How Different are Safeguards from Antidumping? Evidence from US Trade Policies Toward Steel

Transcription

1 How Different are Safeguards from Antidumping? Evidence from US Trade Policies Toward Steel Chad P. Bown, Department of Economics and International Business School Brandeis University July 2004 Abstract How do the trade impacts of a safeguard measure - which is statutorily designed to follow the most-favored-nation (MFN) principle of equal treatment - compare to explicitly discriminatory measures such as antidumping? We address this question empirically by examining the trade effects of the 2002 US safeguard on steel imports and comparing this with the impact of other US trade remedies on steel imports in the 1990s. We first estimate a fixed-effects model on a dynamic panel of product-level US steel imports over and examine the potential discriminatory impact on foreign-produced steel of the 2002 MFN safeguard that used relatively new tools from the policymakers arsenal: country and product exclusions. A unique data set on the excluded products allows us to document the sizable impact on trade of both forms of preferential treatment. We also exploit higherfrequency (i.e., quarterly) data to examine potential differences in the timing of the foreign export response to policies of differential treatment. With respect to safeguard exclusions, we find that while developed country exporters have a quicker response to an exclusion, the developing-country export response is more persistent. Finally, relative to antidumping measures, country and product exclusions from a safeguard allow the protection-imposing country to target preferential treatment more effectively toward specific foreign countries, much like a preferential trade agreement, or even more narrowly toward a specific foreign firm. Thus costly trade diversion could be an even greater concern with a safeguard than with explicitly discriminatory protection such as antidumping. JEL No. F13 Keywords: MFN, Product Exclusions, Country Exclusions, Safeguards, Antidumping, Trade Diversion correspondence: Department of Economics, MS 021, Brandeis University, Waltham, MA USA tel: , fax: , cbown@brandeis.edu, web: cbown/ I gratefully acknowledge financial support from a Brandeis University Perlmutter Fellowship and Mazer Award. Thanks to Tom Prusa, Robert Staiger, Pravin Krishna, Michael Moore, Rachel McCulloch, Meredith Crowley, James Durling, Jeff Campbell and seminar participants at Brandeis, Brown, the Federal Reserve Bank of Chicago, LSE, Rutgers and the 2004 NASMES for helpful comments. Gloria Sheu, Teresa Power and Renee Bowen provided outstanding research assistance. All remaining errors are my own.

2 1 Introduction In March 2002, the United States government implemented a politically controversial safeguard policy under Section 201 of the US trade law. The policy of import tariffs and quotas was designed to shield the domestic steel industry from injurious foreign-produced steel. Conservative estimates presented in table 1 put the aggregate trade impact as a 13.5% reduction in the value of US steel imports in the year following the safeguard in the product categories targeted by the policy, eliminating close to $683 million worth of trade relative to the previous year. While the aggregate trade impact of the massive 2002 steel safeguard is impressive in its own right, the actual impact on imports within the affected categories is perhaps masked by the perception that a safeguard (SG) policy is automatically applied so as to follow the GATT/WTO s most-favored-nation (MFN) principle. One important way through which the SG policy tool is statutorily distinct and perhaps economically preferable from other trade remedies - the antidumping (AD) or countervailing duty (CVD) laws that are historically more popular with domestic petitioners seeking import protection - is that these unfair trade laws apply protection to imports from only one country per petition, thus allowing for the substantial differential and discriminatory treatment across trading partners. On the other hand, the safeguard law is supposed to result in protection being applied through nondiscriminatory tariffs on imports, irrespective of the source country. 1 From a second-best perspective that takes the implementation of some import protection as given, this is one reason why economists frequently argue that an MFN safeguard may be preferable. The alternative use of AD and CVD measures allows for discrimination across export sources which can lead to trade diversion, or to importers switching to the sourcing of products to higher cost (but non-targeted) foreign producers, thus inducing the welfare losses to the domestic economy initially identified by Viner (1950). Nevertheless, while the US s Section 201 safeguard statute has the economic appeal of requiring less discrimination than these other forms of import protection, how nondiscriminatory was the implementation of the actual safeguard in the 2002 steel case? Anecdotally, the 2002 application of tariffs and quotas appears to have allowed for substantial discrimination. First, the US exempted steel imports from several important exporting countries from the safeguard altogether. Second, the US exempted over a thousand firm-specific products from the safeguard after soliciting requests for such product exclusions from domestic steel-consuming industries and foreign exporting firms. Therefore, while overall imports of steel products in affected categories decreased by 13.5% in the twelve months following the safeguard, the magnitude of the import reduction was likely to be far from uniform across export sources and product categories. Indeed, as table 1 indicates, foreign steel exempted in the March 2002 steel safeguard announcement deriving from excluded preferential trade agreement (PTA) countries such as 1 This is not the only distinction between SG and AD/CVD. In addition to issue of fair versus unfair trade, the AD/CVD process is bureaucratic while safeguards allow for Presidential discretion, the injury threshold is higher for SG cases, the duration of safeguards is shorter than AD/CVD, and the use of SG can require compensation to affected countries while AD/CVD does not. For a discussion, see Bown (2002). 1

3 Canada and Mexico ($16 million), from excluded developing countries ($424 million) or from excluded firm-specific product exemptions ($77 million) actually saw the value of their exports to the US increase in the aftermath of the safeguard, as they continued to face low rates of import protection and now less fierce competition from other foreign rivals. 2 One implication is that foreign sources that were not exempted and thus which faced the tariffs and quotas of the March 2002 safeguard saw a much larger reduction in exports than would be expected from looking at the aggregated data. As table 1 confirms, US imports of safeguarded products from producers that did not receive exclusions fell by 30%, or roughly $1.2 billion, from the level of imports received from those same producers in the twelve months prior to the safeguard. In the aftermath of the safeguard, these foreign sources not only faced a competitive disadvantage relative to US steel producers but also relative to other foreign producers that received implicit preferential treatment through exclusions. This paper is the first to econometrically investigate the size and nature of discriminatory treatment across export sources both within and across the steel product categories affected and unaffected by the 2002 safeguard. We use a panel of product-level US steel import data from to investigate whether the implementation of the safeguard had a discriminatory impact, i.e., whether the suggestive evidence provided in table 1 holds up to a formal econometric analysis. We combine the trade data with detailed information on the products affected by the safeguard, as well as the country exclusions and a unique data set of excluded products derived directly from firm-specific petitions filed with the US Department of Commerce. We then compare the pattern of discrimination across countries and products associated with the 2002 safeguard to the explicitly discriminatory earlier acts of protection that the US steel industry received in the 1990s through its appeal to ADDs, CVDs, suspension agreements, and other trade restricting measures. One goal is to investigate whether the 2002 use of an MFN trade policy was any more or less discriminatory than these earlier acts of import protection. This research also contributes to the literature on discrimination versus nondiscrimination in trade policy, initiated by the pioneering work of Viner (1950). More recently, a substantial theoretical literature (including Bagwell and Staiger, 1997, 1999, 2004; Levy, 1997; and Ethier, 2004) examines the role of preferential policy exceptions in multilateral trade agreements. 3 On the other hand, empirical papers such as Prusa (1997, 2001) and Konings, Vandenbussche and Springael (2001) examine related questions of discriminatory treatment by investigating the trade effects and potential trade diversion of earlier use of antidumping in the US and EU, respectively. Furthermore, Bown and Crowley (2004) propose and examine an alternative way through which antidumping and safeguards may have a differential trade impact by considering the effects of examples of such US trade policies on Japanese export flows to third country markets, a phenomenon they term trade deflection. None of these papers, however, examines 2 The data in table 1 underestimates the total differential impact across countries as it does not account for the hundreds of product exclusions granted by the USTR after March 2002 (see table 3 and our discussion below). We will estimate the impact of these exclusions in our formal econometric analysis below. 3 For a survey of other recent papers focusing on different theoretical elements of the interaction between preferential and multilateral agreements, see Krishna (forthcoming). 2

4 the potential discriminatory impact of safeguard policies on the policy-imposing country s imports nor compares the impact to explicitly discriminatory policies such as antidumping. 4 The questions under investigation in this paper regarding the potential differential trade impact of these various forms of trade remedies are important for many policy reasons. As has been well documented (Miranda et al, 1998; Prusa, 2001; Crowley, 2004), the adoption and use of antidumping law has proliferated worldwide, across developed and developing countries alike. There is anecdotal evidence of a similar trend for the use of safeguards. Between 2000 and 2003, 45 definitive safeguard measures were imposed worldwide by WTO Member countries. While this number pales in comparison to the hundreds of antidumping measures imposed over the same time period, it is a substantial increase from the 15 such definitive safeguard measures implemented between 1995 and 1999 (WTO, various issues). 5 Finally, with respect to the form of safeguard protection, a normative policy question under debate is the appropriate form of its use - whether it ought to be applied on an MFN versus a discriminatory basis. This paper can help inform the extent to which even the safeguard policy tool, in this particularly expansive application, has been implemented on a nondiscriminatory basis. We find that the imposition of the 2002 steel safeguard policy had quite a discriminatory impact on the pattern of US imports that in many ways resembled the pattern of protection the US steel industry received under ADDs and CVDs at earlier points in the 1990s. Furthermore, we exploit higher frequency (i.e., quarterly) data which allows us to investigate potential differences in the timing of the impact of the 2002 differential treatment on exporter response. We find that while developed country exporters respond more quickly when granted an exclusion, the developing-country exporters response is more persistent. Furthermore, we also use the quarterly data for the first time to examine the trade impact of antidumping and countervailing duty cases, and we find support for the Staiger and Wolak (1994) results on the differential impact of AD/CVD investigations that conclude in different outcomes. In particular, we document evidence on the length of the investigation effect of AD/CVD petitions; i.e., that exporters that are investigated but which do not even face preliminary duties also see a two quarter reduction in exports before returning to their pre-investigation level. Furthermore, exporters that face preliminary duties but which do not face final duties see a four quarter reduction in exports before 4 Bown and McCulloch (2003; forthcoming) investigate a related question of whether earlier ( ) safeguard actions undertaken by the US and other WTO members have a discriminatory impact across export sources. The focus of these papers, however, is whether safeguard-applying countries have used specific, codified legal exceptions to the principle of nondiscrimination written into the WTO Agreement on Safeguards to apply safeguard protection in a discriminatory way. The approach here is different in both its econometric approach, its comparison to antidumping, and its additional focus on product exclusions, which were neither a feature of the earlier safeguard actions, nor a recognized WTO-consistent exception to the nondiscrimination rule. 5 This simple comparison also fails to control for different global economic conditions between the and time periods. Nevertheless, it should be noted that 31 (between ) is the most safeguard actions taken over any prior four year period since the GATT s establishment in 1947 (Bown and Crowley, forthcoming, table 2). Furthermore, a simple comparison of the number of safeguard and antidumping measures imposed does not provide an accurate reflection of the relative sizes of the importance of the measures, as each individual safeguard action may negatively affect the exports of dozens of countries while each antidumping action will only be targeted toward exports from one country. 3

5 returning to their pre-investigation level. We contrast this with the expected result from a safeguard investigation, where there is no evidence of a statistically robust negative investigation effect. Finally, while we find the full effect of the 2002 safeguard policy with country and product exclusions to be quite discriminatory, our results also highlight an important similarity between safeguards and PTAs. Relative to antidumping measures, country and product exclusions from a safeguard allow the protection-imposing country to target preferential treatment more effectively toward specific foreign countries, much like a PTA, or even more narrowly toward a specific foreign firm. Thus costly trade diversion could be an even greater concern with such features of a safeguard, if they were to become codified standard procedure in the application of the policy, relative to even explicitly discriminatory protection such as antidumping. The rest of this paper proceeds as follows. Section 2 discusses the basic institutional background and theory, and section 3 presents the econometric model and a discussion of data. Section 4 describes our estimation results, and section 5 concludes. 2 Institutional Background and Theory 2.1 The GATT, WTO and US Rules on Safeguards, Antidumping, and Countervailing Duties The use of safeguards, antidumping and countervailing duties is authorized under the rules of the General Agreement on Tariffs and Trade (GATT) and the World Trade Organization (WTO), as well as the laws of the United States. 6 Under the GATT 1947, safeguards were authorized under Article XIX, while antidumping and countervailing duties were authorized under Article VI. With the establishment of the WTO in 1995, these simple Articles laying out the framework for AD, CVD and SG have been expanded into formal agreements: the Agreement on Safeguards, the Agreement on Antidumping, and the Agreement on Subsidies and Countervailing Measures. In the United States, the trade remedy statutes now authorized under the WTO are currently implemented under Section 201 (for safeguards), Section 731 (for antidumping) and Section 701 (for countervailing duties) of the US trade law. 2.2 The US Steel Industry s use of Antidumping, Countervailing Duties and Safeguard Measures over The US steel industry has sought government intervention to shield it from imports almost continuously since the 1960s (Rosegrant, 2002), and it has been the most frequent user of the US s trade remedy laws since the early 1980s. Over the period of our sample, the US industry and unions filed hundreds of petitions against firms from dozens of exporting countries over thousands of steel products 6 Blonigen and Prusa (2004) provide a survey of the economics research literature on antidumping as well as a description of the US AD process, while Bown and Crowley (forthcoming) survey the literature on safeguards. 4

6 alleging that foreign steel was dumped into the US, subsidized by the foreign government or otherwise a threat to injure the domestic industry. Table 2 illustrates how dozens of investigations resulted in the imposition of duties over hundreds of different 10-digit Harmonized Tariff Schedule (HTS) steel product categories during the time period. For a comparison of the frequency of these actions, as we describe below, we have a panel of 1471 unique 10-digit HTS steel product codes from chapters 72 and 73 that are represented in the US trade data over the same period. The first row of this table, for example, suggests that rougly 1 in 6 of these 10-digit steel products were hit with a SG measure during this time period, 1 in 5 were hit with a CVD, and 1 in 4 products were hit with an ADD. 7 In addition to the 2002 steel safeguard (described below) that affected some 272 unique 10-digit HTS categories, in 2000 the US also imposed much smaller tariff-rate quota safeguard on five different 10-digit HTS product categories of steel wire rod and circular welded pipe The 2002 Steel Safeguard Table 3 illustrates the essentials of the process for the events surrounding the imposition of safeguard protection by the US in In July 2001, the US Trade Representative (USTR) requested the US International Trade Commission (ITC) commence a formal safeguard investigation under Section 201 of the US trade law. The USTR requested specifically that the ITC determine whether steel imports under 612 different 10-digit HTS product categories were a substantial cause of serious injury to the domestic steel industry. In October 2001, the ITC made its announcement that it had found that imports in a substantial number of product categories had caused injury to the domestic steel industry, and in December 2001, the ITC announced its non-binding recommendation for safeguard tariffs and quotas. During this time, the USTR was also requesting that US steel-consuming industries and foreign exporters begin submitting petitions to have their products excluded from any upcoming steel safeguard. They could be allocated such an exclusion provided that they could illustrate to the USTR s satisfaction that domestic US production of that particular variety of steel was insufficient to meet domestic demand After 1998, only 1035 unique 10-digit HTS categories from chapter 72 or 73 are in our sample of trade data, which suggests that the ratio for the 3 safeguard cases (all implemented since 1998) is even higher if we factor into account administrative changes in the HTS schedule over time. 8 Note that while Canada and Mexico were also excluded from these particular safeguards, no developing countries were excluded. Furthermore, Korea filed a formal WTO trade dispute over the US safeguard on circular welded pipe and as part of the settlement, the US granted Korea additional access to the quota in September See also Hufbauer and Goodrich (2003). 10 Specifically, the USTR announced that it would use the following criteria in making its decisions of which product exclusion requests to accept: We will grant only those exclusions that do not undermine the objectives of the safeguard measures. In analyzing the requests, we will consider whether it is currently being produced in the United States, whether substitution of the product is possible, whether qualification requirements affect the requestor s ability to use domestic products, inventories, whether the requested product is under development by a U.S. producer who will imminently be able to produce it in marketable quantities, and any other relevant factors. Where necessary, we will meet with parties to discuss the information that was submitted and/or to gain addi- 5

7 In March 2002, a US Presidential Proclamation ordered the application of safeguard tariffs and quotas on 272 different 10-digit HTS product categories. 11 All exporters of a product within a particularly affected 10-digit category would face the same tariff rate assigned to that category unless their country or product were subject to an exclusion from the policy, in which case it would not face any SG restrictions at all. The US safeguard policy then excluded three different categories of imports from the safeguard - two categories based on country definitions, and one in the form of discretionary product exclusions administered by the USTR Developing country exclusions from the 2002 safeguard The first example of countries excluded from the steel safeguard includes a list of 100 developing countries. These developing country exclusions are mandated by the rules of the WTO, where Article 9.1 of the Agreement on Safeguards requires that a safeguard-imposing country exclude developing country suppliers who have less than a 3% share of the affected import market, provided they collectively also have a less than 9% market share. 12 To comply with this element of the law, the US excluded potential imports from 100 developing countries from the safeguard tariffs and quotas Preferential trade agreement country exclusions from the 2002 safeguard The second way through which the US excluded imports from the safeguard tariffs and quotas was by exempting steel products deriving from members of its preferential trade agreements (PTAs) - Canada, Mexico, Israel and Jordan. The WTO-legality of such PTA-country exemptions is doubtful and has been questioned in a number of formal trade disputes litigated under the WTO, without a conclusive answer as to its legitimacy. 14 Nevertheless, the PTA-country exemptions by the US in this case were not unprecedented; the US had exempted PTA members in at least five earlier safeguard actions since 1995 (brooms, wheat gluten, lamb meat, steel wire rod, and circular welded pipe) as had Argentina and Brazil in safeguard actions of their own Product exclusions from the 2002 Safeguard The third way through which the US excluded imports from being subject to the safeguard was through firm-specific product exclusions. These product exclusions were solicited by the USTR, which requested tional information. (USTR website, last accessed on 28 March There were also a handful of products in chapter 84 of the US HTS which we omit, as our focus is only trade policies affecting products in HTS chapters 72 or There is a similar de minimus requirement under the WTO s Agreement on Antidumping that duties should not be imposed on small suppliers. The question of whether that provision is followed in practice is not under investigation here. 13 A subset of steel imported from seven of these developing countries (Brazil, India, Turkey, Moldova, Romania, Thailand and Venezuela) were exempted from the country exclusions and thus faced the safeguard protection. 14 See the discussion in Bown and McCulloch (2003), including the issue of parallelism, or of including imports from PTA members in the domestic industry s injury calculation but not including the PTA members in the remedy application. 6

8 domestic steel consuming industries and/or foreign steel exporters submit petitions requesting the exclusions, and they were asked to demonstrate that their particular product request was otherwise in short-supply in the US market. 15 The USTR received over 2000 such petition requests and granted over 1000 of these requests at various announcements made between March 2002 and March Interestingly, while exclusion product requests were made by exporting firms from dozens of foreign countries, over 90% of the exclusions granted in the various rounds announced by the USTR went to firms from Japan or the EU. 16 Finally, note that the granting of a product exclusion was not then typically extended on an MFN-like basis to close substitutes from competing firms in the same foreign country, let alone competitor firms from other countries who could also produce substitutes. In contrast, a typical exclusion might be so narrowly defined so as to be a trademarked product that, by definition, could only be produced by a particular foreign firm. 17 Thus, many of the foreign products initially subject to higher levels of tariff protection implied by the original March 2002 safeguard application were no longer subject to higher duties or more stringent quantitative restrictions. While this may appear to be a positive development for liberal trade and those concerned with the negative impact of trade remedies, the full implication of these product exclusions may not be so innocuous if the result is preferential treatment granted toward higher cost foreign producers and thus potentially trade diversion. The likelihood of this outcome is as yet unknown but of potential concern given that the wholesale solicitation and granting of product exclusions was unprecedented and without statutory guidance or oversight, allowing the USTR the possibility of exerting tremendous discretion in its decision-making. 2.4 Theory Our discussion of the application of the US trade laws for antidumping, countervailing duties and safeguard measures suggests that there are a variety of ways that exporters can face differential treatment. Through antidumping and countervailing duties, some exporters may not be named in investigations at all, some may be investigated but found innocent of illegal dumping or subsidies and thus not be subject to duties, or even if they are subject to duties, the discriminatory treatment may come through the differential tariffs that they face on their exported products. On the other hand, in practice under 15 US steel producers had the ability to file objections to these petition requests where they could claim that they could produce the products seeking exemption. 16 During this time period, a number of countries including Japan and the members of the EU were challenging the legality of the US safeguard at the WTO, with threats to retaliate with higher tariffs on US exports. One explanation is that some product exclusions were granted in part to informally compensate such aggrieved parties. 17 For example, see product exclusion N granted to the UK firm Somers Forge, Ltd. on 11 June 2002, Forged alloy steel die blocks of round or rectangular cross section. US Trademark No , commonly known as VMC or HYTUF, or exclusion N granted to the Japanese firm Daido Steel on 22 August 2002, A specialized, high grade tool steel, known as Daido s proprietary grade NAK 55, that is used for the construction of plastic molds. See the USTR s website, President Bush Takes Action on Steel, last access date of 29 February

9 the 2002 steel safeguard, the discriminatory treatment arose through a country or exporter receiving an exemption from the policy - either through a PTA or developing country exclusion, or through a specific product exclusion granted by the USTR. The discriminatory application of a trade policy would lead to a decrease in the imports from countries subject to the increased protection and an increase in imports from countries that have been exempted from the increased protection. As Viner (1950) originally noted, the welfare implications of the discriminatory application of trade policies are theoretically ambiguous. The welfare implications of the US application of discriminatory and nondiscriminatory trade remedies are illustrated in figure 1. Suppose for simplicity the US is a small importer of a homogenous good that is produced under conditions of perfect competition by exporters in two foreign countries labeled i = {1, 2}. Assume that the starting point is that the US has applied an MFN tariff on imports of the product from all sources, and this allows it to consume imports from the low cost foreign supplier, which in this case is country 1, thus resulting in the pre-remedy price P. Without loss of generality suppose this initial MFN tariff is zero so that we begin with free trade. We will compare the welfare implications of two different US trade remedy policies: (i) an MFN tariff of size τ on imports from all sources, and (ii) a tariff of size τ on imports from country 1 only, thus excluding imports from country 2 from the remedy. Starting from free trade, the imposition of an MFN tariff of size τ generates the standard loss to consumer surplus and increase in producer surplus and US tariff revenue, resulting in the deadweight loss associated with the inefficiency given by areas B + D + F + H. On the other hand, a discriminatory tariff of size τ on country 1 alone leads to smaller reductions in consumer surplus and increases in producer surplus that leads to a deadweight loss associated with the inefficiency given by areas F + G + H. Comparing the two policies, if G > B + D, a discriminatory trade remedy would be welfare-inferior relative to an MFN trade remedy of the same size. The potential welfare cost of the discriminatory policy (relative to the nondiscriminatory policy) arises from the fact that the exclusion is granted to higher cost exporters in country 2, as US importers now source their product from less efficient foreign producers. Furthermore, note that a discriminatory policy in the form of a trade remedy imposed on the high cost country 2 only (i.e., excluding country 1) would have no effect on welfare or trade flows, as the US would continue to import at the free trade price from the low cost foreign producer. 3 Data and Estimation 3.1 Basic empirical model To investigate the questions of interest regarding the differential trade impact of the discriminatory application of safeguards and other trade remedies, we develop the following reduced-form specification for the quantity of US imports from country i of product h at time t 8

10 ln(m iht ) = α i + α h + α i,t + β 1I ih ln(1 + τ ht ) + β 2ln(m iht 1 ) + ɛ iht, (1) where ln(m iht ) denotes the natural logarithm of the quantity of imports of product h from exporters in country i at time t, ln(1 + τ ht ) is one plus the ad valorem US trade remedy against imports of h at time t, where the ad valorem rate is equal to zero if no remedy is applied. Furthermore, I ih denotes a vector of zero/one indicators that country i was or was not exempt from a particular US trade remedy against imports of h - either because it was an excluded country or was exporting an excluded product in the case of the 2002 safeguard, or because it was not-named in an AD or CVD petition that resulted in duties, etc. Finally, α i and α h are country i and product h fixed effects, respectively, while α i,t is a country-specific effect that may vary over time, and which we use to control for exporting countryspecific covariates (e.g., exchange rates, productivity, domestic subsidies) that may also be affecting the level of US imports. Absent from this estimation equation are any industry-level control variables, which we omit given our concentration on only steel products in our estimation exercise. 3.2 Estimation strategy There are two problems to address in estimating equation (1). First, the autocorrelation of ln(m iht ) implies that least squares estimation of (1) yields biased estimates. Second, in a short panel, the number of parameters to be estimated (α i and γ h ) increases with the number of countries and commodities, so that α i and γ h cannot be consistently estimated. To we address both of these problems we follow Arellano and Bond (1991) and estimate the first difference of (1) and instrument for the lagged change in imports with the second lag of the log level of imports. Taking the first lag of (1) and subtracting this from (1) yields: ln(m iht ) = α i,t + β 1I ih (1 + τ ht ) + β 2 ln(m iht 1 ) + ɛ iht. (2) After first differencing, direct estimation of (2) yields biased coefficients because the lagged difference in the log of imports [ln(m iht 1 ) ln(m iht 2 )] is correlated with the error term (ɛ iht ɛ iht 1 ). We therefore take an instrumental variables approach to address this problem by instrumenting for ln(m iht 1 ) with ln(m iht 2 ). Thus, our final estimation equation is given by 18 ln(m iht ) = α i,t + β 1I ih (1 + τ ht ) + β 2ln(m iht 2 ) + ɛ iht. (3) Instrumental Variable Tests Because of the dynamic panel structure of our data, there are two potential problems with the IV estimator used in equation (3): bias associated with the use of a weak instrument and bias associated 18 Because α i,t is time-varying, it will also pick up changes in the level of US steel imports over time that may be due to changes in US aggregate demand that are common across product categories. In unreported results we have used separate country and time dummies in the estimation, and the qualitative pattern of results in unchanged from that reported below. 9

11 with correlation in measurement error. To address each of these concerns, we perform two separate tests for each of the specifications reported in tables 5 through 8 below. To conserve space, we do not report the coefficient estimates associated with the instrument tests described below, but they are available from the author upon request. The first concern is that ln(m iht 2 ) is a weak instrument for ln(m iht 1 ), and we address this by estimating a first-stage model of the form ln(m iht 1 ) = α i,t + β 1I ih (1 + τ ht ) + β 2ln(m iht 2 ) + ɛ iht. (4) We then compare these estimates with the estimation of a restricted regression of equation (4) under the assumption that β 2 is constrained to be zero. The F-statistic associated with this test in all cases easily surpasses the 99% critical threshold for χ 2 (1) = 6.63, and the statistic is frequently calculated to be larger than Similar results obtain if we instrument for ln(m iht 1 ) with ln(m iht 3 ), but since use of ln(m iht 3 ) further shortens an already relatively short panel, we report results using ln(m iht 2 ) as our instrument. The second potential concern with our use of instruments may be caused by measurement error. If there is measurement error in our series of m iht, then our IV estimator may be biased as measurement error in the regressor [ ln(m iht 1 )] will be correlated with measurement error in the instrument [ln(m iht 2 )]. Thus we can again use the third lag of the log level of imports [ln(m iht 3 )] as a comparison, as it has the advantage of not having measurement error correlated with measurement error in ln(m iht 1 ). While we do not report them here to conserve space, a comparison of specifications using ln(m iht 3 ) versus ln(m iht 2 ) as our instrument suggests that the parameter estimates of interest are not substantially different, so that we conclude that measurement error in m iht is not a significant problem. Since our estimates are robust to a choice of these two instruments, we again choose to use ln(m iht 2 ) as it allows us to estimate the model over a slightly longer panel of data Comparison to the Literature With a number of important caveats, this estimation approach has similarities to the basic approach undertaken by Prusa (1997, 2001), which studies the yearly trade effects of US antidumping measures implemented between First, Prusa focuses only on AD cases for the time period but does examine all (i.e., steel and non-steel) products targeted in those cases. In contrast, our focus is on steel products only, so that our lack of US industry-level covariates that would be needed to control for product-level import variation across industries is less of a concern. Furthermore, our sample covers the time period for which we have a consistent time series of 10-digit HTS import data; and we also consider the impact of the imposition of suspension agreements, SG and CVD cases in addition to AD, as well as the removal of such policies on product-level imports. 19 See also Konings, Vandenbussche and Springael (2001) that apply Prusa s approach to AD cases in the European Union. 10

12 A second important contribution made by our approach is that we have constructed a dynamic panel of all steel products imported by the US over the time period - regardless of whether a particular product was targeted by a trade remedy. On the other hand, Prusa s estimation sample considers only US imports in product categories that were hit with an AD measure, and he uses separate samples to examine the impact of trade remedies on targeted products exported by targeted and non-targeted suppliers. Our approach goes one step further. By using a dynamic panel of all US steel imports, this allows us to take advantage of the substantial variation in the trade data of the products that were targeted and not-targeted by US trade remedies in addition to the exporters that were targeted and not-targeted by US trade remedies within categories of targeted products. Lastly, in a final estimation exercise, we will also present evidence from a specification on a sample of higher frequency (i.e., quarterly) data of US, product-level steel imports to better investigate the question of the timing of the impact of a trade remedy on trade flows. 3.3 Data Trade Data To estimate the model, we use product-level data on US imports of steel at the 10-digit Harmonized Tariff Schedule (HTS) level, where most steel products are allocated to chapters 72 or 73. Import data for the US at the 10-digit HTS level is available from the US International Trade Commission s DataWeb data base for the years Given the need to instrument with ln(m iht 2 ), this allows us to estimate equation (3) on a dynamic panel of yearly, bilateral, product-level US import data from As stated earlier, we have an unbalanced panel of 1471 different 10-digit HTS products in chapters 72 or 73 over the life of the sample. Our initial investigation will estimate the model on yearly data for product-level US steel imports. Nevertheless, below we will also utilize higher-frequency, quarterly data (also available from DataWeb) in some specifications for the full period in order to examine the timing of the export response to the imposition of trade remedies as well as the duration of the persistence that the imposition of a remedy has within a given year Policy Data This paper investigates the potentially differential effects of various instruments of protection affecting steel imports into the US during the period. We have collected detailed product-level changes in trade policies associated with antidumping and countervailing duty investigations, the removal of AD and CVD orders after revocation orders or sunset reviews, the imposition of suspension agreements and the imposition and removal of acts of safeguard protection. The information regarding the implementation and removal of AD or CVD measures or suspension 20 See last access date of 30 January

13 agreements is available from the US Federal Register, as well as public documents made available at either the USITC or the Department of Commerce s official websites. We have collected data on the dates of preliminary dumping or subsidy and injury determinations, final dumping or subsidy and injury determinations, the size of the duties imposed in either preliminary or final affirmative decisions, the 10-digit HTS products investigated and being subject to preliminary and/or final duties, the dates of and HTS 10-digit products subject to revocation orders where imposed duties were removed, and the foreign countries whose exports were directly affected by all of these policy changes. Much of the data for the safeguard cases is also publicly available from either the Federal Register or the ITC s official website. The US President imposed safeguard protection over various 10-digit HTS products in the steel industry on three occasions during the period: circular welded pipe and steel wire rod in 2000 (five different 10-digit HTS products), as well as the 272 different HTS-products targeted by the March 2002 broad-based policy that we refer to as the 2002 Steel Safeguard. In each case, the 10-digit HTS products as well as any excluded countries are made publicly available in the Presidential Proclamation announcing the safeguard policy. The one important piece of policy information that is not readily available from electronic sources for the 2002 steel safeguard is the 10-digit HTS categories of the product exclusions granted by the USTR. Descriptions of the physical characteristics of the products excluded from the safeguard are publicly available from the USTR s website; however, this information is not useful when trying to systematically match excluded products to the 10-digit HTS coded import data used in the estimation of our model. Information on the 10-digit HTS codes of the excluded products is available in the actual surveys that petitioners had to fill out to request that their product be excluded. Such petitions are publicly available, though the data had to be transcribed by hand from hard copy surveys available in the International Trade Administration s Central Records and Subsidy Library in the Department of Commerce in Washington, DC. There were over 2000 petitions from firms requesting to have their products excluded from the safeguard. 21 Finally, we also note that a product exclusion was typically granted to a product from a single exporting firm which would be even more narrowly defined than a 10-digit HTS product category. Thus, not all products within a 10-digit product category would be excluded. Table 4 provides summary statistics for the yearly data used in the estimation of specifications reported in the next section. To conserve space, we do not present summary statistics for the tariffs associated with the 2002 safeguard, which we use to estimate specifications presented in table 6 below. For the digit HTS codes which faced a tariff in the first year after the safeguard, 185 (69%) received a 30% tariff, 60 (22%) received a 15% tariff, 15 (6%) received a 13% tariff and 7 (3%) received an 8% tariff in the first year. In March 2003, the tariffs for each of the categories were stepped-down to 24%, 12%, 10% and 7%, respectively. 21 There is some measurement error with the survey data as petitioners sometimes left the entry for 10-digit HTS code blank, or instead entered an incorrect product code that was not subject to the safeguard. 12

14 4 Empirical Results 4.1 Results for the 2002 steel safeguard Table 5a presents results for our first estimation of equation (3) where we use indicators for our trade remedy policy variables. Consider first specification (1) in which the dependent variable is the yearly growth rate of the quantity of imports of product h from country i. The top rows of explanatory variables contain estimates for the first set of variables of interest, i.e. those related to the March 2002 steel safeguard. The effect of the imposition of the safeguard is negative, as expected, with the safeguard leading to a 19.9 percentage point reduction in export growth for 10-digit products that were hit, relative to investigated products that were not hit. As we have also included a 2002 indicator for the HTS products that were merely investigated as ordered by the USTR and the export growth of these products in 2002 fell by 10.1 percentage points, the combination of these estimates is evidence that, relative to an imported steel product that was not under investigation in the Section 201 case, a steel product hit with the safeguard saw export growth fall by 30.0 (= ) percentage points. It is useful to have as a benchmark other steel imports h that were investigated but which were not part of the safeguard order, as there is some concern that steel imports of all investigated products surged in the latter half of 2001 and first quarter of 2002, in the attempt to get the imports into the US market before any safeguard went into effect. 22 While the positive and significant estimate (0.111) for the 2001 indicator for investigated products reported in specification (1) would seem to provide evidence to support this concern; nevertheless, we will see below that this result is not robust to the use of alternative specifications which allow for entering/exiting exporters, import values instead of quantities, and/or resorting to higher frequency data. The estimates of the next three rows present evidence of the differential impact that country and product exclusions had on imports of products exempted from the safeguard. For example, relative to the safeguarded products, import growth from US PTA members that were excluded (Canada, Mexico, Israel and Jordan) increased by 61.4 percentage points. Furthermore, a developing country whose 10- digit product was excluded saw its exports increase by 53.9 percentage points in 2002 relative to a safeguarded product that was not excluded. Finally the specific HTS-10 product exclusions that were announced on a product-by-product basis from specific foreign firms led to an increase in imports of 25.8 percentage points in 2002 relative to the non-excluded, but safeguarded products. The estimates of these impacts in specification (1) are evidence that not only was there a significant differential effect of the safeguard on excluded versus non-excluded products, but compared to other investigated products h that were not subject to the March 2002 safeguard order, the net trade impact of the safeguard on excluded products for each of the three categories was positive. Products excluded from the safeguard because of PTAs saw exports increase by 41.5 percentage points ( ), products 22 Unlike AD/CVD investigations, there are no preliminary duties imposable at the beginning of the investigation and there are also no retroactive duties imposed in a SG investigation. 13

15 excluded from the safeguard deriving from developing countries saw exports increase by 34.0 percentage points ( ), and products excluded from the safeguard because of USTR announcements as to firm-specific exclusions saw exports increase by 5.9 percentage points ( ). Again, the positive net impact for the USTR-granted product exclusions is somewhat of a surprise given that there may be many exported products within a 10-digit product category that were not exempted from the safeguard, and that many of the excluded products were granted in June, July and August of 2002 and thus would only have an impact on the last six months worth of trade data for Finally, even when compared to steel product imports not subject to the safeguard investigation, imported products excluded from the safeguard because of PTAs increased by 30.4 percentage points ( ) and products excluded from the safeguard deriving from developing countries increased by 22.9 percentage points ( ). The second set of estimates in the next four rows of specification (1) in table 5a illustrates the 2003 impact of the products affected by the 2002 (t 1) steel safeguard. Even though the safeguard tariffs were reduced in March 2003 from their March 2002 level across all affected products, as is mandated by the statute and was unannounced in the March 2002 Presidential Proclamation, the 2002 safeguard is again associated with a substantial negative impact on import growth from the prior year. Products targeted by the 2002 steel safeguard are associated with an additional 47.4 percentage point decrease in imports in 2003 from even the 2002 level. On the other hand, the developing country exclusions (0.505) and product exclusions (0.233) had a significantly different and positive impact on imports relative to the non-excluded products. This is not true for the products from the US s PTA partners, as there was not a significantly differential impact between these products and the non-excluded products in The middle rows of table 5a report estimates for the impact of the smaller, year 2000 imposition of safeguards on the five 10-digit HTS product categories of steel wire rod and circular welded pipe, as well as the Canada/Mexico country exclusions announced with those safeguards, and the 2002 exemption that Korea received as a settlement after a WTO trade dispute concerning the SG. These two safeguards are not associated with any statistically significant impact on US imports until two years after the safeguard went into effect, when the 2002 expansion of the tariff rate quota led to a substantial increase in safeguarded products from the prior year. Furthermore, Korea s exemption in 2002 is associated with a large increase in imports of safeguarded products from Korea as well. 4.2 Results for AD/CVD policies imposed and removed over Table 5b presents expanded results from this same estimation equation that could not all fit into table 5a; in particular, describing the impact of the imposition and removal of US AD/CVD policies on steel taken over First note that we use the following 6 categories to characterize the different circumstances facing exporters from country i of a 10-digit HTS product h under investigation in a US AD/CVD case: (a) AD/CVD investigation of i terminated with the imposition of preliminary and final duties, (b) AD/CVD investigation of i terminated with the imposition of a suspension agreement, 14