers to Trade. [14] World Bank. (2004) Technical Barriers to Trade Database: Washington, DC. [15] Vella, Francis. (1993) "A Simple Estimator for Simultaneous Models with Censored Endogenous Regressors". International Economic Review 34: 441-457. [16] Vogel, David. (1995) Trading Up: Consumer and Environmental Regulation in A Global Economy. Harvard University Press, Cambridge, Massachusetts. 37