非技術的總體措施是使用傳統的保護手段來限制進口貿易，并通過多邊貿易談判來獲得支持減低關稅和設立貿易總協定(GATT),世界貿易組織(WTO)合作事宜也是通過此措施來實現的。根據全球關稅下降和定量措施,增加使用非關稅壁壘和應急保護措施可以促使一些評論家來關注替代品貿易的經濟發展。反傾銷措施在關鍵的位置可以用來保護進口商品以及多邊貿易協定的限制之下少受到約束。

非關稅貿易壁壘的重要性正在逐步受到人們的關注,特別是貿易應急保護措施已成為一個全球貿易政策的重要特征,尤其是在發達國家和發展中國家之中表現最為突出。減少關稅保護或自行設立保護措施越來越被視為另類貿易政策，交易過程中利用貿易工具來保護國內生產商并且可以促進行業發展(Blonigen Prusa,2003;Konings Vandenbusche,2005)。我們可以把最初的設想設計為“公平”貿易措施。作為反傾銷的特點就是商品貿易可以受到保護，由于反傾銷政策的建立和實施在相當大的程度上是受到經濟政策的影響,因此所產生的另一種渠道就是要保護游稅政策(納爾遜,2006)。

Non-Technical Summary The use of traditional measures of import protection has been increasingly restricted by multilateral trade negotiations under the auspices of the General Agreement on Tariffs and Trade (GATT), and the World Trade Organisation (WTO). In light of globally declining tariffs and quantitative measures, the remarkable increase in the use of non-tariff barriers and contingent protection measures has prompted several commentators to express concerns over their use as potential protection substitutes. Antidumping measures feature prominently amongst the forms of alternative import protection and their use tends to be much less constrained by coordinated, multilateral trade agreements. Given the way in which antidumping policies and procedures are designed and implemented, an alternative channel for protection lobbying emerges providing ample room for political-economy influences and hence a rationale for the substitution of trade policy instruments. This paper investigates the hypothesis of trade policy substitution, following a major, coordinated trade reform, by examining the relationship between multilateral tariff concessions and the subsequent use of antidumping actions by the European Union (EU). Motivated by Anderson and Schmitt’s (2003) theoretical contribution, which provides a theoretical framework for an increasing use of antidumping measures following binding tariff and quota restrictions, we address the substitution hypothesis by focusing on the impact of the Uruguay Round tariff cuts on subsequent EU antidumping investigations at a detailed HS 8-digit product-country level. The Uruguay Round substantially reduced bound MFN tariffs for most signatory countries and also required its signatory countries to ‘tariffy’ and reduce quantitative restrictions. Given that antidumping actions largely remain WTO-unconstrained, we argue that the Uruguay Round trade agreements may represent a suitable policy setting for a potential substitution effect of declining tariff protection for an enhanced use of product-country level antidumping measures. Our results provide support for Anderson and Schmitt’s (2003) theoretical contribution. A statistically highly significant, however in absolute size limited, positive impact of bound MFN tariff cuts on the probability of subsequent antidumping investigations is found; having controlled for other influences. Employing a variety of different econometric techniques, including random-effects and a ChamberlainMundlak approach to control for unobserved heterogeneity, this finding is shown to be robust to a series of sensitivity tests.

The growth in importance of non-tariff trade barriers in general and contingent protection measures in particular has become a remarkable feature of the conduct of global trade policy, both in the developed and more recently also in the developing world (Prusa, 2001 and 2005; Zanardi, 2004; Bown, 2008). With reduced scope for tariff protection contingent protection measures are increasingly seen as alternative trade policy instruments to protect domestic producers and industries (Blonigen and Prusa, 2003; Konings and Vandenbusche, 2005). Originally devised as ‘fair’ trade measures antidumping features prominently amongst the forms of contingent protection. Given the way in which antidumping policies are set up and implemented considerable room for political-economy influences tends to be created, thus generating an alternative channel for protection lobbying (Nelson, 2006).1 Support for the hypothesis of declining tariff protection being replaced by an enhanced use of antidumping investigations may be found in early descriptive studies identifying anti-dumping as “a major loophole in the free-trading disciplines of the world trading system” (Lindsay and Ikenson, 2001:5).2 Thorough empirical evidence on the subject matter is however still scarce and tends to be industry analysis and characterized by mixed results.3 Focusing on the impact of the Uruguay Round tariff concessions, Feinberg and Reynolds (2007) analyse subsequent antidumping investigations in 19 different industries for several countries between 1996 to 2003. They find evidence for trade policy substitution mostly in developing countries.4 Traditional users of anti-dumping measures (i.e. Australia, Canada, New Zealand, the EU and US) are not found to show a positive correlation between tariff protection and anti-dumping proceedings. Moore and Zanardi (2011) further add to these findings by examining the relationship between the probability of AD investigations and applied (rather than bound) tariffs between 1991 and 2002.

Nelson (2006:554) reviews of the literature on the political economy character of antidumping measures highlighting that “[..] antidumping is a much worse a problem than its small coverage and marginal contribution to aggregate protection would imply.” 2 Vandenbussche and Zanardi (2010), moreover, find in this context that antidumping measures considerably affect trade in industries which are not directly involved in the investigation thereby characterising antidumping investigations as a potentially very powerful tool of alternative import protection. A view which is also held by Blonigen and Prusa (2003:253) who state that most people “agree that AD has nothing to do with keeping trade ‘fair’ […] It is simply another form of protection”. 3 There is a related literature that focuses on the political choice between tariffs and other forms of (non-AD related) import protection. Hillman (1990), Hillman and Ursprung (1988) and Feenstra and Lewis (1991) analyse the use of tariffs and Voluntary Export Restraints (VERs) showing that the latter may, under certain assumptions, be preferred to tariffs. Limão and Tovar (2011) provide theoretical and empirical evidence for a substitution scenario of tariffs for non-tariff-barriers (NTBs) in general. 4 Feinberg and Reynolds (2007) focus on AD petitions in HS 1-digit industries.

ISIC 3-digit manufacturing industries and numerous countries, the authors are not able to confirm the findings of a positive correlation between tariffs and antidumping, with the exception of a small group of developing economies.5 By contrast, the sole study conducted at a detailed product level, Bown and Tovar (2011) provides support for the hypothesis of tariffs being substituted by more frequent AD investigations when analysing India’s antidumping proceedings in the face of a major tariff reform programme.6 We seek to contribute to the existing literature by examining a potential product-level link between (bound) mfn tariff cuts conceded by one of the world’s largest traders – the EU – and the latter’s subsequent antidumping investigations. Our study contrasts to much of the previous empirical evidence by focusing on detailed and country-specific HS 8-digit productlevel AD investigations, and (bound) MFN tariff concessions for a large and developed economy.7 As pointed out by Feinberg and Reynolds (2007) the fact that industry classifications usually include several hundreds of individual product lines, industry and country level studies may lead to biased results, since sectors with a large variation in product level tariff cuts and possibly very small aggregate tariff reductions are be more likely to attract subsequent AD investigations than industries with a large aggregated degree of tariff liberalization but no extreme product level tariff reductions. Our research is motivated by Anderson and Schmitt’s (2003) theoretical contribution which analyses the effect of binding tariff reductions on the use of quantitative import restrictions and anti-dumping measures.8 Based on Brander and Krugman’s (1983) reciprocal dumping model, these authors derive a theoretical framework of preference progression for different forms of trade policy protection. They show that in an unrestricted trade policy。

These include Argentina, Brazil, China, India, Mexico, Peru and South Africa (Moore and Zanardi, 2011). Moore and Zanardi (2011) find evidence for antidumping deflection and retaliation as well as the importance of the size of import-flows when determining the likelihood of AD investigations across all country subsamples. 6 Bown and Tovar (2011) analyse cross-sectional HS 6-digit imposed antidumping data for India’s pre- and post- IMF imposed reform period (i.e. 1990 and 2000 to 2002). Based on Grossman and Helpman’s (1994) import protection model they find that India’s 1990 tariff policy is in line with the latter model’s prediction whereas India’s post-reform tariff data is not. Re-estimating the post-reform model including tariffs as well as imposed antidumping and safeguard duties, however, again results in theory-consistent significant estimates pointing to a substitution effect of trade policies following the IMF imposed tariff reform programme. 7 Feinberg and Reynolds (2007) as well as Moore and Zanardi (2011) use rather broad industry or country-level data when analysing the tariff-antidumping nexus for developed economies. 8 Further theoretical contributions on the substitution of different trade policies include Limão and Tovar (2011) who show in a political choice model that governments may benefit from coordinated tariff constraints through a higher bargaining power towards domestic special interest groups which then enhances the latters’ efforts to lobby for alternative forms of protection. Moreover, political pressure deflection by governments committed to tariff liberalization as a further rationale for explaining the substitution of declining tariffs by more antidumping investigations has been analysed by Anderson and Zanardi (2009) as well as Moore and Zanardi (2011). Moore and Zanardi show that political decision makers may increasingly try to shift protectionist demands towards more administered forms of protection in order to reduce pressure from domestic interest groups.
 

Environment tariffs tend to be the most preferred protectionist trade policy tool followed by quotas and antidumping measures. As a result, when constraining the use of tariffs by coordinated negotiations, policy-makers are likely to resort to the use of quantitative trade policy instruments which are again superseded by the use of antidumping actions in the presences of additional agreements on ‘quota tariffication’. Restrictions on the use of tariffs and quotas will thus result in an enhanced use of antidumping protection. This trade policy preference progression tends to be in line with some stylized facts regarding the historical use of trade policy instruments. Coinciding with the end of the Kennedy Round (1964-1967), the 1960’s witnessed an upsurge of quantitative import barriers which was followed by an increasing trend towards antidumping measures since the 1980s (Renner, 1971; Finger and Olechowski, 1987). The Uruguay Round (1986-1994) finally established a guideline for the ‘tariffication’ of quantitative import restrictions for all GATT-signatory countries and additionally required them to restrict the use of quotas in the future, whereas the use of AD measures remains largely WTO-unconstrained. Import protection following the Uruguay Round (UR) tariff commitments therefore represents an interesting testing environment for a potential substitution effect of greater use of antidumping measures in response to falling tariffs. Focusing on the UR trade policy outcome, our findings show a highly significant, albeit small, positive impact of bound MFN tariff concessions on the probability of subsequent antidumping investigations; having controlled for other influences. Employing a variety of different econometric techniques, including random-effects and a ChamberlainMundlak approach to control for unobserved heterogeneity, this finding is robust to a series of sensitivity tests. The reminder of the paper is organized as follows. Section 2 describes the legal framework of the EU’s anti-dumping policy and provides some descriptive statistics. Section 3 briefly sketches out the conceptual framework which motivates our study, while section 4 introduces the empirical methodology followed by a discussion of the results in section 5. Section 6 concludes.
European Anti-dumping Policy and Uruguay Round Tariff Concessions.

 Legal framework The EU’s trade policy is governed by the European Council and the European Commission. While the Commission proposes and enforces trade policy actions, the Council, consisting of Member States’ representatives, decides about approval or rejection of the Commission’s propositions. Antidumping measures represent a major component of the EU’s trade policy mix (Rovegno and Vandenbusche, 2011). Guided by Article 207 of the Treaty on the Functioning of the European Union as well as Council regulation 1225/2009, the EU’s antidumping legislation is embedded in the WTO’s antidumping policy framework allowing GATT signatory countries to impose discriminatory trade protection measures if foreign exporters sell their goods at a price lower than their ‘normal value’,9 and if the latter results or threatens to result in ‘material injury’ for the domestic industry. The initiation of an antidumping investigation on part of the EU’s antidumping authorities requires an officially lodged complaint by a Community industry which needs to provide evidence of dumping and the resulting causal material injury. Additionally, any antidumping complaint must be supported by enough EU producers responsible for at least 25% of the EU’s product-specific production. EU regulations further specify a timeframe of 45 days for the Commission to decide whether to open an investigation or not. Preliminary measures may be imposed after an initial investigation period of 9 months, during which (mostly questionnaire-based) consultations are held with EU producers and importers as well as the investigated exporters. The time span from the opening of an investigation to the publication of the final decision may therefore take up to 15 months.

The ‘normal value’ of a product is in general defined as the country of origin’s production costs plus reasonable profit margins and additional costs for selling and administration. In calculating the normal value the European Commission distinguishes between whether the investigated country is a market economy or not. If it is not, an analogue country, often already proposed by the complaining industry, serves as a proxy (Liu and Vandenbusche, 2002). In light of the difficulties of estimating production costs, the European Commission often uses domestic sales prices in the exporting country to calculate the normal value. Price information of the analogue country is also used if domestic sales in the exporting, or analogue, country are too small to be representative. For more detailed information on the determination or ‘construction’ of the normal value see Macrory et al. (1991). 10 The EU can initiate anti-dumping investigations against all non-EU member countries, with an almost complete exception of goods stemming from Iceland, Lichtenstein and Norway (i.e. the EEA countries).

Comparing foreign suppliers’ export prices with ‘normal values’, the European Commission first investigates whether there is enough proof for the existence of dumping following a complaint of a Community industry. While the investigated export price refers to the ex-factory price - i.e. the price for goods sold to the EU net of rebates, discounts, taxes, etc. (Macrory et al., 1991), the normal value of a product is most often calculated on the basis of domestic sales prices of the like product in the exporting country. The difference between the latter two – i.e. the dumping margin – is then calculated according to one of three alternative measures specified in the WTO’s Antidumping Agreement (ADA).11 The determination of causal material injury to the domestic industry, or a threat thereof, includes an economic analysis of various domestic industry factors such as, output, productivity, profits, utilisation capacity, stocks, sales, market share, cash flow return on investment and employment, and also compares the foreign producers’ export prices to the prices charged by the domestic industry (i.e. the injury margin).12 If the Commission considers the evidence for dumping and material injury to be sufficient as well as potential trade defence actions to be in line with the general interest of the Community, the former finally proposes antidumping measures which may either take the form of price-undertakings or additional duties to offset the injury caused by the dumped products.13 Despite the fact that antidumping investigations directly target exporting firms and tend to impose firm-specific trade remedy duties, not investigated firms originating from the same country are most often also subjected to additional duties even if the latter were not involved in dumping activities. In the EU the duty imposed on so-called non-named or potential exporters amounts to the highest duty imposed on all investigated firms 。

The difference between a calculated normal value and the foreign firm’s export price determines the dumping margin. The WTO’s Antidumping Agreement (Article II) specifies three alternative approaches for contrasting the latter two prices: (i) comparing weighted averages of both price indices, (ii) comparing both price indices for each (product-level) transaction averaging the latter to compute the overall dumping margin, or (iii) contrasting weighted normal values with individual transaction based foreign producers’ export prices if the latter vary substantially across purchasers, time periods or regions. The latter method is also followed by the averaging of all transaction-to-transaction based dumping margins. Closely associated with the calculation of dumping margins is the methodology of ‘zeroing’. ‘Zeroing’ denotes the replacement of negative dumping margins by zeros which may finally results in larger average dumping margins. For a recent discussion regarding the different approaches of zeroing and associated WTO litigations see Prusa and Vermulst (2010). 12 When calculating the material injury of alleged dumping activities, the EU, like many other users of AD actions, often applies the principle of cumulation, which allows considering the combined impact of all imports from the investigated exporting countries on the domestic industry. Hansen and Prusa (1996) as well as Tharakan et al. (1998) find that cumulation significantly increases the probability of finding evidence for material injury. 13 The imposed duty rate in most cases reflects the dumping-margin unless the material injury could also be withdrawn with a smaller duty rate (‘lesser duty rule’). The anti-dumping import tax may either be an advalorem duty, a specific duty or a variable duty (i.e. a minimum import price). Moreover, in line with WTO regulations antidumping measures are in most cases imposed for a period of 5 years. Targeted parties may however ask for an interim review which may result in lower duty rates.
Same exporting country (Macrory et al., 1991). Newcomers, which did not export to the EU at the time of the investigation, are also subjected to the latter antidumping duties (mostly in order to prevent circumvention). Given the nature of this process of evaluation and duty setting, and that the mere initiation of an AD investigation may affect firms’ behaviour, the Commission has considerable discretion to eliminate foreign competition and to protect the domestic industry against foreign producers.

Uruguay Round Tariff Commitments and EU Antidumping Investigations During the Uruguay Round the European Union agreed to reduce its bound tariffs by almost a third, with considerable variation across industries and individual product lines. Table 1 (below) provides an overview of the EU’s bound Uruguay Round MFN tariff cuts, per industry. The sector with the largest average decline in tariff protection was the tobacco industry, with a cut of around 24 percentage points.15 Containing a much larger number of individual HS 8-digit product lines, the iron and steel sector comes second showing an average reduction in tariff protection of approximately 5.1 percentage points, followed by the processed food, furniture, paper, beverages and chemicals industries.16 In addition, coefficients of variation displayed in Column (4) also reveal that the tariff cuts within individual industries were not conducted uniformly and were subject to considerable intraindustry (i.e. product-level) variations. Table 1 further documents the EU’s country- and product-specific use of antidumping measures for the 28 manufacturing industries over the period 1996-2008.17 The most successful issuer of dumping complaints was the iron and steel industry with 491 investigated product-country pairs, followed by the textiles (232), industrial chemicals (114), fabricated metals (112), footwear (99) and electrical machineries (69) industries.

The potential of antidumping constraints to provide import protection to the domestic industry has also been highlighted by Messerlin and Reed (1995), who find that 90% of all AD measures are implemented on the basis of rather loose injury criteria - such as simple differences in prices rather than actual predatory pricing behaviour. 15 The relatively large average tariff reduction in the tobacco sector has to be interpreted with some caution as the tobacco sector only counts 6 HS 8-digit product lines, whereas the iron and steel industry includes 573 HS 8digit products. 16 The latter industries show average tariff cuts of 4.3, 4.2, twice 3.9 and 3.4 percentage points, respectively. Despite the much smaller average reductions in the latter sectors (relative to the tobacco sector), the former are still considerably above the average MFN tariff reduction which amounts to 2.7 percentage points for all manufacturing industries. 17 A list of the countries targeted by an EU antidumping investigation over the considered time horizon is provided in Annex table 3. 18 A similar ordering emerges when analysing the final imposed antidumping duties, with the iron and steel sector being the prime user of antidumping measures counting 218 product-country pairs subject.

Number of AD-targeted product lines instead of product-country pairs (Table 1, Column 6) results in a very similar ordering with the iron and steel, chemicals and textile industries representing the sectors with the highest number of AD targeted product lines.
Food Products 4.3 Beverages 3.9 Tobacco 23.8 Textiles 2.6 Wearing apparel 1.7 Leather products 1.6 Footwear except rubber 0.8 Wood products 3.2 Furniture except metal 4.2 Paper and products 3.9 Printing and publishing 3 Manufacture of industrial 351 177.633 928 2.4 2.7 1.1 43 114 chemicals 352 Other chemicals 92.187 294 3.4 2.9 0.9 6 7 353 Petroleum refineries 13.306 69 1.3 0.9 0.7 1 1 354 Misc. Petroleum and coal 2.362 11 1.2 1.1 1 0 0 355 Rubber products 29.598 75 1.6 1.1 0.7 0 0 356 Plastic products 54.209 116 2.5 1.8 0.7 15 32 361 Pottery china earthenware 16.012 21 1.9 1.1 0.6 0 0 362 Glass and products 46.443 135 2.5 1.3 0.5 0 0 369 Other non-metallic mineral 39.926 113 2.3 1 0.4 9 12 371 Iron and Steel 92.062 573 5.1 2.2 0.4 147 491 372 Non-ferrous metals 64.325 245 1.6 1.5 1 4 10 381 Fabricated metal products 190.094 425 2.6 1.5 0.6 22 112 382 Machinery except electrical 374.185 946 2.3 1.3 0.6 10 12 383 Machinery electrical 236.631 472 2.5 2 0.8 28 69 384 Transport equipment 101.818 298 2 1.8 0.9 6 12 385 Professional and scientific 159.825 314 3.1 1.9 0.6 3 3 390 Other manufactured 118.016 248 3.2 1.8 0.6 8 26 All Manufacturing Industries 2.680.408 7777 3.4 2.1 0.6 408 1273 Notes: The above statistics are based on the author's own calculation using product-country level import data from Comext, bound Uruguay Round tariff changes from the WTO's schedule of concessions and antidumping data from the World Bank’s global antidumping database. The statistics displayed in Table 1 are based on 2,680,408 observations. It is worthwhile noting that while the above table includes all country-specific HS 8-digit EU import flows between 1996 and 2008 our estimations only include countries and 4-digit ISIC industries where at least one antidumping investigation had been initiated over the considered time horizon. Introducing lagged regressors and growth variables further reduces the estimating sample to 701,272 observations including 1061 antidumping.
Antidumping duty. Further sectors with a rather high incidence of imposed duties are the fabricated metal (95), industrial chemicals (51), footwear (42) and electrical machineries (20) sectors. 19 The exact industry ordering is iron and steel (147), industrial chemicals (43), textiles (41), footwear (36), electrical machinery (28), fabricated metals (22). Moreover, analysing the distribution of imposed preliminary and final duties per industry (Annex table 4) delivers further interesting insights. The highest preliminary duties were, on average, imposed in the non-metallic minerals (66.1), the non-electrical machinery (50.1) and the leather products (48.3) industries, while the sectors with the highest average of imposed final duties were the electrical (51.1) and non-electrical (48.0) machinery, as well as the footwear (47.6) and wearing apparel industries (46.4).

Targeted product-country pairs. Annex table 4 displays the distribution of AD investigations per industry for the estimating sample and shows an almost identical frequency distribution across different industries.

Our empirical analysis is motivated by Anderson and Schmitt’s (2003) theoretical framework which provides a rationale for investigating the impact of tariff liberalization on the use of quotas and antidumping measures. Building on Brander and Krugman’s (1983) model of reciprocal dumping, they show that countries tend to resort to antidumping measures when tariffs and quotas are credibly restricted by coordinated (e.g. multilateral) commitments. Focusing on a two-country, two-firm, Cournot framework in which each firm sells the same good in both countries, Anderson and Schmitt (2003) analyse the choice between different trade policy instruments by means of a government objective function in the presence and absence of a multilateral liberalization commitment.20 The governments’ objective functions in both countries are thereby defined as: 21 U(τi, τi*) = [(β (τi), Π(τi, τi*)].

The term β denotes consumer welfare including the provision of public goods financed by tariff revenue and Π represents domestic industry profits earned at home and abroad. τ and τ* are protection parameters respectively set by domestic and foreign policy makers.22 Each government has tariffs, quotas and antidumping measures at its disposal; the latter options thereby defining τ: τi ∈ {ti, qri, adi}.

where ti represents an ad-valorem tariff rate for product i and qri and adi denote advalorem tariff equivalents of a binding quota or an antidumping restriction, respectively.

For simplicity the authors exclude the potential entry and exit of firms. Since some form of market imperfection is need in order to explain the use of quotas and antidumping constraints, the authors assume an oligopolistic market structure. Adopting a strategic Cournot interaction implies a tariff quota equivalence if quotas are auctioned off (see for instance Hwang and Mai, 1988). Anderson and Schmitt (2003) include transportation costs as an additional trade barrier in their model, but we simplify here by omitting these barriers. 21 Due to a symmetry assumption we focus in the following only on one country. 22 τ is not explicitly mentioned in Anderson and Schmitt (2003) but has here been introduced for illustrative reasons. It is further worth noting that the domestic industry’s profit does not only depend on the home government’s (τi) but also on the foreign government’s protection parameters (τi*), through its sales abroad. Moreover, the utility function is assumed to apply to each industry individually and to be strictly increasing in each argument as well as to be characterised by a strictly diminishing marginal rate of substitution. U(.) is modelled by a Cobb-Douglas function of the form: U= β1-α Πα, where α denotes the weight the government puts on producers’ profits. Following Stigler (1971) and Peltzman (1976), U(.) thereby takes into account the government’s concern for consumer as well as producer surplus - a concern for voters and political campaign contributions.

Assuming complete discretion for the government to set trade policy tools as freely as it chooses,23 Anderson and Schmitt establish that tariffs, when set unilaterally, are the most efficient protectionist trade policy tool. The intuition is that, while all three alternatives are likely to exert a similar impact on domestic prices, and by consequence on domestic producers’ profits, non-tariff trade barriers are assumed to be more costly for governments since the latter won’t generate any revenue gains.24 Anderson and Schmitt also show that, given any tariff rate t, unilaterally imposed quantitative constraints are preferred to antidumping constraints, since the latter may exert a negative impact on the domestic industry’s export profits whereas the former leave export profits unaffected.

Allowing both countries to commit to (reciprocal) trade liberalization via an internationally binding agreement, it is assumed that the tariff commitment reduces and restricts the use of external tariff protection relative to the unconstrained Nash equilibrium. Enhanced trade flows and declining local market power lead to Pareto improvements for both signatory countries [i.e. Uc(.) > Un(.)]. Given a government’s (still present) incentive to change the terms-of-trade to its own advantage the former may, however, decide to explore alternative (and potentially more subtle) ways of import protection following the binding tariff agreement.27 In this context, the government is assumed to first negotiate binding multilateral tariff cuts in order to internalize terms-of-trade effects.

Several authors note that the use of antidumping measures is influenced by political-economy forces and thus may lead to a less strict interpretation of the (anti-)dumping legislation (Moore, 1992; Baldwin and Steagall, 1994; Zanardi, 2004). 24 Anderson and Schmitt (2003) assume that quotas cannot be licensed off and are thus lost to foreign producers. The authors therefore find that any quota-cum-tariff or antidumping-cum-tariff protection can be achieved by using a (higher) optimal tariff. Since the latter additionally generates tariff revenue the use of tariffs maximizes U(.). 25 Faced with an antidumping constraint imposed by country A, a supplying exporting firm (located in country B) then decides whether to exit the market or whether to supply A without dumping and thus complying to the condition: pA≥ pB + t. An antidumping constraint imposed on the foreign firm reduces the latter’s output in country A, however, increases its output in country B (in order to comply with the above antidumping constraint), which will finally result in declining prices in B and increasing prices in A. An antidumping constraint imposed by country A on the foreign firm domiciled in B, therefore, not only protects the domestic industry but also reduces the domestic industry’s export profits since profits are a decreasing function of the opponent’s output. 26 Anderson and Schmitt (2003) further point out that the preference ordering also holds in a Cournot setting with several domestic and foreign firms. 27 It is thereby assumed that both countries are not able to deviate from the lower negotiated tariff rates, as is the case when bound MFN tariff reductions are negotiated in GATT/WTO trade rounds. Furthermore, Anderson and Schmitt (2003) assume that trade agreements are formed to overcome negative terms-of-trade effects as illustrated in Bagwell and Staiger (1999). Others, like Maggi and Rodriguez-Clare (1998) and Limão and Tovar (2011), also include motives of better fending off lobbying pressure by having access to a commitment technology.

Alternative ways to influence these to its own benefit.28 Given a restriction on the use of tariffs, the choice is between quotas or antidumping measures. In presence of an additional constraint on the use of quotas, antidumping actions prevail further limiting the scope of the protection parameter τi. The probability of using an antidumping measure on product i (illustrated by the variable adi), is thus conditional on a coordinated agreement on the use of tariffs (here illustrated by the absolute value of bound tariff reductions ?tic).
In this context, it is interesting to note that the Uruguay Round substantially reduced bound MFN tariffs for most signatory countries thereby representing a credible internationally-binding commitment on the use of tariffs. Moreover, the trade round also required its signatory countries to ‘tariffy’ quantitative restrictions and to limit their use in the future.29 The use of NTBs, and in particular of antidumping measures, is however much less regulated. In light of very limited WTO-restrictions on the use of antidumping actions, we hence argue that the Uruguay Round trade agreements may represent a suitable policy setting for a potential substitution effect of declining tariff protection for an enhanced use of productcountry level antidumping measures. We do not consider trade barriers stemming from technical and safety regulations given their rather less precise nature and the prevalent difficulties in finding adequate product level measures.30 Moreover, by focusing on the Uruguay Round we based our empirical examination on an institutional framework in which policy-makers were enabled to credibly commit to binding tariff and quota restrictions. As a result, the policy context we study investigates the relationship between different forms of trade policy, and guided by the theoretical framework illustrated in this section, serves as a vehicle to address the question of trade policy substitution following major, coordinated trade reforms.

Anderson and Schmitt (2003:89) assume “[..] a certain degree of myopia in the cooperative phases. In particular, trade negotiators do not consider how governments may later resort to other policies [..]”. The authors further note that loopholes are in practise mostly closed in subsequent negotiations and only after a substantial amount of violations. 29 Aiming to achieve greater transparency regarding trade restrictions a US-led proposal of ‘tariffication’ was adopted in the UR (Whalley, 1995). Moreover, the UR also terminated the use of Voluntary Export Restraints (VERs) within four years after its conclusion and banned their use in the future. Low and Yeats (2007) consider the latter trade policy constraint as an important achievement or the Uruguay Round. 30 To the best of our knowledge there is no coherent time series data for technical barriers at the HS 8-digit level for the European Union. UN-TRAINS provides this information, at the HS 8-digit level only for the year 2009.

Identification In the previous section we argued that substantial bound tariff concessions and imposed restrictions on the use of quotas may lead to increasing incentives for the use of antidumping protection. To adequately account for the nature of antidumping measures we choose product-country pairs as the unit of our analysis. By focusing on product-country pairs – i.e. the use of antidumping measures targeting particular imported goods from a particular exporting country – we aim to account for the fact that antidumping policies are country-specific implying the existence of exporter-source directed factors which are likely to affect the antidumping process. Our objective is to estimate the impact of the Uruguay Round bound tariff cuts on the probability of subsequent EU antidumping investigations between 1996 and 2008. We use linear, as well as non-linear, binomial panel data modelling techniques and define the dependent variable (Yijt) as an indicator variable taking the value one if the EU initiated an antidumping investigation against a particular product-country pair ij in year t, defining the response probability.

Where Xn,i(j)t represents a vector of n explanatory variables and βn the respective parameter estimates; cij denotes an unobservable individual-specific and time-invariant effect, while Φ(.) describes the underlying distribution function. Analysing product-country pairs in a panel data framework additionally allows accounting for year-specific information which, alongside country-directed variation, is likely to represent an important element when analysing antidumping investigations.

We define the binary dependent variable Yijt at the HS 8-digit product level. Our main explanatory variable is the variable Δti representing the absolute value of the (bound) MFN tariff change negotiated during the Uruguay Round. Based on the conceptual framework presented in section 3, the main theoretical prediction is that the coefficient of Δti is positive 11.

And that the size of the coefficient captures the probability of an antidumping investigation following the Uruguay Round. We additionally introduce a series of product-country and industry control variables captured by the vector Zi(j)t which are not directly based on the conceptual prediction in equation (4) but have been suggested by the relevant literature.31 First, the lagged level of HS 8-digit import flows may represent an important determinant for the initiation of an antidumping investigation given that a higher level of import flows from a particular trading partner may a priori increase the potential for rent destruction and squeezed profit margins for the domestic industry. Second, given that EU law directly refers the investigating authorities to a consideration of a potential increase in allegedly dumped products as well as to an examination of a potentially depressing effect on domestic prices, we also add the lagged growth of product-country specific import flows and unit values, both in percentage terms. Third, to account for a potential retaliatory character of antidumping investigations – i.e. a higher probability of antidumping actions against countries and industries which initiated their own antidumping investigations against EU producers in the past,32 we construct a measure which aims to account for the latter. We define an indicator variable taking the value one if an exporting country’s industry initiated an antidumping procedure at the HS 6-digit product level in the same ISIC 4-digit industry within a time period of five years preceding the EU’s own antidumping investigation. Moreover, accounting for further unobserved industry characteristics which are likely to determine the probability of material injury and thus the finding of dumping, as well as to control for factors such as market-specific demand and supply shocks, we additionally include industry dummies (δn) at the ISIC 4-digit level. Finally, in order to account for unobserved timespecific factors and the possibility that some countries may be more likely to face antidumping investigations we also include year (?t) and exporting-country specific (ϑk) dummies in the model. 33 Our final estimation sample includes 47 countries which were targeted by the EU in 36 separate ISIC 4-digit categories over the time horizon 1996 to 2008. Including product31.

It is worth noting that we introduce all control variables with a one-year lag. Using a slightly lagged expression also limits potential endogeneity concerns due to reverse causality given that past or prospective antidumping measures may contribute to more aggressive tariff liberalisation (cf. the ‘safety valve’ argument). The current scarce literature on the antidumping ‘safety valve’ hypothesis has however not found any empirical evidence for anti-dumping measures making tariff reductions more likely. On the contrary, Moore and Zanardi (2011) find evidence for the opposite (i.e. AD resulting in less tariff liberalisation) when analysing a group of heavily AD-using emerging economies. 32 Bloningen and Bown (2003), and Feinberg and Reynolds (2006), for example, provide evidence for the retaliatory character of antidumping measures. 33 Bown (2010) points out that countries like China, for instance, may be more likely to face antidumping investigations across different products.

country level growth- as well as lagged variables, our empirical analysis is based on a sample of 701,272 importing product-country-year observations and 1061 antidumping investigations. The investigation excludes agricultural products because of the heavy incidence of non-AD measures in that sector.

Data The exact definition of the variables in the empirical analysis and their data sources are presented in Annex Table 1. Summary statistics are provided in Annex Table 2. In this section we describe some of the dataset’s most salient features. We use product-country specific EU antidumping data retrieved from the World Bank’s global antidumping dataset (GAD) for the post-UR time period 1996 to 2008.34 The EU’s HS 8-digit bound Uruguay Round tariff commitments are obtained from the WTO’s schedule of concessions, while the information on product-country specific trade value and quantity import flows are retrieved from the EU’s Comext database. Taking into account the potential retaliatory character of antidumping actions we construct the retaliation indicator variable using antidumping information of countries targeting the EU within a preceding 5-year window. The antidumping data on countries targeting the EU is also from the World Bank’s global antidumping database.35 Finally, in order to link the information on product-country level antidumping investigations from 1996 to 2008 to the EU’s product-level Uruguay Round tariff concessions concordance tables from the EU’s Ramon database have been employed. Using concordance tables and merging the antidumping data with time and product-country level import data results in 1273 antidumping targeted product-country pairs. Including lagged values and introducing the import and unit-value growth variables as additional regressors in the model reduces the number of antidumping targeted observations to 1061.

Using a slightly lagged expression also limits potential endogeneity concerns due to reverse causality given that past or prospective antidumping measures may contribute to more aggressive tariff liberalisation (cf. the ‘safety valve’ argument). 35 The World Bank’s Global Antidumping Dataset (GAD) contains AD information for approximately 47 countries. Moreover, a small fraction of AD investigations used in this study were initiated at the HS 10-digit product level. The latter observations have been transformed into HS-8 digits, providing a potential, although considerably small, bias when estimating the impact of tariff cuts on AD use.

Our estimation strategy is based on three modelling techniques. We first use a linear panel data specification which serves as a benchmark. In spite of representing a valid approximation of the average partial effects (APEs) of our independent variables, using linear estimation techniques for binary response models may also imply certain shortcomings such as nonsensical probability values – i.e. values that lie outside the unit interval (Wooldridge, 2002). We therefore employ two alternative non-linear modelling strategies. Assuming an underlying standard normal distribution function Φ(.) we additionally estimate pooled as well as panel data probit models.36 In order to control for potentially omitted variables at the product-country level and given a restriction on the use of fixed effects estimation techniques due to the time-invariant nature of our main explanatory variable (i.e. the UR bound tariff cuts), we additionally combine the above mentioned modelling techniques with an alternative estimation method suggested by Chamberlain (1980) in the specification of Mundlak (1978). The latter estimation framework is based on the assumption that the time-invariant unobserved effect cij is a function of the means of the time-varying explanatory variables: cij ? α0 ? α1 Zij ? ηij .37 Using the latter property we implement a correlated random effects (CRE) model by introducing a vector of variables consisting of the time means of the time-varying regressors. Mundlak (1978) argues that the introduction of the latter time averages as additional controls explicitly allows for the individual specific effect being correlated to (at least) some elements of Xn,i(j)t. The latter estimation framework allows us to take into account potential unobserved heterogeneity concerns as well as a possible correlation between the individual-specific unobserved components with the, in the model, introduced characteristics. In order to gauge our findings we employ a random-effects (RE) linear as well as two alternative probit modelling techniques and test their sensitivity to the use of the Chamberlain-Mundlak correlated random effects estimation framework.

The most commonly used distributions are the probit and logistic functions which both assume that Φ only takes values between 0 and 1. 37 Note that it is also assumed that the time-constant unobservable features are determined by a conditional normal distribution (i.e. (cij|Zij)~ N(α 0 ? α1 Zij , σ 2 η) .

Omitting the random-effects in the linear model specifications does not change the results. The latter findings are available upon request.

Main Findings Table 2 contains the main findings for the effect of bound MFN tariff concessions on subsequent antidumping investigations at the product-country level. Column (1) presents the estimation results using the linear RE panel data model, while Columns (2) and (3) contain the estimated coefficients and average partial effects (APEs) from the pooled and panel probit specifications, respectively, without controlling for unobserved heterogeneity. Columns (4) to (6) display the results for the latter three techniques including unobserved heterogeneity using the Chamberlain-Mundlak approach. With and without controlling for unobserved time-constant factors the coefficients as well as the estimated average partial effects, for the UR bound tariff variable are positive and statistically highly significant in all model specifications; indicating on average a positive impact of the EU’s Uruguay Round tariff concessions on the probability of subsequent antidumping investigations. Reporting average partial effects that vary between 0.002 and 0.011, our findings are consistent with the theoretical predictions and tend to provide empirical support for the hypothesis of trade policy substitution following the last successfully concluded multilateral trade round. This result stands in contrast to Feinberg and Reynolds (2007) as well as Moore and Zanardi (2011) who, analysing tariff reductions at rather broad industry levels, find either the opposite when analysing filed antidumping petitions in a sample of traditional AD using developed economies (including the EU) or no significant relationship for developed economies at all.39 Analysing the estimation results across different econometric specifications shows that the largest average marginal effects of around 0.010 and 0.011 were reported in the linear probability model (LPM), while the pooled and panel maximum likelihood estimations (MLE) show slightly smaller, but still highly significant, APEs of around 0.009 and 0.002. Controlling for unobserved heterogeneity using the Chamberlain-Mundlak device does not affect the latter results, again indicating robust findings (Table 2, Columns 4 to 6). Despite playing a crucial role in determining the probability of a future EU antidumping investigation.

Feinberg and Reynolds (2007) find an average probability increase of subsequent AD fillings of 0.0042 percentage points per one percent tariff cut when focusing on a sample of 24 WTO-member countries. Looking only at industrial countries the findings of Feinberg and Reynolds (2007) suggest a higher probability for antidumping investigations in industries with rather small UR tariff concessions. Moore and Zanardi (2011) do not find consistent and statistically robust evidence for a trade policy substitution effect in developed countries when analysing the link between sector-level applied tariff and subsequent AD measures. The authors however find, for a group of developing economies, that a one percent increase in tariff cuts increases the average probability of an industry-level AD petition between 0.24% and 0.40%.

From a statistical point of view, the economic impact of the Uruguay Round (bound) tariff reductions tends to be rather limited. A 0.00002 to 0.00011 average percentage point probability increase for a future AD investigation for each one percentage point reduction in bound MFN tariffs indicates a rather small effect, given that the EU’s bound tariffs were reduced by around 2.7 percentage points on average. The very disaggregate focus of our product-country level study may, however, partially explain the limited magnitude of our baseline findings – a hypothesis which tends to be supported by larger coefficients when excluding the country dimension (see section 5.2). The magnitude and statistical significance of the ISIC 4-digit industry dummies provides further interesting insights. They indicate a highly significant influence in oligopolistic sectors such as the textile (in particular ISIC sector 3211 and 3215), the footwear (3240), industrial chemicals (3512, 3513), iron and steel (3710), fertilizers and pesticides (3710) and the fabricated metal and electrical machinery industries (3819 and 3832).40 An important impact of more oligopolistic organised sectors also tends to be highlighted when only focusing on the ISIC 4-digit industries with at least 50 productcountry level antidumping investigations over the considered time horizon. For these regressions, displayed in Annex Table 6, the main explanatory variable coefficients have a higher magnitude (compared to the baseline specification in Table 2). The results also show statistically highly significant and in size larger average marginal effects that vary between 0.019, 0.018 and 0.005 for the linear, non-linear pooled and panel data specifications, respectively, which remain largely unaffected when using the Mundlak-Chamberlain estimation approach (Annex Table 6, Columns 1 to 6). Computing marginal effects which are evaluated at fixed (mean) values of each explanatory variable (MEMs), instead of calculating the average of discrete changes over the whole sample (APEs), provides an alternative method to estimate marginal effects. 41 Given that MEMs may still provide an “asymptotically valid approximation” of average partial effects (APEs) (Greene, 1997:876), we additionally estimate equation (6) with MEMs. The results are reported in Annex Table 5 and corroborate the earlier findings by showing the same values for the linear model estimations and slighter smaller marginal.

Note that these results are not reported in Table 2, but are available upon request. To the best of our knowledge, the existing literature does not clearly favour one estimation method over the other. Some studies favour APEs over MEMs in particular in the presence of dummy variables (Long, 1997; Greene, 1997). Long (1997), for instance, points out that the presence of indicator variables among the explanatory variable may make the computation of MEM refer to inherently nonsensical observations.

Effects at mean values (MEMs) of 0.002 for the pooled and panel MLE regressions compared to the average partial effects (APEs) displayed in Table 2. The signs and significance of the remaining control variables in Table 2 are generally in line with the literature on the factors determining antidumping investigations. The retaliation indicator variable shows positive parameter estimates and average partial effects in all model specifications presented in Table 2, which are, however, only significant when using probit estimation techniques.42 Average partial effects of around 0.002 to 0.001, when estimated with pooled or panel MLE, may therefore point to a higher probability of EU antidumping investigations against imports from trading partners whose industries had previously been targeted by EU exporters in the same ISIC 4-digit industry (Table 2, Columns 2 and 3). The latter result may therefore, under certain circumstances, point to some evidence of a possible retaliatory character of EU antidumping protection.43 The lagged value of product-country specific imports is also shown to have estimated coefficients which are consistent with the theoretical predictions. They show a positive and statistically significant impact on the propensity of subsequent antidumping investigations in all model specifications. This may indicate that products which tend to be exposed to a high degree of foreign import competition are more likely to be protected by an EU antidumping investigation. While the linear model specifications show the largest APE approximations for the import value variable of 0.118 and 0.075 (Table 2, Columns 1 and 4, respectively), the pooled and panel probit models show considerably smaller values which lie between 0.020 and 0.002. The effect of the lagged growth of product-country specific imports shows mixed results which are, however, not significant at the usual levels. Positive parameter estimates can only be reported for the estimations using the pooled or panel probit model and controlling for unobserved heterogeneity. As a result, our findings tend to provide no support for the argument that the EU’s antidumping authorities are more likely to launch an antidumping investigation against products from partner countries with preceding import growth.44 Analysing the effect of the lagged unit value change, used as a proxy for domestic price evolutions which are likely to play an important role according to the EU’s regulatory.

Note that the industry retaliation indicator have positive coefficients and APEs, which are however not significant at the usual levels when using a linear panel estimation approach (Table 2, Columns 1 and 4). 43 Note that the retaliation results remain the same when using Chamberlain’s model modifications (Table 2, Columns 5 and 6). 44 Using two or three-year lags or lagged averages over 2 and 3 years for the product-country level import growth variable results in qualitatively identical findings. The results are available upon request.

Investigation framework, shows a negative sign for the average partial effect calculations. These are however only statistically significant when using a panel data estimation technique with or without controlling for unobserved heterogeneity.45 Our estimation results may thus partially point to a significant effect of declining prices on the probability of an EU antidumping investigation.

Despite reporting negative parameter estimates and average partial effects, the latter are not significant at the usual levels when using a pooled probit estimation approach.

Table 2: The Impact Uruguay Round Bound Tariff Concessions on subsequent Antidumping Measures: Average Partial Effects (1) (2) (3) (4) (5) Chamberlain's Chamberlain's RE Model Linear Probit Probit Linear RE Probit RE Estimation Method UR (bound) tariff cuts Industry Retaliation Indicator Import Value.

Notes: Standard errors are in parentheses below all coefficients or Average Partial Effects (APEs). *, **, *** respectively denote the 10%, 5%, 1% significance levels. ζ and ξ indicates that the respective variable has been re-scaled by 10,000,000 and 1,000 respectively. For the linear RE model, pooled probit, and Chamberlain’s pooled MLE estimation, the serial-correlation robust standard errors were computed by using clustering at the product-country level. Due to prohibitive estimation times standard errors using bootstrapping estimation techniques were not computed for the panel data MLE estimations (column 6). The superscript (a) at the bottom of the table indicates that the p-values reported for the linear probability models are based on F-tests.

Robustness Tests and Additional Specifications We examine the robustness of the results by conducting a series of sensitivity tests using alternative estimation techniques and additional model specifications. Table 3 presents a summary of these findings. First, we introduce an additional explanatory variable to test the sensitivity of the results to the inclusion of the lagged level of applied MFN tariffs in each year. Since bound tariffs reflect the highest MFN rate possible and thus a ceiling value agreed upon in multilateral – i.e. coordinated – tariff negotiations, focusing on the latter provides a consistent framework for our theoretical predictions set out in section 3. Nevertheless, including applied MFN tariffs as an alternative trade liberalization measure may provide some additional insights on the impact of existing tariff protection.46 Columns (1) to (3) in Table 3 show the computed average marginal effects for all three modelling techniques used in the previous section including the lagged level of applied MFN tariffs. Columns (1) to (3) confirm the main results of Table 2 by showing a positive effect of applied as well UR bound tariff rates, the latter however at a slightly lower significance level.47 Additionally introducing the (lagged) difference between the upper bound and applied tariff rates in percent of the applied MFN rate as a further control variable in the model, which may reflect the EU’s flexibility to increase the (applied) MFN tariff in order to provide additional import protection, results in qualitatively similar results for the applied as well as the bound tariff rates when estimated with linear or non-linear probit estimation techniques (Columns 4 to 6, Table 2). The estimation results for the tariff overhang variable report positive coefficients which are however only significant when estimated with the linear model. This result is rather surprising as it indicates a lower probability for an antidumping investigation where bound-applied tariff margins are small.

It should be noted that, in contrast to developing countries, the difference between the latter two types of tariffs is in the case of developed economies generally rather small. 47 The results for the lagged applied MFN tariff rate indicate that products with higher applied MFN tariffs in the previous year also are more likely to be protected by antidumping actions. These results are however not surprising if we assume that the use of some kind of formulaic approach, in the UR, led to larger tariff cuts on initially high bound (and also applied) MFN tariffs thereby reducing but not eliminating the within-industry variation of tariff rates. The positive correlation between applied and bound tariff rates, in particular in the case of developed economies may then explain why products with higher applied MFN tariffs may also have been subjected to larger bound cuts. Larger cuts for initially high tariff rates as well as similar inter-industry distribution of pre- and post-UR tariff levels tend to be confirmed by a visual inspection of the inter-industry tariff distribution (graphs available upon request). Due to the presumably strong correlation between the URnegotiated tariff cuts and applied MFN tariffs, the latter specification represents a suitable robustness check but is not our preferred model specification. 48 Omitting the lagged level of applied tariffs in this specification results in an even more significant effect of the UR tariff cuts but renders the impact of the tariff overhang variable insignificant in all estimations.

Moreover, we also perform an additional robustness test by following Feinberg and Reynolds (2007) and inter-acting the EU’s negotiated (UR) tariff reductions with a log trend variable in order to proxy the impact of the tariff cuts over time.49 They allow for the fact that the agreed tariff cuts were phased in over several years. Given a potential time lag for the tariff reductions to result in an increased import competition and injury for the domestic industry as well as the possibility of industries adjusting to increased competition over time, we are controlling for any time-varying impact of the negotiated tariff concessions. While the regressions in Column (7) are based on a random-effects probit panel data model, the estimations results in Columns (8) and (9) use fixed-effects logit and linear panel data regression techniques. The results, displayed in Table 3, Columns (7) to (9), tend to corroborate previous findings by showing a highly significant impact of the EU’s UR tariff concessions on the probability of future antidumping measures in all model specifications. Finally, we also test the EU’s antidumping bound MFN tariff relationship by dropping the country dimension and transforming the dataset into an HS 8-digit product-level panel. Examining the product-level link between EU tariff cuts and the subsequent use of alternative forms of import protection may provide further support for trade policy substitution following the Uruguay Round. The results are displayed in Annex Table 7 and confirm previous results in favour for a significant positive relationship between the size of the EU’s bound external tariff commitments and the likelihood of a subsequent anti-dumping investigation in all model specifications. The results show average partial effects that vary between 0.029 and 0.052. The signs and significance of the control variables are also broadly in line with the earlier product-country level findings.

Suggests significant inter-correlation when including the applied tariff level and the tariff overhang variable in the same specification. Moreover, dropping the UR tariff cut variable and only focusing on the tariff overhang variable leads to non-significant results in a fixed effects linear probability model specification. 49 Feinberg and Reynolds (2007:953) introduce this interaction “in order to capture the change in the impact of the reductions over time. For example, one would expect there to be a lag in the impact of tariff reductions on antidumping filings both because the reductions were phased in between 1996 and 1999 and because of the time it would take for industries injured by tariff reductions to file a petition. One might also expect that the impact of the tariff reduction may diminish over time as industries adjust to the new, lower tariff rates.”

Table 3: Robustness Analysis: EU bound tariff concessions and antidumping investigations: Average Partial Effects (1) (2) (3) (4) (5) Model Linear Probit Probit Linear Probit Estimation Method
RE Coefficient Pooled MLE APE 0.006* (0.003) 0.013*** (0.002) 0.002*** (0.001) 0.019*** (0.004) 0.087** (0.039) -0.129 (0.081) MLE APE 0.004* (0.002) 0.009*** (0.002) 0.001*** (0.0004) 0.014*** (0.003) 0.062 (0.088) -0.092*** (0.034) RE Coefficient 0.011** (0.006) 0.0002*** (0.00003) 0.0004** (0.0002) 0.005 (0.003) 0.285*** (0.071) -0.003 (0.035) 0.00001 (0.00001) Pooled MLE APE 0.008* (0.004) 0.0002*** (0.00002) 0.0003 (0.0002) 0.002*** (0.001) 0.045*** (0.009) 0.111 (0.071) -0.229** (0.107)
(6) Probit
MLE APE 0.006* (0.003) 0.0001*** (0.00002) 0.0002 (0.0002) 0.002*** (0.001) 0.033*** (0.005) 0.079 (0.095) -0.167*** (0.055)
(7) Probit(RE)
MLE APE 0.001*** (0.0004) 0.001*** (0.0002) 0.006*** (0.001) -0.007 (0.033) -0.024** (0.010)
(9) Logit(FE)
MLE APE 1.185*** (0.404) 3.868** (1.377) 1.271 (32.960) -2.949 (4.079 )
(10) Linear
FE Coefficient 0.012** (0.005) 0.067*** (0.015) -0.034 (0.137) -6.57*10-7 (0.0002)

UR (bound) tariff cuts UR (bound) tariff cuts*ln(T) Applied MFN tariff Tariff Overhang Industry Retaliation Indicator Import Valueζ Import Value Growthζ Unit Value Growthξ Mundlak Transformations Observations Log likelihood Pseudo R-squared Wald(chi2) p-value
0.007* (0.004) 0.015*** (0.002) 0.003 (0.002) 0.10831*** (0.02835) 0.009 (0.037) -0.00002* (0.00001)
No No No Yes Yes Yes No No No 575365 509737 575365 483635 406263 483635 701272 9445 701272 -5113.95 -5109.95 -4280.42 -4278.75 -6252.22 -2034.62 0.19 0.19 0.20 0.20 0.003 0.10 0.000(a) 0.000 0.000 0.000(a) 0.000 0.000 0.000 0.000 0.000(a) Notes: Standard errors are in parentheses below all coefficients or Average Partial Effects (APEs). *, **, *** respectively denote the 10%, 5%, 1% significance levels. All regressions in Table 3 include timespecific fixed effects. The results displayed in the columns (1) to (6) are based on estimations which include exporting country as well as (ISIC 4-digit) industry-level dummies. For the linear RE models and the pooled probit serial-correlation robust standard errors were computed by using clustering at the product-country level. Due to prohibitive estimation times standard errors using bootstrapping estimation techniques were not computed for the panel data MLE estimations (columns 3 and 6). Columns (7) to (9) report the estimation results when interacting the UR tariff concessions with a log trend variable in order to proxy the latter’s over time changing impact (i.e. phasing-in schedules). ζ and ξ indicates that the respective variable has been re-scaled by 10,000,000 and 1,000 respectively. The superscript (a) at the bottom of the table indicates that the p-values reported for the linear probability models are based on F-tests.

Using data at the product-country level, this paper examines the impact of bound MFN tariff cuts, negotiated during the Uruguay Round, on subsequent EU anti-dumping actions. Our findings tend to provide empirical evidence for Anderson and Schmitt’s (2003) theoretical contribution, which predicts an enhanced use of antidumping protection following coordinated (e.g. multilateral) tariff liberalization when coupled with an additional constraint on the use of quantitative protectionist measures. Based on a persisting incentive to alter the terms of trade, Anderson and Schmitt’s theoretical framework shows that policy-makers are more likely to resort to alternative WTO-permitted forms of import protection following major multilateral trade reforms. In light of the Uruguay Round’s tariff commitments as well as its restrictions on the future use of quotas, we consider the latter to represent a suitable testing ground for the above theory. We test the effect of the UR negotiated tariff reductions on subsequent product-country level EU anti-dumping investigations initiated between 1996 and 2008. Our results point to a substitution of different forms of trade policy instruments following the Uruguay Round, with a persistent and statistically highly significant impact of bound MFN tariff concessions on the probability of future antidumping investigations being identified. The average partial effects vary between 0.010 and 0.002, which indicates a positive, but quantitatively limited, probability increase of up to 0.00010 percentage points per one percentage point tariff reduction agreed upon during the UR. Our findings stand in contrast to previous studies in the literature. These studies either find no statistically viable relationship between tariff liberalization and antidumping measures for developed economies, or even a weak negative link between the latter two forms of import protection as shown by Moore and Zanardi (2011) and Feinberg and Reynolds (2007), respectively.50 Common to both of the latter two studies is that they focus on rather broad industry-level tariff and antidumping actions. However, antidumping measures are, in general, imposed on a very disaggregated product-level, giving rise to a potential bias. Interestingly, our results tend to corroborate those of Bown and Tovar (2011) who analyse HS 6-digit product-level tariff cuts in India and find evidence for tariffs being substituted by an enhanced use of antidumping and safeguard measures.

Note that Moore and Zanardi (2011) focus on sectoral applied and not on bound MFN tariff changes between 1991 and 2001. Feinberg and Reynolds (2007) find a significant positive link between bound MFN tariff cuts and subsequent antidumping actions mainly for developing countries and a rather weak negative relationship for traditional AD using countries (i.e. Australia, Canada, EU, New Zealand, and the USA).

To sum up, our findings tend to point to a substitution effect of bound MFN tariff protection for more antidumping investigations in the European Union. Our results show that coordinated tariff liberalization is partially reversed by the use of alternative trade remedy instruments, and caution against only focusing on tariffs as a country’s trade liberalization indicator. Moreover, our empirical analysis also provides evidence that the substitution of tariffs for more antidumping protection is an aspect of trade policy in emerging economies, as indicated in previous studies, but is also to be found in developed economies.

Anderson, S. P., Schmitt N., 2003, Nontariff barriers and trade liberalization, Economic Inquiry, 41, 80–97. Anderson, J. E., Zanardi M., 2009, Political pressure deflection, Public Choice, 141, 129–50. Bagwell, K., Staiger R. W., 1999, An economic theory of the GATT, American Economic Review, 89, 215-248. Baldwin, R. E., Steagall J.W., 1994, An analysis of ITC decisions in antidumping, countervailing duty and safeguard cases, Review of World Economics, 130, 290-308. Blonigen, B. A., Prusa T. J., 2003, Antidumping’, in E.K. Choi and J. Harrigan (eds.), Handbook of International Trade (Oxford, UK and Cambridge, MA: Blackwell Publishers), 251-284. Blonigen, B. A., Bown C.P., 2003, Antidumping and retaliation threats, Journal of International Economics, 60, 249–273. Brander, J., Krugman P., 1983, Reciprocal dumping model of international trade, Journal of International Economics, 15, 313–321. Bown, C. P., 2008, The WTO and antidumping in developing countries, Economics and Politics, 20, 255-288. Bown, C. P., 2012, Global antidumping database, available at Bown, C. P., 2010, China’s WTO Entry: Antidumping, safeguards, and dispute settlement, in Robert C. Feenstra and Shang-Jin Wei (Eds.), China’s Growing Role in World Trade, University of Chicago Press for the NBER, 281-337. Bown, C. P., Tovar, P., 2011, Trade liberalization, antidumping, and safeguards: Evidence from India's tariff reform, Journal of Development Economics, 96, 115-125. Chamberlain, G., 1980, Analysis of covariance with qualitative data, Review of Economic Studies, 47, 225-238. Feenstra, R.C., Lewis, T.R., 1991, Negotiated trade restrictions with political pressure, Quarterly Journal of Economics, 106, 1287-1307. Feinberg, R. M., Reynolds K.M, 2006, The spread of antidumping regimes and the role of retaliation in filings, Southern Economic Journal, 72, 877–890. Feinberg, R.M., Reynolds K.M., 2007, Tariff liberalisation and increased administrative protection: Is there a quid pro quo?, The World Economy, 30, 948-961. Finger, M., Olechowski A., 1987, The Uruguay Round: A Handbook for the Multilateral Trade Negotiations, The World Bank, Washington D.C. Greene, W. H., 1997, Econometric Analysis, 3rd ed., Upper Saddle River, NJ: Prentice Hall. Grossman, G., Helpman, E., 1994, Protection for sale, American Economic Review, 84, 833-850. Hansen, W. L., Prusa T.J., 1996, Cumulation and ITC decision-making: The sum of the parts is greater than the whole, Economic Inquiry, 34, 746-69. Hillman, A.L., Ursprung, H., 1988, Domestic politics, foreign interests, and international trade policy, American Economic Review, 78: 729-745. Hillman, A.L., 1990, Protectionist policy as the regulation of international industry, Public Choice, 67, 101-110.

Hwang, H., Mai C., 1988, On the equivalence of tariffs and quotas under duopoly, Journal of International Economics, 24, 373-380. Konings, J., Vandenbussche, H., 2005, Antidumping protection and markups of domestic firms, Journal of International Economics, 65, 151-65. Limão, N., Tovar, P., 2011, Policy choice: Theory and evidence from commitment via international trade agreements, Journal of International Economics, 85, 186-205. Lindsay, B., Ikenson D., 2001, Coming Home to Roost: Proliferating Antidumping Laws and the Growing Threat to US Exports, Cato Institute, Washington D.C. Liu, X., Vandenbussche H., 2002, European Union anti- dumping cases against China: An overview and future prospects with respect to China’s World Trade Organization membership, Journal of World Trade, 36, 1125–44. Long, J. S., 1997, Regression Models for Categorical and Limited Dependent Variables, Thousand Oaks, CA: Sage. Low, P., Yeats A. J., 2007, Non-tariff measures and developing countries: Has the Uruguay Round levelled the playing field, The World Economy, 18, 51–70. Macrory, P., Vermulst E., Waer P., 1991, United States and European community antidumping law: Similarities and differences, University of Miami Yearbook of International Law, 74, 74-142. Maggi, G., Rodriguez-Clare A., 1998, The value of trade agreements in the presence of political pressures, Journal of Political Economy, 106, 574-601. Messerlin, P., Reed G., 1995, Antidumping policy in the United States and the European Communitiy, Economic Journal, 105, 1565-75. Moore, M.O., 1992, Rules or Politics?: An empirical analysis of ITC antidumping decisions, Economic Inquiry, 30, 449-466. Moore, M., Zanardi M., 2011, Trade liberalization and antidumping: Is there a substitution effect?, Review of Development Economics, 15, 601–619. Mundlak, Y., 1978, On the pooling of time series and cross section data, Econometrica, 46, 69-85. Nelson, D., 2006, The political economy of antidumping: A survey, European Journal of Political Economy, 22, 554-590. Peltzman, S., 1976, Toward a more general theory of regulation, Journal of Law and Economics, 19, 211-240 Prusa, T. A., 2001, On the spread and impact of antidumping, Canadian Journal of Economics, 34(3), 591–611. Prusa, T. A., S. Skeath, 2004, Modern commercial policy: Managed trade or retaliation?, in E. K. Choi and J. C. Hartigan (eds.), The Handbook of International Trade. Vol. II: Economic and Legal Analysis of Trade Policy and Institutions (Oxford: Blackwell). Prusa, T. A., 2005, Antidumping: A growing problem in international trade, The World Economy, 28, 683–700. Prusa, T. J., Vermulst E. A., 2011, United States-continued existence and application of zeroing methodology: The end of zeroing? World Trade Review, 10, 45-61. Renner, J.C., 1971, National restrictions on international trade, In: United States International Economic Policy in an Interdependent World, vol.1, Commission on International Trade and Investment Policy, Washington D.C: US Government Printing Office, 663-675.

Rovegno, L., Vandenbusche H., 2011, A Comparative Analysis of EU Antidumping Rules and Application, In Liberalising Trade in the EU and the WTO: Comparative Perspectives, Cambridge University Press. Tharakan, P., Greenaway D., and Tharakan J., 1998, Cumulation and injury determination of the European Community in antidumping cases, Review of World Economics, 134, 320-39. Stigler, G.J., 1971, The theory of economic regulation, Bell Journal of Economics, 4, 295-298. Vandenbussche, H., Zanardi M., 2010, The chilling trade effects of antidumping proliferation, European Economic Review, 54,760–777. Whalley J., 1995, Developing countries and system strengthening in the Uruguay Round, In Martin W., Winters L.A., (Ed.), The Uruguay Round and Developing Economies, The World Bank Washington, D.C. Wooldridge, J.M., 2002, Econometric Analysis of Cross Section and Panel Data, The MIT Press, Cambridge, MA. Zanardi, M., 2004, Antidumping: What are the numbers to discuss at Doha? The World Economy, 27, 403–433.

Variable Abbreviation Exact definition Dependent variable Anti-dumping indicator variable adijt Indicator variable that equals one if the EU initiated an antidumping investigation on a particular productcountry pair ij Main Explanatory variable Bound MFN tariff rate reductions ?ti Bound ‘Most Favoured Nation’ (MFN) tariff reductions agreed upon during the Uruguay Round Control Variables Import trade value impijt Trade value of HS 8-digit import flows in 1000 ecu Indicator variable which takes the value one if the foreign investigated ISIC 4-digit industry had filed an antidumping investigation against exports from a European Member State in the same sector during the past 5 years HS 8-digit Import trade value change in 1000ecu COMEXT WTO + authors’ own calculations Bown (2012) Source.

HS 8-digit unit value changes calculated as product level import value over import quantities Interaction between a log trend variable and the bound MFN rate UR tariff concessions at the HS 8-digit product level (i.e. UR-reduction*ln(T), where the year 1995 represents t=1)(a) HS 8-digit product-level difference between bound MFN and applied MFN tariff rates in percent of the applied rate.

Final imposed punitive (ad-valorem) tariff duty at the 8digit HS product level (preliminary imposed duties were used when the final duties were missing).

Notes: (a) Defining a UR concession trend variable follows the estimation approach chosen by Feinberg and Reynolds (2007:953) and aims at capturing “the change in the impact of the reductions over time.” Using the year 1994 as t=1 (and thus 1996 as t=3), results in qualitatively identical findings.

Variable Antidumping indicator Uruguay Round tariff change Industry retaliation indicator Import value Import value growth Unit value growth UR concession trend variable Tariff overhang Applied tariffs Imposed AD duties Mean 0.002 0.027 0.003 0.001 2.03*10-6 0.002 0.052 -0.234 0.050 29.16 Std. Dev. 0.039 0.020 0.051 0.005 0.0004 0.289 0.042 0.324 0.038 19.64 Min 0.000 0.000 0.000 0.000 -1.00*10-7 -0.001 0.000 -1.000 0.000 0.000 Max 1.000 0.268 1.000 0.808 0.253 148.59 0.401 3.333 0.406 96.80.
Notes: The summary statistics of the explanatory variables used in the main specifications estimated in table 2 are based on a sample of 701,272 observations. Including the applied tariff rate reduces the sample to 575,365 year product-country observations. The summary statistics of the latter variable and the tariff overhang are thus based on a slightly smaller dataset. It is further worth noting that the import value and import value growth variables have been re-scaled by 10,000,000, whereas the unit value-growth variable has been re-scaled by a factor of 1,000.

Annex Table 3: Descriptive Statistics - European Antidumping Investigations between 1996 and 2008: Targeted Countries
Antidumping Investigations (1) Prod.-Country Pairs (2) Product Lines (3) Targeted Products (4) Targeted Prod.Country Pairs.
No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47

5.823 1.783 1 Algeria 1 1.518 690 1 Armenia 1 31.347 4.319 2 Australia 2 13.481 2.941 9 Belarus 9 11.708 2.868 3 Bosnia Herzegovina 3 34.006 4.514 5 Brazil 5 24.114 4.007 13 Bulgaria 13 42.811 4.801 0 Canada 0 50.899 5.037 280 China 305 27.788 4.127 14 Croatia 19 30.048 4.697 18 Czech Republic 18 19.646 3.553 36 Egypt 50 15.975 3.514 4 Estonia 4 34.357 4.303 2 Hong Kong 2 26.557 4.534 13 Hungary 14 42.708 4.805 113 India 131 26.821 3.892 29 Indonesia 43 10.969 2.639 10 Iran 10 50.358 4.949 20 Japan 26 4.508 1.504 4 Kazakhstan 4 9.843 2.784 2 Latvia 2 2.525 962 13 Libya 13 11.899 3.080 11 Lithuania 11 5.032 1.449 0 Macao 0 10.510 2.713 2 Macedonia 2 27.048 3.871 28 Malaysia 36 27.294 4.088 3 Mexico 3 5.728 1.910 8 Moldova 8 14.484 2.749 21 Pakistan 40 17.608 3.062 7 Philippines 7 28.958 4.692 15 Poland 15 28.183 4.229 20 Romania 25 35.213 4.567 46 Russia 51 14.407 2.977 2 Saudi Arabia 2 27.937 3.988 1 Singapore 1 20.526 3.936 21 Slovakia 21 22.177 4.054 2 Slovenia 2 33.597 4.490 17 South Africa 17 39.043 4.561 53 South Korea 60 39.466 4.482 55 Taiwan 63 31.158 4.116 29 Thailand 36 43.006 4.803 51 Turkey 67 60.686 5.213 34 USA 34 23.263 3.971 36 Ukraine 41 2.392 945 0 Uzbekistan 0 14.843 2.845 46 Vietnam 46 12.453 3.080 10 Yugoslavia 10 All Manufacturing Industries 1.114.721 7353 1111 1273 Notes: The above statistics are based on the author's own calculations using product-country level import data form Comext and antidumping data from the World Bank’s global antidumping database. Column (1) reports the total number of imports per country over the total time horizon, while Column (2) denotes the number of different products imported by the EU from the respective partner country. Column (3) shows the number of different products subjected to an AD investigation and column (4) reports the total number of observations characterized by an antidumping investigation.

Annex Table 4: Descriptive Statistics - European Antidumping Decisions between 1996 and 2008
Antidumping Investigations (1) (2) Targeted Targeted Prod.Products Country Pairs 8 0 0 39 0 11 35 4 0 2 0 41 5 1 0 0 15 0 0 9 94 4 22 9 26 6 1 8 340 8 0 0 227 0 11 97 4 0 14 0 103 7 1 0 0 29 0 0 12 317 8 106 11 68 12 1 25 1061 Preliminary AD Duties (3) (4) Product Average lines Duties 4 0 0 124 0 5 45 3 0 14 0 60 3 0 0 0 3 0 0 6 151 5 84 4 19 4 0 5 539 23.7 28.8 48.3 17.8 41.6 18.2 27.9 22.8 66.1 33.1 34 37 50.1 32 24.6 34.7 33.8 Final AD Duties (5) (6) Product Average lines Duties 2 0 0 24 0 5 45 3 0 12 0 56 0 0 0 0 8 0 0 6 196 6 95 11 21 7 0 2 499 12.9 25.5 46.4 12.5 47.6 15.8 26.5 22.2 39.9 30.6 25.1 33.4 48 51.1 25.1 34.1 31
ISIC Code 311 313 314 321 322 323 324 331 332 341 342 351 352 353 354 355 356 361 362 369 371 372 381 382 383 384 385 390
Industry Food Products Beverages Tobacco Textiles Wearing apparel Leather products Footwear except rubber Wood products Furniture except metal Paper and products Printing and publishing Manufacture of industrial chemicals Other chemicals Petroleum refineries Misc. Petroleum and coal Rubber products Plastic products Pottery china earthenware Glass and products Other non-metalic mineral Iron and Steel Non-ferrous metals Fabricated metal products Machinery except electrical Machinery electrical Transport equipment Professional and scientific Other manufactured All Manufacturing Industries.

Notes: The above statistics are based on the estimating sample of 701,272 observations including 1061 antidumping targeted product-country pairs. The antidumping data stems from Bown (2012). Columns (1) and (2) display the number of targeted products and product-country pairs per ISIC 3-digit industry, while columns (3) to (6) illustrates the number of product lines that were subject to a preliminary or final antidumping tax, including the respective average of the punitive import duties.
 
Annex Table 5: Uruguay Round Tariff Concessions and EU Antidumping Investigations: Marginal Effects at Mean Values (MEM)
(1) Model Linear RE Estimation Method Coefficient (2) Probit Pooled MLE Coefficient MEM (3) Probit MLE Coefficient MEM (4) Chamberlain's Linear RE RE Coefficient (5) Chamberlain's RE Probit Pooled MLE Coefficient MEM (6) Chamberlain's RE Probit MLE Coefficient MEM
UR (bound) tariff cuts Industry Retaliation Indicator Import Valueζ Import Value Growth Unit Value Growthξ Constant Industry FE Year and Country Dummies Mundlak Transformations Observations Log likelihood Pseudo R-squared Wald(chi2) p-value:
0.010*** (0.004) 0.003 (0.002) 0.118*** (0.029) -0.033 (0.047) -0.00002* (0.00001) 0.002** (0.001) Yes Yes No 701272 0.000(a)
1.829*** (0.572) 0.379*** (0.107) 4.295*** (0.965) -4.049 (8.439) -21.447 (13.152) -3.276*** (0.336) Yes Yes No 634378 -6401.87 0.19 0.000
0.002*** (0.001) 0.001** (0.0004) 0.004*** (0.001) -0.004 (0.008) -0.020* (0.011) Yes Yes No 634378 -6395.88 0.19 0.000
2.030*** (0.729) 0.492*** (0.143) 5.417*** (0.850) -6.857 (31.762) -23.388*** (9.031) -4.050*** (0.472) Yes Yes No 701272 -6264.13 0.18 0.000
0.002*** (0.001) 0.005*** (0.001) 0.0004*** (0.0001) -0.006 (0.028) -0.021*** (0.008) Yes Yes No 701272 -6254.87 0.18 0.000
0.011*** (0.004) 0.003 (0.002) 0.075*** (0.029) -0.017 (0.037) -0.00002* (0.00001) 0.001* (0.001) Yes Yes Yes 701272 0.000(a)
1.901*** (0.571) 0.382*** (0.107) 1.831*** (0.487) 10.734 (8.035) -21.473 (13.207) -3.284*** (0.336) Yes Yes Yes 634378 -6386.24 0.19 0.000
0.002*** (0.001) 0.001** (0.0004) 0.002*** (0.001) 0.010 (0.007) -0.019* (0.011) Yes Yes Yes 634378 -6380.44 0.19 0.000
2.135*** (0.730) 0.496*** (0.143) 2.126* (1.123) 10.534 (31.858) -23.220** (9.039) -4.062*** (0.474) Yes Yes Yes 701272 -6248.26 0.18 0.000
0.002*** (0.001) 0.002* (0.001) 0.001*** (0.0001) 0.010 (0.029) -0.021*** (0.008) Yes Yes Yes 701272 -6239.34 0.18 0.000
Notes: Annex Table 5 reports the estimated parameter estimates and marginal effects evaluated at the mean value of the explanatory variables. Standard errors are reported in the parentheses below the estimated coefficients and calculated marginal effects (MEMs). *, **, *** illustrate the 10%, 5%, 1% significance levels, respectively. ζ and ξ indicates that the respective variable has been rescaled by 10,000,000 and 1,000 respectively. For the linear RE model, pooled probit, and Chamberlain’s RE probit estimated by pooled MLE, the serial-correlation robust standard errors were computed by using clustering at the product-country level. Due to prohibitive estimation times standard errors using bootstrapping estimation techniques were not computed for the APE of the MLE probit model in column (6). (a) The reported p-value for the linear probability models is based on an F-test.

Annex Table 6: Tariff Concessions and Antidumping Measures - Most Targeted Industries: Average Partial Effects
(1) Model Linear (2) Probit (3) Probit (4) Chamberlain's Linear RE MLE Coefficient 2.873*** (0.799) 5.887*** (1.080) 0.501*** (0.147) -10206 (37.193) -20.828** (9428) -3.539*** (0.475) Yes Yes No 359316 -5204.69 0.01 0.000 APE 0.005*** (0.001) 0.010*** (0.002) 0.001*** (0.0003) -0.017 (0.061) -0.034** (0.016) Yes Yes No 359316 -5204.69 0.01 0.000 RE Coefficient 0.019*** (0.006) 0.127** (0.065) 0.002 (0.002) -0.008 (0.045) -0.00003 (0.00004 0.005** (0.002) Yes Yes Yes 359316 (5) Chamberlain's RE Probit Pooled MLE Coefficient 2.534*** (0.621) 2.204*** (0.638) 0.389*** (0.107) 35.948*** (13.106) -18641 (13.560) -2.762*** (0.331) Yes Yes Yes 331341 -5320.17 0.16 0.000 APE 0.019*** (0.005) 0.017*** (0.005) 0.003*** (0.001) 0.269*** (0.099) -0.14 (0.102) Yes Yes Yes 331341 -5320.17 0.16 0.000 (6) Chamberlain's RE Probit MLE Coefficient 3.028*** (0.802) 2.570* (1.460) 0.513*** (0.148) 52603 (43.699) -20.055** (9.400) -3.556*** (0.478) Yes Yes Yes 359316 -5180.51 0.01 0.000 APE 0.005*** (0.002) 0.004* (0.003) 0.001*** (0.0003) 0.089 (0.075) -0.034** (0.017) Yes Yes Yes 359316 -5180.51 0.01 0.000.

Notes: Standard errors are in parentheses below all coefficients or average partial effects (APEs). *, **, *** respectively denote the 10%, 5%, 1% significance levels. ζ and ξ indicates that the respective variable has been re-scaled by 10,000,000 and 1,000 respectively. For the above regressions only industries with at least 50 product-country level AD investigations have been considered. For the linear RE model, pooled probit, and Chamberlain’s pooled probit model, the serial-correlation robust standard errors were computed by using clustering at the product-country level. Due to prohibitive estimation times standard errors using bootstrapping estimation techniques were not computed for the APEs of the MLE probit model (column 6). (a) The reported p-value for the linear probability models is based on an Ftest.

Annex Table 7: Tariff Concessions and Antidumping Measures - Product Level Analysis: Average Partial Effects
(1) Model Linear (2) Probit (3) Probit (4) Chamberlain' s Linear RE RE APE 0.030** (0.013) 0.004*** (0.001) 0.016*** (0.005) -0.00002 (0.00004) -0.002** (0.001) Yes Yes No 50890 -1747.47 0.33 0.000 Coefficient 0.052** (0.026) 0.016*** (0.006) 0.0001 (0.027) 0.000 (0.000) -0.00001 (0.00001) 0.006*** (0.002) Yes Yes Yes 50890 (5) Chamberlain's RE Probit (6) Chamberlain's RE Probit.

Pooled MLE Coefficient 2.666** (1.178) 0.321*** (0.093) -0.236 (0.318) -0.001 (0.002) -0.142* (0.076) -2.606*** (0.152) Yes Yes Yes 46782 -1756.77 0.15 0.000 APE 0.049** (0.022) 0.006*** (0.002) -0.004 (0.006) -0.00002 (0.00004) -0.003* (0.001) Yes Yes Yes 46782 -1756.77 0.15 0.000.
MLE Coefficient 2.803** (1.226) 0.367*** (0.098) -0.295 (0.749) -0.001 (0.003) -0.156*** (0.059) -2.847*** (0.189) Yes Yes Yes 50890 -1740.51 0.33 0.000 APE 0.029** (0.013) 0.004*** (0.001) -0.003 (0.008) -0.00001 (0.00003) -0.002** (0.001) Yes Yes Yes 50890 -1740.51 0.33 0.000.
Notes: Standard errors are in parentheses below all coefficients or Average Partial Effects (APEs). *, **, *** respectively denote the 10%, 5%, 1% significance levels. ζ and ξ indicates that the respective variable has been re-scaled by 10,000,000 and 1,000 respectively. For the linear RE model, pooled probit, and Chamberlain’s pooled probit model, the serial-correlation robust standard errors were computed by using clustering at the product-country level. Due to prohibitive estimation times standard errors using bootstrapping estimation techniques were not computed for the APE of the pooled probit MLE model (column 6). (a) The reported p-value for the linear probability models is based on an F-test.