WPS6796 Policy Research Working Paper 6796 Import Dynamics and Demands for Protection Russell Hillberry Phillip McCalman The World Bank Development Research Group Trade and International Integration Team March 2014 Policy Research Working Paper 6796 Abstract What kinds of changes in foreign competition lead favors imports from named countries. Second, a negative domestic industries to seek import protection? To address demand shock hits imports from all sources just prior to this question this paper uses detailed monthly U.S. domestic industries’ petition for protection. The relative import data to investigate changes in import composition supply shock is a broad one; it applies both to named during a 24-month window immediately preceding commodities and to the comparison product group. The the filing of a petition for protection. A decomposition import demand shock, by contrast, is narrow, hitting only methodology allows a comparison of imports from two named products. The latter shock is also large: import groups of countries supplying the same product: those growth over the two-year window is 15 percentage points that are named in the petition and those that are not. lower in named products than in reference products, The same decomposition can be applied to products with most of this gap arising in the final two quarters quite similar to the imports in question, but not subject before the petition. The negative import demand shock to a petition. The results suggest that industries typically appears to be a key event in the run-up to the filing of seek protection when faced with a specific pattern of a petition. It has been missed by previous studies using shocks. First, a persistent positive relative supply shock more aggregated data. This paper is a product of the Trade and International Integration Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at rhillberry@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Import Dynamics and Demands for Protection∗† ‡ Russell Hillberry § Phillip McCalman Key words: Import dynamics, endogenous trade policy, anti-dumping JEL codes: F13, F14, F53 Sector Board: Economic Policy (EPOL) ∗ An earlier version of this paper circulated under the title ‘What Triggers an Anti-dumping Petition? Finding the Devil in the Detail.’ † We would like to thank Bruce Blonigen, Chad Bown, Robert Feinberg, James Hartigan, John Ries, Georg Schaur, and Maurizio Zanardi for useful comments on earlier drafts, as well as audiences at the Universities of Adelaide, California-Santa Cruz, Melbourne, and Tennessee, Deakin and Purdue Universities, the U.S. International Trade Commission, the World Bank, the Australasian Trade Workshop, the Ljubljana and Rocky Mountain Empirical Trade conferences, and the Midwest International Economics Group conference. ‡ Development Research Group, World Bank, 1818 H Street NW, Washington DC, 20433, United States. e-mail: rhillberry@worldbank.org § Department of Economics, Level 4, FBE Building 105, 111 Barry Street, University of Melbourne, Melbourne, VIC 3010, Australia, e-mail: mccalman@unimelb.edu.au 1 Introduction One of the main achievements of contemporary international economic cooperation has been the liberalization of international trade. Starting with the General Agreement on Tariffs and Trade, the world trading system has evolved into one which has seen the complete elimination of trade barriers on some products and widespread constraints on the use of trade policy in relation to most others. Given the general commitment to lower trade barriers, deviations from this norm assume great importance and are only possible if certain criteria are met.1 These exceptions are designed to be quite narrow - imports of very specific products can face high tariff barriers while very similar products remain unprotected. The most important exception, anti-dumping duties (AD), are applied discriminately, affecting some countries’ exports of the specified product and not other countries’ exports. The narrowness of the targeted products and the policy’s discriminate use provide a setting in which relative comparisons between affected and unaffected imports can be used to gain insight into the nature of the pressures that animate trade policy. In this paper we use monthly U.S. import data at the detailed product level to characterize the evolution of imports in a two-year window leading up to U.S. industries’ formal requests for protection via AD. A decomposition technique developed by Hummels and Klenow (2005) allows a comparison of imports of the specified products from countries that will, and will not, face formal complaints of unfair trade. This within-product decomposition can also be applied to closely related products in which no such complaints arise. The comparison of the time profiles of imports from ‘subject’ and ‘non-subject’ countries among ‘subject’ and closely-related ‘non-subject’ products generates novel insights about the developments in import markets that typically precede formal requests for protection.2 The evidence points to AD petitions being filed in the aftermath of a specific combination of eco- nomic shocks. Our analysis suggests that the typical case appears to be preceded by a broad-based, and persistent, relative supply shock favoring exports from the countries named in the petition. Despite sizable increases in imports from subject countries, aspects of the status quo remain intact; the overall level of imports from all sources and the average import price remain relatively constant in the face of the relative supply shock. However, just prior to the filing of a petition, it appears 1 There is a large literature on the design of trade agreements and their exceptions. See for example Bagwell and Staiger (1999), Bagwell and Staiger (2002), Bagwell and Staiger (2012), Saggi and Sara (2008) and Saggi (2006). 2 We appropriate the terms ‘subject’ and ‘non-subject’ from the language used in the case law on AD. The petition identifies the countries and products that are subject to the petition. 1 that a negative import demand shock affects imports from all sources, placing the preceding rel- ative supply shock in a different light. This interpretation is underscored by a comparison with closely-related reference commodities in which no petition is filed. Among these commodities, the relative supply shock is also evident, and of similar magnitude and duration. However, the reference commodities do not experience the negative demand shock, which suggests that a narrow negative demand shock is a key event in the run-up to a petition. Overall, it appears that an export supply shock is a necessary but not sufficient condition for a petition to be filed, and that it is a particular combination of shocks that triggers the filing of an AD petition. It is the narrow import demand shock that is the most novel finding of the paper, and the shock is a sizable one. In the subject commodities total imports fall by approximately eight percent in the six months prior to the petition. At the same time, imports of the reference commodities that will not be subject to a subsequent petition continue to grow. Overall imports in the subject commodities fall by eight percent over the two year window, while imports of reference commodities grow by seven percent, producing a 15 percentage point gap in the relative import growth rates between subject and non-subject commodities over the two years leading up to the petition. Our work complements the existing literature, which can be divided into two broad types of papers: those that study the determinants of a petition and/or the applied duties and those that consider the consequences of protection once applied. This paper falls into the former group but takes a different approach; one distinguished by the level of detail embodied in the trade data we employ. Our detailed data allow us to match exactly the definition of a product under investigation and the exact timing of initiation of the process. Most papers addressing the determinants of filing utilize data on domestic industry performance. However, because domestic data are aggregated over products and time, these studies are constrained to deal with industry level information on an annual basis. In contrast, we utilize product-level import data that is reported on a monthly basis. The additional detail in both these dimensions turns out to be critically important since one of the shocks is common to both subject and non- subject products and long lived, but the other is narrow and occurs within the year of filing. This implies that the previous more aggregate approach puts all the emphasis on the common export supply shock (since it is also evident in the year before filing at the industry level) but misses the pronounced import demand shock that occurs immediately before filing. 2 Our work differs from the the earlier literature in several additional ways.3 First, we investigate the cumulative impact of all imports from subject countries, whereas most authors investigate the partial effects of changes in individual countries’ imports. The latter approach is consistent with econometric estimation of a panel regression, while ours accounts for cumulative changes in imports from countries named in a petition. Cumulative changes are important in the case law governing authorities’ decisions to grant protection. Second, we differ methodologically from many papers investigating the filing and/or the application of AD in that we investigate imports in great detail. But, lacking key variables in the cross-section at our level of disaggregation, we are unable to exploit panel data to test specific theories. Instead we provide side-by-side comparisons of imports of affected commodities from affected countries, and compare these to relevant reference groups. In the recent language of Gelman and Imbens (2013), one can view our work as investigating the ‘causes of effects’, while studies that employ econometric techniques operate within the more familiar ‘effects of causes’ framework. Relative to the earlier literature the key lesson our approach reveals is the apparent importance of the narrow import demand shock just prior to the filing of an AD petition. Finally, our work differs from some other papers in the literature, including Bown and Crowley (2013b), in that we focus exclusively on the filing decision, which lies only in the hands of the domestic industry, whereas those authors focus on the government’s eventual decision to apply protection and how much protection to apply. This outcome depends not only on decisionmakers within the government but also, initially, on domestic industries’ decision to file a petition.4 The paper is organized as follows. The next section reviews the existing literature on the deter- minants of filing that combine the trade and domestic data. We then describe the nature of the data set we construct, and how it is mapped into the AD petitions. To analyze this data and identify the underlying shocks, a decomposition methodology is set out in section four. Section five provides results, and section six concludes. 3 A prominent recent article that is representative, methodologically, of the literature to which we refer is Bown and Crowley (2013b). 4 Following the filing of a petition by the industry, a range of outcomes are possible, including a negotiated settlement between the parties. Given the relatively small number of cases our comparison framework is not well suited to evaluation of multiple outcomes, so we focus exclusively on the initial decision to file a petition and the details of both the timing of the filing and the products contained in the petition. 3 2 U.S. Anti-Dumping Institutions and Filing Behavior The world trading system is overseen by the WTO, an institution that has at its core the twin principles of non-discrimination and reciprocity. Guided by these values, the global trading system has been liberalized through an ongoing process of negotiations. While these principles have proven effective, the governance structure provides for a number of exceptions in specific circumstances. Some of these exceptions allow for a subset of countries to liberalize to a greater extent than the WTO requires (through preferential trade agreements), and others allow scope for countries to raise protection in excess of their commitments (safeguards and AD).5 Given the success of multilateral liberalization, any exceptions to the key principles are rightly subject to extensive scrutiny and analysis. For our part, we focus on AD. The general motivation for AD laws is to provide domestic firms with a facility to assess their claims of injury arising from “unfair” foreign competition. Since the initiation of a petition for pro- tection is from a domestic industry there is a strong similarity to a legal case where an injured party (domestic firms) seeks relief from another party (foreign firms). A critical element for the process to operate smoothly is that the product must be well defined. This leads to a key characteristic of an AD case: the products identified in a case are very specific. Such specificity reflects the quasi-judicial nature of the injury claim since it provides a clear demarcation of those who are injured and those who are potentially responsible. In the U.S. AD laws are administered by the Department of Commerce and the International Trade Commission, each with a specific role in assessing the merits of a petition. The unfair com- petition component is assessed by the Department of Commerce. In doing so it determines whether the subject product is being sold at Less Than Fair Value (LTFV): defined as selling a product in the U.S. at less than “normal” value, which is typically based on the foreign firm’s sale price in its home market. The binary nature of this comparison allows an assessment of whether pricing at LTFV is occurring for each firm named in a petition.6 The International Trade Commission deter- mines whether the domestic industry has been materially injured, or is threatened with material injury, due to the imports from the sources named in the petition. When determining injury the International Trade Commission typically aggregates the imports from all sources that are named 5 Bagwell and Staiger (1990) develop a model that examines the relationship between trade liberalization and temporary increases in protection. Looking at US trade policy, Bown and Crowley (2013b) find evidence consistent with this theory. 6 Consequently dumping margins can vary not only across countries named in a petition but also across firms within a country. 4 in a petition or named in related petitions. That is, rather than having to parse out the injury that might arise from each separate firm or country named in a petition(s), its focus is on the aggregate level of imports from the subject countries and the associated injury to the domestic industry.7 The literature has identified a number of circumstances where the domestic industry might feel justified in filing an AD petition. In general these conditions are divided into shocks that occur either in the foreign market or in the domestic market. Staiger and Wolak (1992) focus on shocks to the foreign market and construct a model of stochastic foreign demand that potentially generates an excess supply that is dumped on the domestic market when foreign demand realizations are sufficiently low. A case can also be made that a positive foreign demand shock leads to dumping as in Clarida (1993) (due to excessive entry) and Hartigan (1996) (due to an interaction with financial frictions that constrain the home firm). Despite these mechanisms Knetter and Prusa (2003) fail to find a statistically significant relationship between changes in source country GDP and filing rates in the domestic country. However, the level of aggregation seems to be important for this result. Using more detailed 3 digit ISIC data Crowley (2009) finds that a negative foreign demand shock is a significant determinant of the probability of a domestic industry filing an AD petition. While foreign shocks are viewed as a likely motivation for foreign firms to alter their pricing in the domestic market, the weight placed on the material injury criterion suggests that other shocks may also be relevant. In particular, there may be scope for a negative domestic shock to be used as a basis for filing a petition since it is likely to be associated with both a reduced domestic price (which raises the likelihood of a domestic/foreign price differential) and injury to the domestic industry. Both Knetter and Prusa (2003) and Crowley (2009) find that domestic shocks play an important role in filing behavior, along with a number of other studies (see also Bown and Crowley (2013a), Staiger and Wolak (1994), Finger (1981), Herander and Schwartz (1984), Feinberg and Hirsch (1989), Hansen (1990), Krupp (1994), Lichtenberg and Tan (1994), Furusawa and Prusa (1996), Blonigen (2000), Sabry (2000)). While the current literature provides insight into the source of shocks that are likely to precede an AD petition, their reliance on aggregate data means that they can only estimate propensities to file among broadly-defined domestic industries. Yet AD petitions relate to very specific products. For example the 3-digit ISIC classification defines 28 manufacturing sectors whereas the HS defines over 15,000 products. Given that the typical case involves approximately eight separate HS-10 codes, the 7 This process is formally known as cumulation. 5 ability of the previous literature to identify why a petition is filed against some HS-10 products and not others is hampered by the aggregate level of the analysis. Some sense of the difference made in the level of aggregation is apparent in the contrasting results obtained by Knetter and Prusa (2003) and Crowley (2009). Furthermore, the reliance on annual data is an additional constraint when identifying shocks since a petition can be filed at any point in the year. If two cases in the same HS-4 category are filed in January and December of the same year, then the standard econometric approach would be to assume that the same aggregate data are responsible for both petitions being filed. This lack of precision in terms of the products covered and the time of filing undermines the confidence in the inferences drawn from the aggregate analysis.8 An important recent paper that addresses these issues in a slightly different manner is Bown and Crowley (2013b). These authors focus on variation in the level of applied protection under AD. Following the theoretical framework proposed by Bagwell and Staiger (1990), terms of trade shocks can lead countries to defect from their cooperative tariffs. The theory makes cross-sectional predictions about the likelihood of a breakdown in cooperation, but in the empirical study of Bown and Crowley (2013b) the only predictive variable that varies over time is import growth at the product-country level. In principle this would allow the authors to exploit more disaggregated data, but the model also relies on export supply elasticities that are only available at a higher level of aggregation. As a result, and because they also wish to consider industry level controls in the robustness exercises, Bown and Crowley (2013b) estimate their model for 283 industries that correspond roughly to HS4 digit international trade categories. Their data are also aggregated to the annual level. Relative to our aims of describing import dynamics and separating the named products from nearby neighbors in product space these sector definitions are still quite aggregated. In principle they aggregate over products that do and do not see an investigation, and they also aggregate over time periods that may and may not be relevant in predicting a filing. The cross- sectional focus of Bown and Crowley (2013b) also precludes an investigation of the role of cumulative impacts of imports from more than one source country. We address these deficiencies by utilizing both the HS-10 data associated with a petition and the month in which a petition was filed. The starting point is the motivation underlying the theoretical 8 A small literature uses detailed import data to document trade dynamics following the conclusion of an AD case. Prusa (1997) exploits product-level U.S. trade data, while Pierce (2011) adds a plant-level dimension. Konings and Vandenbussche (2013) investigate the impact of European AD cases using firm-level export data from France and product-level trade data from the European Union. Besedeˇ s and Prusa (2013) use a survival model to estimate the breaking of trade links following an US AD case. 6 models that economic shocks are the basis for filing behavior. However, we extend the notion to consider a setting with sectors that contain more than one product (not all of which are filed against) and more than one source country (again not all of which are filed against). The disaggregated nature of the data helps to isolate the likely source of the various shocks (foreign and specific or domestic and common to all sources and products) and also to address a number of issues on which the theoretical models are silent. For example; what happens if more than one shock occurs and how persistent do shocks need to be to induce filing?9 While our focus is on dissecting the import data, it is also possible that filing is primarily about political factors. For example, it may be that domestic characteristics (industry concentration, location in a marginal electorate, etc.) will be more important than changes in imports.10 Another set of arguments suggests that target countries are chosen in an international game of tit-for-tat. If such political factors dominate filing behavior then decomposing the import data is likely to yield little information. We focus on economic phenomena observable in the import data to provide insight into why some countries are filed against and not others, and why some products are filed against and not others. 3 Data To achieve these goals we utilize the information on the nature of the products named in the AD petitions and combine it with detailed U.S. trade data.11 Information on the individual anti-dumping cases is drawn from Bown (2012). This dataset contains information on the case numbers, countries filed against, HS codes and the date a petition is filed. The presence of the HS codes allows us to link this information to detailed U.S. trade data. Our trade data are the HS product level information on U.S. imports available in the monthly reports of the U.S. Imports of Merchandise Data.12 These data contain substantial information on the value and quantity of imports at the product level. From this value and quantity information we derive unit values, providing us with a very detailed dataset for analysis. We use Imports for Consumption, the classification of U.S. imports used in 9 The focus on supply and demand shocks is similar to that in Grossman (1986), Pindyck and Rotemberg (1987) and Kelly (1988). However, their methods and data differ from ours along with the specific questions they address. 10 We employ product-specific fixed effects, which should sweep out average differences in the underlying propensity to file. 11 Unfortunately the relevant trade data contained in the petition is not made available to the public. 12 Despite the public availability of these data they are relatively underutilized. Exceptions include Hummels and Schaur (2010), Chor and Manova (2012), and within the AD literature, Besedeˇ s and Prusa (2013). 7 U.S. antidumping investigations.13 The fields we employ include ‘Customs value’, which measures the value of imports at the foreign port. We typically identify prices by deflating these values by a quantity measure, usually the ‘first unit of quantity’ as reported in the data. Note that these data allow us to identify not only the products and countries that are subject to a petition, but also imports in the subject products from all sources (i.e. both named and non- named countries). This offers an opportunity to investigate the extent to which imports from the subject countries in the subject products are typical and in what ways they differ from the imports coming from other non-named sources. In addition we can identify a reference group of products that are similar to the subject products but that are not named in the petition. The group of related products serves as an important reference group in our attempt to understand why some products are subject to an AD petition and why others are not. In making these comparisons we track the value and quantity of imports in both named and reference products from subject and non-subject countries over the 24 months preceding the filing of an AD petition. One issue that arises when following HS codes over time is that definitions change relatively frequently (for example over 4,500 new HS-10 codes were introduced between January 1994 and January 1996).14 Since we analyze the shipments occurring in the previous 24 months, changes in product definitions can have a large impact on the amount of data available (for instance if the definition of a product changes in the month before an AD case is filed then only one month’s worth of data is available). To mitigate these problems we identify a period of time where the definitions of HS-10 products are relatively stable, 1998-2004.15 Consequently, we are able to track almost all HS-10 product codes associated with the AD cases filed between 1998 and 2004 for 24 months preceding the filing of an AD petition.16 We consider 106 investigations initiated in the 1998-2004 period.17 An investigation often in- cludes multiple countries, and each investigation-country pair is associated with the filing of a pe- 13 The U.S. also reports General Imports which measures total physical arrivals into the United States, including shipments bound for bonded warehouse or transshipment to other countries. 14 A single HS-10 code can be replaced by a number of new codes or it maybe amalgamated with a number of other HS-10 codes. In either case the identifier associated with the old HS-10 code is deleted. 15 This period is also of interest because it encompasses the adoption of the Byrd Amendment, a legal change that affected the incentives to file a petition. 16 To evaluate the impact of HS-10 changes that do occur within our sample, we use the Pierce and Schott (2012) concordance mapping HS-10 changes over time. Since very few commodities are affected, there is little effect on the results. The bridging techniques in Pierce and Schott (2012) are not well suited to our exercise, as they tend to expand the product scope of the listed HS-10 products. We report results from a more conservative approach, which involves dropping those commodities that appear or disappear within the two-year window prior to each case we consider. 17 111 investigations were initiated over this period. Three pairs of investigations are near duplicates, and we keep only one investigation of each pair. We also drop two investigations from the data set for lack of quantity information in the trade data. See the data appendix for a more complete discussion of our treatment of the data. 8 tition. In our data there are 306 petitions filed against 56 countries. The most frequently targeted countries in the sample are China (45 petitions), Japan (21), India (18) and the Republic of Korea (18). Altogether 139 (45%) of the petitions in 59 (56%) of the investigations resulted in an affir- mative determination.18 Of the 139 investigation, 32 (30%) involve steel products, while 65 (61%) occurred under Byrd Amendment rules. In terms of the U.S. import data, the sample includes 764 HS-10 products that are included in an investigation. The sample includes 113 HS-4 categories that contain commodities named in a petition. As we describe below, we use products within the affected HS-4 categories, but not named in a petition, as reference commodities against which to evaluate changes in the named commodities. There are 2,514 products within this group of reference commodities.19 4 Methodology 4.1 Research Design Since our objective is to isolate the distinctive characteristics of the products and countries named in an AD petition, we pursue a methodology based on the detail of the AD process. Consequently, our analysis is predicated not only on the very specific products and countries named in a petition but also the very specific filing date used to evaluate the merits of the claim (i.e. import competition is evaluated prior to this particular date). As a result, all three of these dimensions play a critical role in our research design. The quasi-judicial nature of the AD process requires a very precise definition of the products in question. The boundaries of the product space are defined sharply, excluding commodities that can be quite similar in design, construction and use. Following the terminology used in AD cases, the products included in the petition shall be identified as ‘subject products.’ The set of countries named in the petition (subject countries) is also typically limited to a subset of all national sources of the imported product. The distinctions between subject and non-subject products and between subject and non-subject countries are central to our analysis as they provide two natural comparisons that we exploit to isolate the unique features of the subject countries and products. A natural reference group for the subject imports are shipments of the same product from non- 18 Some investigations see affirmative determinations against only a subset of the countries that were filed against. 19 We experimented with the definition of related products by narrowing it to non-named commodities within the same HS-6 category. The results are almost identical. We report results from the HS-4 definition since it allows us to include all cases (some cases are defined at the HS-6 level and would not have a reference group at the HS-6 level). 9 subject countries (i.e. those not named in the petition). As noted previously, only a subset of countries supplying the U.S. market is actually named in a petition. The ability to compare the evolution of import values provides a perspective on why the subject countries are likely to be singled out. In particular, we are looking for any changes that are pronounced enough to potentially motivate the domestic industry to file an AD petition. Further detail is provided by making additional comparisons based on changes in unit values and imported quantity. This additional level of detail helps to provide a narrative in terms of the source of shocks since positive or negative correlations in price and quantity map into distinct sources of shocks (i.e. demand shocks are typically associated with a positive correlation while a negative correlation is more closely aligned with supply shocks). To address the question of why these products and not others, we turn to the comparison between subject and non-subject products. The architecture of the U.S. import statistics provides a natural basis for the comparison, as it identifies very good candidates for inclusion in the group of reference commodities. Products within the same HS-4 category as the subject products are treated as the relevant reference group.20 (The reference group contains only those products that were not included in a petition over the entire period of our analysis.) In the analysis of the reference products, the set of subject countries is the same as with subject products. Shocks that hit national economies within broader HS-4 category can then be observed as common movements across subject and non-subject commodities. Hence, differences in the evolution of subject and non-subject products in terms of value, price and quantity will provide insight into the question of why only certain products were targeted and not others. 4.2 Decomposition Since our inference is based on comparisons we adopt a decomposition methodology developed by Hummels and Klenow (2005) that facilitates comparisons both across groups and over-time.21 The decomposition technique involves two steps. First, the trade data within each month is summa- rized in terms of aggregates representing subject countries/products and the appropriate comparison group. Second, a log-linear regression is used to summarize common movements in the aggregates. Our regression technique is designed to isolate common, within commodity movements in the de- 20 As an example, the 2001 petition involving mussels from Canada included two HS-10 products (‘farmed mussels’ and ‘other mussels’) within the larger HS-4 category covering mollusks and aquatic invertebrates other than crustaceans that are fit for human consumption. There were 29 HS-10 products within the HS-4 category that were non-subject products, including HS-10 products that track imports of oysters, cuttlefish, octopus, snails and clams. 21 Other applications of this methodology include Hillberry and McDaniel (2002), Bernard et al. (2007) and Hillberry and Hummels (2008). 10 composition variables, and to summarize these movements in terms of cross-commodity shifts in a conditional mean. The top level decomposition allows a summary of shifts in subject countries’ import market share and total imports. The second level decomposition describes shifts in prices and quantities, which we interpret in terms of import supply and demand shocks. In the initial, top-level decomposition, imports from subject countries are represented as the product of the subject countries’ share of total imports and total imports: SMit SMit = Mit (1) Mit where i indicates the product, t indicates a time period, Mit indicates the total value of imports in product i at time t, and SMit is the value of imports from subject countries.22 While a direct comparison between imports from subject and non-subject countries is possible the decomposition methodology incorporates additional information. In particular, we norm subject imports against imports in subject products from all sources. Such a comparison not only isolates any distinctive behavior of subject imports (through changes in the subject-country market share) but it does so relative to a base that is directly responsive to changes in domestic market conditions. That is, while we may lack the data to directly measure domestic quantities, imports from all sources do embody elements of domestic market conditions. Hence, this comparison provides insight into differences between subject and non-subject sources as well as information on market level outcomes. The second level decomposition separates import value in terms of prices and quantities. Total trade Mit is represented as Mit Mit = Qit = Pit Qit (2) Qit where Pit is the (per unit) price, and Qit is measured quantities of imports. This decomposition can also be applied to imports from subject countries: SMit = SPit SQit . (3) Substituting (2) and (3) into (1) produces a representation of imports from subject countries as a 22 Products are defined at the lowest level of aggregation available in the US tariff classification. This is usually the HS-10 level, but in some cases is HS-6 or HS-8. The time subscript here represents a particular month (i.e. January 2000). 11 series of comparisons between each component of aggregate imports and its relative counterpart: SPit SQit SMit = Pit Qit (4) Pit Qit The first and second terms represent, respectively, the relative import price for subject countries and the import price from all sources. The third and fourth terms represent the relative subject country import quantity and the quantity of imports from all sources. The terms of each of these decompositions (SMit , Mit , etc.) are random variables that are observed in a given month, and in a given narrowly-defined commodity. A regression framework is useful for summarizing common movements in these variables over time. We adopt a simple OLS framework, regressing each of the (logged) variables from the decomposition on dummy vari- ables representing time periods prior to the filing date, with fixed effects included to account for heterogeneity associated with particular commodities and petitions, along with a number of other dimensions. The coefficients on the dummy variables representing time lags capture average shifts in the conditional mean of the LHS variables in question. Let Xikst represent one of the random variables from the decomposition. Our regression takes the form: 8 X ln(Xikst ) = βs lags + αik + ηt + yr + iks , (5) s=1 where, lags is a quarterly dummy capturing three months in the sth lag from the filing date.23 αik is a dummy variable indicating the HS-10 code-case pairing (with k indicating the ‘product’ as defined in the dumping case), and ηt is a dummy variable associated with a month, where t represents a month.24 yr is a dummy for the year, while iks is assumed to be an i.i.d. error term. X coefficients. These parameters measure shifts, over The key parameters of interest are the βs time, in the conditional mean of the left hand side variable. The reference period is the eighth X represent an increase in the level of the given X quarter before the filing. Positive values of βs variable, relative to its level in the initial period. Changes in these parameters will be understood as representing changes in market fundamentals, and interpreted as the outcome of economic shocks. 23 We retain the monthly detail in the data, but report quarterly averages defined by the coefficients on quarterly dummy variables. Import shipments are lumpy when observed at this level of detail. The move to quarterly averages smooths out this lumpiness without requiring an aggregation of the data over time. Quarterly movements are still sufficient to represent the relevant dynamics of each variable. 24 The product described in an anti-dumping petition frequently encompasses more than one HS-10 code. αik controls for cross-commodity differences at the HS-10 level, while ηt captures seasonal changes across the entire set of HS-10 commodities that are included in the sample. 12 So, for example, when the average relative price of subject country imports falls while relative quantities rise, we shall interpret this change as the outcome of a supply shock. As noted in Hummels and Klenow (2005), log-linear regressions of this sort have a quite useful SM be the estimated coefficient associated with property. Consider the components of (1). Let βs M the estimated coefficient associated with the M term, and β sm the the sth lag of the SM term, βs s coefficient associated with the share of subject imports ( SM M ). Independent regressions of each of these terms using (5) will produce estimated coefficients with the property that the coefficients on the SM = β M + β sm . two right hand side variables will add up to the coefficient on the left hand variable, βs s s SM ) can be decomposed in Consequently changes in the left hand side variable (summarized by βs terms of shares attributable to each of the components on the right hand side. This feature applies to all the decompositions calculated. 5 Results 5.1 Decomposing Imported Values Given the targeted nature of an AD petition, in terms of products and countries, we focus our analysis on behavior that might clearly distinguish the countries that are filed against from those that are not. Concentrating on products that are named in a petition, equation (1) decomposes the changes in the value of subject country imports into changes in common with all source countries and also changes in the share of imports associated with the subject countries. This allows us to understand the degree to which filers are responding to a general surge of imports (all imports increase together), or to specific pressure from the subject importers (a change in the share of imports from subject countries). Since we are pooling across all cases initiated between 1998 and 2004, the coefficients provide estimates for the typical HS-10 commodity associated with a petition. The results of these decompositions appear in Table 1. Column (1) reveals that there is a significant increase in the value of subject imports during the two-year window preceding the filing of a petition. The β SM coefficient on the dummy variable associated with the last quarter before the filing date indicates, on average, an 11 percent increase in subject country imports across all HS-10 codes that were part of an AD petition. The timing of this increase is also quite distinct, beginning about five quarters before the filing, peaking about two quarters before the filing, and then subsiding to the level of an 11 percent increase. In any case, the regressions show a sustained 13 Table 1: Just a surge from subject countries or from all sources? (1) (2) (3) VARIABLES ln(SMit ) = ln( SM Mit ) it + ln(Mit ) 1 Qtr before file 0.11*** 0.19*** -0.08*** (0.030) (0.023) (0.022) 2 Qtrs before file 0.16*** 0.19*** -0.03 (0.029) (0.023) (0.021) 3 Qtrs before file 0.14*** 0.14*** -0.00 (0.029) (0.023) (0.021) 4 Qtrs before file 0.13*** 0.08*** 0.05** (0.028) (0.022) (0.020) 5 Qtrs before file 0.11*** 0.10*** 0.01 (0.029) (0.023) (0.020) 6 Qtrs before file 0.04 0.08*** -0.03 (0.029) (0.023) (0.021) 7 Qtrs before file -0.04 0.02 -0.06*** (0.030) (0.024) (0.021) Constant 12.68*** -1.43*** 14.11*** (0.033) (0.026) (0.024) Observations 16596 16596 16596 R-squared 0.839 0.805 0.889 Estimates include month, year, petition and HS10 fixed effects Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 increase in the level of imports from subject countries over the 24 month window preceding a filing. While increased subject country imports might be suggestive of a damaging import surge, without a benchmark it is not obvious that this increase in imports is associated with an increase in import competition. To provide some context column (3) reports the average behavior of imports from all sources of the HS-10 products named in a petition. As is evident, there is no tendency for total imports to rise. Indeed, in the quarter before the filing there is a general tendency for imports to decline, falling 8 percent below the benchmark level, on average. Since total imports from all sources is a market level characteristic it provides insight into conditions for the market as a whole. One potential interpretation is that the products in which AD petitions are filed are likely to have experienced a negative demand shock which would account for the pronounced drop in spending on total imports of subject commodities immediately before a petition is filed. The contrast in outcomes in columns (1) and (3) is very stark. Column (2) confirms that the underlying compositional change is very dramatic, with the subject countries’ imports growing over 14 the 24 month window by 19 percent more than all sources. Some back-of-the-envelope calculations put this shift in context. The simple average subject-country market share across the subject HS-10 products in the reference period, quarter 8, is 0.47. A 19 percent increase in market share over the 24 month window increases the average market share to approximately 0.47 ∗ e0.19 = 0.57. Thus, it appears that subject country imports typically represent a significant and growing share of total imports of subject commodities. An 11 percent increase in imports from subject countries at the same time overall imports in the named products drop by 8 percent clearly indicate that subject countries look very different to other importers immediately before a petition is filed. This suggests that the subject countries are singled out for a particular reason. However, a focus on this quarter alone neglects an important phenomenon apparent in Table 1. In particular, Table 1 provides insight into the relative timing of the increase in subject imports and the decline in imports from all sources. Five quarters before filing there is a pronounced shift toward the named countries, while imports from all sources are relatively constant. This indicates that the compositional switch pre-dates the decline in expenditure on total imports in the quarter before filing. Is this differential enough to warrant an AD petition being filed against these countries in these products? While the differential apparent in Table 1 is suggestive, it is also possible that the same features occur in other products but nonetheless these products were not subject to an AD petition. If this is the case, then it begs the question of why AD petitions were not filed in these related products as well. To gain insight into how typical these patterns are we replicate the analysis on a related set of products. Specifically, define related products as those that are in the same HS-4 product group but are not subject to AD activity over the sample period. To complete the analysis we allocate subject and non-subject countries in this group in a way that follows the original demarcation contained in the subject products. That is, if a country is named in a petition for a product, then we treat the same country as potentially subject for the related product. If the underlying drivers are relatively broad in both the subject country and within the U.S., then the pattern apparent in Table 1 will also be evident in this reference group of related products. Table 2 presents the results of this decomposition. Similar to Table 1 the increase in imports from the subject countries is consistent with a foreign advantage that boosts imports. In fact the increase in imports in the reference products is even greater than documented in Table 1. This suggests a relatively general shock in the subject countries rather than a narrow product-specific export 15 Table 2: Are there compound shocks in related products? (1) (2) (3) 4) = SM 4 4) VARIABLES ln(SMit ln( M 4it ) + ln(Mit it 1 Qtr before file 0.17*** 0.10*** 0.07*** (0.021) (0.019) (0.014) 2 Qtrs before file 0.15*** 0.09*** 0.06*** (0.021) (0.019) (0.014) 3 Qtrs before file 0.12*** 0.07*** 0.04*** (0.021) (0.019) (0.014) 4 Qtrs before file 0.10*** 0.03* 0.07*** (0.020) (0.019) (0.013) 5 Qtrs before file 0.09*** 0.02 0.06*** (0.021) (0.019) (0.013) 6 Qtrs before file 0.04* 0.00 0.03** (0.021) (0.019) (0.014) 7 Qtrs before file 0.00 0.00 0.00 (0.021) (0.019) (0.014) Constant 11.66*** -2.02*** 13.68*** (0.022) (0.021) (0.015) Observations 35328 35328 35328 R-squared 0.814 0.794 0.912 Estimates include month, year, petition and HS-10 fixed effects Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 shock. However, the second feature (the pronounced decline in total imports) is not evident in the reference products. Inspection of column (3) reveals that imports from all sources are consistently higher than in the base period. Allowing ourselves to speculate, this suggests that AD petitions are associated with very narrow, product-specific demand shocks, rather than economy- or sector-wide demand shocks. Figure 1 plots these coefficients to make this point graphically. While the blue lines (imports from subject countries) mirror each other in the two graphs, there is no such tendency for the red lines (total imports). Consequently an import surge by itself does not seem to be sufficient to trigger an AD petition. However, a relatively modest increase in imports when accompanied by a subsequent domestic demand shock (as proxied by the level of imports from all sources), does appear to be necessary for the filing of an AD petition. 16 All Subject Products Non-Subject Products .2 .2 .1 .1 filing filing 0 0 -.1 -.1 -25 -20 -15 -10 -5 0 -25 -20 -15 -10 -5 0 months months subject countries all countries subject countries all countries Figure 1: Subject and all sources imports: HS-10 vs HS-4 17 Table 3: Is the surge driven by lower prices or higher quantities? (1) (2) (3) (4) (5) VARIABLES ln(SMit ) = ln( SP Pit ) it + ln(Pit ) + ln( SQit Qit ) + ln(Qit ) 1 Qtr before file 0.11*** -0.03** -0.03*** 0.22*** -0.06** (0.03) (0.01) (0.01) (0.03) (0.02) 2 Qtrs before file 0.16*** -0.03** -0.02* 0.22*** -0.01 (0.03) (0.01) (0.01) (0.03) (0.02) 3 Qtrs before file 0.14*** -0.04*** -0.01 0.18*** 0.01 (0.03) (0.01) (0.01) (0.03) (0.02) 4 Qtrs before file 0.13*** -0.03** -0.00 0.12*** 0.05** (0.03) (0.02) (0.01) (0.03) (0.02) 5 Qtrs before file 0.11*** -0.01 -0.00 0.11*** 0.01 (0.03) (0.02) (0.01) (0.03) (0.02) 6 Qtrs before file 0.04 -0.01 -0.01 0.09*** -0.02 (0.03) (0.02) (0.01) (0.03) (0.02) 7 Qtrs before file -0.04 0.01 0.00 0.01 -0.06** (0.03) (0.02) (0.01) (0.03) (0.02) Observations 16596 16596 16596 16596 16596 R2 0.839 0.583 0.979 0.745 0.925 Estimates include month, year, petition and HS-10 fixed effects Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 5.2 Decomposition of Imports: Prices and Quantities To this point the interpretation of demand and supply shocks as underlying the behavior of subject imports and total imports is relatively provisional since we have only considered information on expenditures. To scrutinize this interpretation we further decompose the behavior of imports into changes in price and quantity. These results appear in Table 3. Is the decline in total imports immediately before the filing likely to be due to a negative demand shock? It appears that this is the case with evidence from the first row of columns (3) and (5) of Table 3 confirming a decline in both average unit price and average quantity from all sources in the quarter immediately before filing (i.e. in equilibrium both price and quantity have declined).25 The picture that emerges for imports from all sources is consistent with a negative demand shock 25 Because this shock affects imports from all sources it is likely that its origins are domestic. Observed movements in price and quantity are consistent with a domestic demand shock. There is scope for other explanations on the supply side, however, such as the breakdown of collusive arrangements in the domestic market. We return to this argument later. 18 - import price and quantity variables both fall just prior to the filing. However, if one were to focus instead on just the subject countries then a very different story applies. In fact the behavior of prices and quantities from these sources is consistent with a positive export supply shock - a price decline of 6 percent (derived from (2)+(3)) and an increase in quantity of 16 percent (derived from (4)+(5)) in the quarter immediately before filing.26 This suggests that the subject countries receive a persistent favorable export supply shock that more than offsets the negative domestic demand shock. Moreover, the interaction of these two shocks generate market outcomes that are prima facie consistent with the criteria for an AD petition - aggressive pricing by foreign firms (from the export supply shock) that is correlated with domestic injury (potentially associated with the negative demand shock). The relative timing of these shocks is also informative. If one takes the initial emergence of the foreign export supply shock as approximately 15 months before filing, then it seems that the transmission of the shock into the domestic market is handled in such a way as to keep the overall extent of foreign competition relatively constant. That is, columns (3) and (5) indicate that aggregate imported quantities and prices are essentially unchanged from five quarters before the filing through the penultimate quarter before filing. However, a negative demand shock in the final quarter before filing places the foreign export supply shock into a harsher light for the domestic industry.27 This suggests that it is not necessarily a miscalculation on the part of foreign firms that induces an AD petition, but rather a subsequent demand shock which changes domestic circumstances for the worse. Table 4 provides the same decomposition of the value of imports into price and quantity compo- nents for the related products that were not filed against. The analysis of imported values suggested that the major difference between the subject products and the related products was the absence of a domestic demand shock amongst related products. This is confirmed by columns (3) and (5) which show that quantities shipped from all sources are uniformly higher than in the base period. Moreover, the average price is also lower. These outcomes are more consistent with a pronounced supply shock and there is little evidence of a demand shock. This speaks directly to the breadth of the various shocks, with the relative supply shock apparent in both subject and related products for subject countries, but the negative demand shock is only documented in the subject products. Imports from the countries that are filed against do seem to exhibit characteristics that might make 26 There may be many potential sources of the relative export supply shock. For example, Debaere (2010) documents how changes in third country standards can motivate a relative export shock. Other potential explanations include foreign demand shocks (Staiger and Wolak (1992), Blonigen and Wilson (2010)) and foreign supply shocks (Clarida (1993)). 27 An analysis using monthly dummies suggests that the negative domestic shock becomes evident sometime in the second quarter prior to the filing. 19 Table 4: Are similar shocks evident in related but non-subject products? (1) (2) (3) (4) (5) VARIABLES 4) = ln(SMit ln( SP Pit ) it + ln(Pit ) + ln( SQit Qit ) + ln(Qit ) 1 Qtr before file 0.17*** -0.03* -0.04*** 0.13*** 0.10*** (0.02) (0.01) (0.01) (0.02) (0.02) 2 Qtrs before file 0.15*** 0.00 -0.03*** 0.09*** 0.09*** (0.02) (0.01) (0.01) (0.02) (0.02) 3 Qtrs before file 0.12*** -0.00 -0.03*** 0.08*** 0.07*** (0.02) (0.01) (0.01) (0.02) (0.02) 4 Qtrs before file 0.10*** 0.01 -0.03*** 0.02 0.10*** (0.02) (0.01) (0.01) (0.02) (0.02) 5 Qtrs before file 0.09*** 0.00 -0.02** 0.02 0.08*** (0.02) (0.01) (0.01) (0.02) (0.02) 6 Qtrs before file 0.04* -0.01 -0.00 0.01 0.04** (0.02) (0.01) (0.01) (0.02) (0.02) 7 Qtrs before file 0.00 -0.01 -0.02** 0.01 0.02 (0.02) (0.01) (0.01) (0.02) (0.02) Observations 35328 35328 35328 35328 35328 R2 0.814 0.567 0.963 0.743 0.940 Estimates include month, year, petition and HS-10 fixed effects Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 20 Table 5: Are similar shocks evident in Steel products? (1) (2) (3) (4) (5) VARIABLES ln(SMit ) = ln( SP Pit ) it + ln(Pit ) + ln( SQit Qit ) + ln(Qit ) 1 Qtr before file 0.14*** -0.04** -0.03*** 0.28*** -0.06* (0.04) (0.02) (0.01) (0.04) (0.03) 2 Qtrs before file 0.18*** -0.04** -0.00 0.28*** -0.06* (0.04) (0.02) (0.01) (0.04) (0.03) 3 Qtrs before file 0.15*** -0.05** -0.00 0.23*** -0.03 (0.04) (0.02) (0.01) (0.04) (0.03) 4 Qtrs before file 0.13*** -0.06*** -0.01 0.18*** 0.02 (0.04) (0.02) (0.01) (0.04) (0.03) 5 Qtrs before file 0.16*** -0.04* -0.01 0.19*** 0.02 (0.04) (0.02) (0.01) (0.04) (0.03) 6 Qtrs before file 0.05 -0.03 -0.02 0.13*** -0.03 (0.04) (0.02) (0.01) (0.04) (0.03) 7 Qtrs before file -0.06 0.00 0.00 0.02 -0.09*** (0.04) (0.02) (0.01) (0.04) (0.03) Observations 10294 10294 10294 10294 10294 R2 0.791 0.567 0.942 0.696 0.889 Estimates include month, year, petition and HS-10 fixed effects Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 them likely candidates for an AD petition, but a product specific negative demand shock appears to be an important factor in the choice of subject products. 5.3 Robustness Exercises The previous section argued that at least two shocks precede an AD petition: a broad and sustained relative supply shock favoring exports from subject countries, followed by a relatively narrow demand shock that hits imports from all sources. These results were based on pooled data, so it is of interest to examine whether these patterns are evident amongst cases with different characteristics. With this in mind we consider two broad subsamples, one based on product characteristics and the other based on a temporal consideration. In both cases the results presented are based on estimates associated with equation (4). Consider first a subsample limited to cases involving the steel industry, which is a heavy user of the anti-dumping law. Studies of anti-dumping activity have shown differences between cases 21 Table 6: Are similar shocks evident under the Byrd Amendment? (1) (2) (3) (4) (5) VARIABLES ln(SMit ) = ln( SP Pit ) it + ln(Pit ) + ln( SQit Qit ) + ln(Qit ) 1 Qtr before file 0.05 -0.02 0.00 0.18*** -0.11*** (0.04) (0.02) (0.01) (0.04) (0.03) 2 Qtrs before file 0.08** -0.01 0.00 0.15*** -0.07** (0.04) (0.02) (0.01) (0.04) (0.03) 3 Qtrs before file 0.10** -0.05** 0.01 0.17*** -0.02 (0.04) (0.02) (0.01) (0.04) (0.03) 4 Qtrs before file 0.11*** -0.03 0.02 0.09** 0.04 (0.04) (0.02) (0.01) (0.04) (0.03) 5 Qtrs before file 0.09** 0.02 0.02 0.05 0.01 (0.04) (0.02) (0.01) (0.04) (0.03) 6 Qtrs before file 0.00 -0.01 -0.01 0.07** -0.05 (0.04) (0.02) (0.01) (0.04) (0.03) 7 Qtrs before file -0.07* 0.01 0.01 -0.03 -0.07** (0.04) (0.02) (0.01) (0.04) (0.03) Observations 8921 8921 8921 8921 8921 R2 0.861 0.567 0.980 0.789 0.923 Estimates include month, year, petition and HS-10 fixed effects Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 involving the steel industry and other cases. We break out cases involving the steel industry, which constitute 34 of the 106 cases we consider. Table 5 presents the results, which are consistent with the compound shock thesis. The pronounced relative supply shock is evident from six quarters prior to filing with higher relative quantities imported (column 4) and lower relative prices (column 2). Nevertheless the average import price and quantity remain relatively constant until the penultimate quarter before filing. In contrast, in the quarter before filing the average steel product experiences a decline in import prices of 3 percent and a decline in import volume of 6 percent for subject steel products. This strongly suggests that both a relative export supply shock followed by a sharp decline in domestic demand precede the filing of AD petitions involving steel products.28 Consider next a period defined by a significant change in AD law. The Continued Dumping and Subsidy Offset Act of 2000, also known as the “Byrd Amendment,” significantly changed the 28 We ran the decompositions over all non-steel cases and found evidence of the compound shock pattern that is consistent with what we found in the aggregate sample and in the steel subsamples. 22 incentives to file anti-dumping cases, as it directed the collected tariff revenues to the filing firms.29 We focus on cases filed after January 1, 2001, the day the Act went into effect.30 The results are contained in Table 6 and are once again are broadly consistent with the compound shock thesis. A relative supply shock is evident in column (4), though its origin appears to be slightly later than documented for steel products. The negative domestic demand shock is also documented with a decline in imports from all sources of 11 percent. Evidence from the steel and Byrd cases illustrates that the compound shock thesis is consistent across important subsamples. Of course, deviations from the pattern identified above could occur within other subsamples. Nonetheless, these two seem to be natural and obvious subsamples in which to evaluate the robustness of the pattern observed in the complete sample. On the basis of these exercises, the compound shock thesis appears to be a robust aspect of filing behavior. 6 Conclusion While the global trading system generally supports open trading arrangements, it also allows room for substantial policy exceptions. These exceptions are quite narrow, and can be applied in a discriminating manner. The narrowness of the product space in which policy can be applied, and the discriminating use allows a research design that exploits relative differences in import dynamics - across countries and across similar products - to relate changes in import competition to industry demands for protection. Knowing more about the use of exceptions can help us better understand the pressures that drive trade policy, and perhaps even the benefits of the wider system of restraint. Our study of filing behavior in U.S. AD policy indicates that domestic industries’ demands for protection, in the form of AD petitions, typically follow a pattern of compound shocks. Approx- imately 15 months prior to the filing of a petition, an initial positive relative supply shock favors exports from those countries that will subsequently be named in a petition, but not exports from other countries. In terms of product space this shock is a broad one, hitting the products named 29 Reynolds (2006) finds evidence that the Byrd Amendment increased the number of petitions filed. Reynolds and Liebman (2006) identify a relationship between firms that gain from the amendment, their contributions to members of Congress and members’ support for the Amendment. Reynolds and Liebman (2006) contains a description of the amendment as well as a legislative history. 30 The sample split captures more than the change in the law of course. There is a change in presidential administra- tions at this time. There are also changes in macroeconomic phenomena such as a reduced average GDP growth rate and a shift in the direction of the USD exchange rate that roughly coincide with this split. The subsample considered also contains relatively few steel cases, as the steel safeguard action implemented in March of 2002 limits the number of AD cases in steel. Our sample split is thus a rough robustness check against the possibility that any of these phenomena, including the Byrd Amendment, might affect our results. 23 in the petition, but also other products in the HS4 that are not named. The second shock, a nega- tive import demand shock, hits roughly 3 months prior to the filing. The second shock is notably different than the first in that a) it hits the product in question but not the other products in the HS4 category, and b) it affects all imports, not only those from the countries that are subsequently named in the petition. The initial relative supply shock appears similar to those identified elsewhere in the literature, but the later, narrow import demand shock has not been identified empirically as an important trigger of AD. We are able to observe this second shock because we exploit the granularity of U.S. import data. In particular, we use these data to observe changes over shorter times and narrower product classes than other authors in the literature. A limitation of exploiting these data is that they do not have an analogue in domestic production, so we cannot make statements about changes in domestic supply. Nonetheless we allow ourselves to speculate. The second shock affects imports from all sources, and therefore might well have a domestic origin. The narrowness of the shock in product space may suggest that a positive domestic supply shock is more viable than a negative domestic demand shock as a source of the import demand shock. One possible explanation of this pattern, similar to that in Feinberg (1989), is that the initial shock unsettles domestic collusive arrangements, and that the breakdown hits a narrow set of products, with the AD case employed as an attempt to restore the collusive order. This is one potential mechanism, others are also possible. We leave these questions for future research. 24 References Bagwell, Kyle, and Robert W Staiger (1990) ‘A theory of managed trade.’ American Economic Review 80(4), 779–95 Bagwell, Kyle W, and Robert W Staiger (1999) ‘An economic theory of GATT.’ The American Economic Review 89(1), 215–248 (2002) The economics of the world trading system (Cambridge, USA: MIT Press) (2012) ‘Profit shifting and trade agreements in imperfectly competitive markets.’ International Economic Review 53(4), 1067–1104 Bernard, Andrew B., J. Bradford Jensen, Stephen J. Redding, and Peter K. Schott (2007) ‘Firms in international trade.’ Journal of Economic Perspectives 21(3), 105–130 s, Tibor, and Thomas J. Prusa (2013) ‘Antidumping and the death of trade.’ National Bureau Besedeˇ of Economic Research Working Papers No. 19555 Blonigen, Bruce A. (2000) ‘US antidumping filings and the threat of retaliation.’ manuscript Blonigen, Bruce A., and Wesley W. Wilson (2010) ‘Foreign subsidization and excess capacity.’ Jour- nal of International Economics 80(2), 200–211 Bown, Chad P. (2012) ‘Global antidumping database version 5.0.’ available at http://econ.worldbank.org/ttbd/gad/ Bown, Chad P, and Meredith A Crowley (2013a) ‘Import protection, business cycles, and exchange rates: Evidence from the great recession.’ Journal of International Economics 90(1), 50–64 (2013b) ‘Self-enforcing trade agreements: Evidence from time-varying trade policy.’ The American Economic Review 103(2), 1071–1090 Chor, Davin, and Kalina Manova (2012) ‘Off the cliff and back? Credit conditions and international trade during the global financial crisis.’ Journal of International Economics 87(1), 117–133 Clarida, Richard H. (1993) ‘Entry, dumping, and shakeout.’ The American Economic Review 83(1), 180–202 Crowley, Meredith A. (2009) ‘Cyclical dumping and US antidumping protection: 1980-2001.’ Work- ing paper, Federal Reserve Bank of Chicago Debaere, Peter (2010) ‘Small fish–big issues: the effect of trade policy on the global shrimp market.’ World Trade Review 9, 353–374 Feinberg, Robert M. (1989) ‘Imports as a threat to cartel stability.’ International Journal of Indus- trial Organization 7(2), 281–288 Feinberg, Robert M., and Barry T. Hirsch (1989) ‘Industry rent seeking and the filing of ‘unfair trade’ complaints.’ International Journal of Industrial Organization 7(3), 325–340 Finger, J. Michael (1981) ‘The industry-country incidence of less than fair value; Cases in US import trade.’ In ‘Export Diversification and the New Protectionism: The Experience of Latin America’ NBER Chapters (National Bureau of Economic Research, Inc) pp. 260–279 Furusawa, Taiji, and Thomas J. Prusa (1996) ‘Antidumping enforcement in a reciprocal model of dumping: Theory and evidence.’ manuscript 25 Gelman, Andrew, and Guido Imbens (2013) ‘Why ask why? Forward causal inference and reverse causal questions.’ National Bureau of Economic Research Working Papers No. 19614 Grossman, Gene M. (1986) ‘Imports as a cause of injury: The case of the US steel industry.’ Journal of International Economics 20(3-4), 201–223 Hansen, Wendy L. (1990) ‘The International Trade Commission and the politics of protectionism.’ The American Political Science Review 84(1), 21–46 Hartigan, James C. (1996) ‘Predatory dumping.’ Canadian Journal of Economics pp. 228–239 Herander, Mark G., and J. Brad Schwartz (1984) ‘An empirical test of the impact of the threat of US trade policy: The case of antidumping duties.’ Southern Economic Journal 51(1), 59–79 Hillberry, Russell, and Christine McDaniel (2002) ‘A decomposition of North American trade growth since NAFTA.’ USITC Working Paper 2002-12-A Hillberry, Russell, and David Hummels (2008) ‘Trade responses to geographic frictions: A decom- position using micro-data.’ European Economic Review 52(3), 527–550 Hummels, David, and Peter J. Klenow (2005) ‘The variety and quality of a nation’s exports.’ Amer- ican Economic Review 95(3), 704–723 Hummels, David L., and Georg Schaur (2010) ‘Hedging price volatility using fast transport.’ Journal of International Economics 82(1), 15–25 Kelly, Kenneth (1988) ‘The analysis of causality in escape clause cases.’ The Journal of Industrial Economics 37(2), 187–207 Knetter, Michael M., and Thomas J. Prusa (2003) ‘Macroeconomic factors and antidumping filings: Evidence from four countries.’ Journal of International Economics 61(1), 1–17 Konings, Jozef, and Hylke Vandenbussche (2013) ‘Antidumping protection hurts exporters: Firm level evidence from France.’ Review of World Economics 149(2), 295–320 Krupp, Corinne (1994) ‘Antidumping cases in the US chemical industry: A panel data approach.’ The Journal of Industrial Economics 42(3), 299–311 Lichtenberg, Frank, and Hong Tan (1994) ‘An industry-level analysis of import relief petitions filed by US manufacturers, 1958-1985.’ In Troubled industries in the United States and Japan, ed. Hong Tan and Haruo Shimada (St Martin’s Press) pp. 161–188 Pierce, Justin R (2011) ‘Plant-level responses to antidumping duties: Evidence from US manufac- turers.’ Journal of International Economics 85(2), 222–233 Pierce, Justin R., and Peter K. Schott (2012) ‘Concording U.S. harmonized system categories over time.’ Journal of Official Statistics 28(1), 53–68 Pindyck, Robert S, and Julio J Rotemberg (1987) ‘Are imports to blame? Attribution of injury under the 1974 trade act.’ Journal of Law & Economics 30(1), 101–22 Prusa, Thomas J. (1997) ‘The trade effects of US antidumping actions.’ In The Effects of US Trade Protection and Promotion Policies, ed. Robert C. Feenstra (University of Chicago Press) pp. 191– 213 Reynolds, Kara M. (2006) ‘Subsidizing rent-seeking: Antidumping protection and the Byrd Amend- ment.’ Journal of International Economics 70(2), 490–502 26 Reynolds, Kara M., and Benjamin J. Liebman (2006) ‘The returns from rent-seeking: Cam- paign contributions, firm subsidies, and the Byrd Amendment.’ Canadian Journal of Economics 39(4), 1345–1369 Sabry, Faten (2000) ‘An analysis of the decision to file, the dumping estimates, and the outcome of antidumping petitions.’ The International Trade Journal 14(2), 109–145 Saggi, Kamal (2006) ‘Preferential trade agreements and multilateral tariff cooperation.’ International Economic Review 47(1), 29–57 Saggi, Kamal, and Nese Sara (2008) ‘National treatment at the WTO: The roles of product and country heterogeneity.’ International Economic Review 49(4), 1365–1394 Staiger, Robert W., and Frank A. Wolak (1992) ‘The effect of domestic antidumping law in the presence of foreign monopoly.’ Journal of International Economics 32(3-4), 265–287 Staiger, Robert W., and Frank A. Wolak (1994) ‘Measuring industry-specific protection: Antidump- ing in the United States.’ Brookings Papers on Economic Activity. Microeconomics 1994, 51–118 27 7 Data Appendix In this appendix we detail the decisions that were made in the construction of the dataset. We analyze the AD petitions filed between 1998 and 2004. Over this period there were 315 AD petitions filed, where a petition consists of a country-case pair. In terms of International Trade Commission investigation numbers this covers the sequence from 731-TA-776 to 731-TA-1089. As described above, the process of cumulation allows the International Trade Commission to combine petitions for the purpose of determining injury. This process is governed by the similarity of the products involved and the filing of the petitions on the same date (amongst other considerations). Through this process there were 111 injury investigations initiated over this period. However, not all of these cases satisfy the criteria for inclusion in our dataset. Two cases we dropped since, for administrative reasons, the original case was dropped and a very similar petition was almost instantly refiled (relating to the same products and countries). To avoid double counting due to this refiling the following cases were eliminated: • 44’-Diamino-22’-Stilbenedisulfonic Acid Chemistry • Live Cattle In addition two very similar investigations were conducted simultaneously into “Certain Tissue Paper Products” and also “Crepe Paper Products” covering an almost identical set of products and an identical set of countries. To avoid double counting we only include “Crepe Paper Products” in the dataset. Finally two other cases were dropped since no quantity information was available in the U.S. trade data. These two cases are: • Ironing Tables And Certain Parts Thereof • Aperture Masks A critical component of the analysis is an ability to link an AD petition with the relevant trade data. This exploits the listing of cases and HS codes in Bown (2012). However, for a few cases the HS codes listed are not consistent across countries. This occurs because some countries are dropped after the initial investigation and the definition of subject “product” is subsequently varied. For the sake of consistency, the definition of the subject “product” in the initial investigation is applied to all subject countries. The affected cases are: • Honey • Magnesium • Structural Steel Beams • Certain Circular Welded Carbon Quality Line Pipe • Foundry Coke The table below lists the 106 cases that are included in the dataset. 28 Case Year Case Year Certain Preserved Mushrooms 1998 *Stainless Steel Bar 2001 Butter Cookies 1998 *Welded Large Diam. Line Pipe 2001 *Stainless Steel Round Wire 1998 Folding Gift Boxes 2001 Extruded Rubber Thread 1998 Auto Glass Windshields 2001 *Stainless Steel Plate In Coils 1998 Oleoresin Paprika 2001 Emulsion Styrene-Butadiene Rubber 1998 Mussels 2001 *Stainless Steel Sheet And Strip 1998 Greenhouse Tomatoes 2001 *Elastic Steel Sheet And Strip 1998 Spring Table Grapes 2001 *Hot Rolled Carbon Steel Flat Products 1998 Softwood Lumber 2001 Live Cattle 1998 Silicomanganese 2001 DRAMS One Megabit & Above 1998 Folding Metal Tables & Chairs 2001 Creatine Monohydrate 1999 Polyethylene Terephthalate Film 2001 *Cut-To-Length Carb Steel Plate 1999 *Structural Steel Beams 2001 Polyester Staple Fiber 1999 *Circ-Weld Non-Alloy Steel Pipe 2001 Nitrile Rubber 1999 Indiv Frozen Red Raspberries 2001 Aspirin 1999 Gum Arabic 2001 *Cold-Rolled Carbon Steel Products 1999 Blast Furnace Coke 2001 Non-Frozen Apple Jce Concen 1999 *Carbon & Alloy Steel Wire Rod 2001 *Crude Petroleum Oil Products 1999 *Cold-Rolled Steel Products 2001 *Small Diameter Pressure Pipe 1999 Sulfanilic Acid 2001 *Large Diameter Pressure Pipe 1999 Ferrovanadium 2001 Synthetic Indigo 1999 Pneumatic Directional Cntrl Vlves 2002 *Structural Steel Beams 1999 *Ball Bearings 2002 Ammonium Nitrate 1999 *Non-Mall Cast Iron Pipe Ftgs 2002 Paintbrushes 1999 *Silicon Metal 2002 *Seamless Stnless Steel Hollow 1999 *Oil Ctry Tubular Goods 2002 *Tin Mill Products 1999 Urea Ammonium Nitrate Solution 2002 Expnble Polystyrene Resins 1999 *Lawn & Garden Steel Fence Posts 2002 Citric Acid And Sodium Citrate 1999 Coldwater Pink Shrimp 2002 *Butt-Weld Pipe Fittings 2000 Frozen Fish Fillets 2002 *Steel Wire Rope 2000 Saccharin 2002 *Steel Concrete Rebar 2000 Polyvinyl Alcohol 2002 Anhydrous Sodium Sulfate 2000 Durum & Hard Red Spring Wheat 2002 Desktop Note Count & Scan 2000 Barium Carbonate 2002 *Stainless Steel Angles 2000 *Malleable Iron Pipe Ftgs 2002 Foundry Coke 2000 Refined Brown Aluminum Oxide 2002 Honey 2000 Crtn Ceramic Station Post Insltrs 2003 Ammonium Nitrate 2000 *Prestress Concrete Steel Wire Strnd 2003 Pure Magnesium 2000 Allura Red Coloring 2003 *Hot-Rolled Carbon Steel 2000 44’-Diamino-22’- Acid Chem 2003 Low Enriched Uranium 2000 Hydraulic Mag Circuit Breakers 2003 * denotes a steel investigation * denotes a steel investigation 29 Case Year *Lt-Walled Rect. Pipe And Tube 2003 *Certain Aluminum Plate 2003 Certain Processed Hazelnuts 2003 Wooden Bedroom Furniture 2003 Hand Trucks 2003 Carbazole Violet Pigment 23 2003 Kosher Chicken 2003 Crt Frz & Can Wmwater Shrimp 2004 Outboard Engines 2004 Crepe Paper Products 2004 Magnesium 2004 *Crt Circ Weld Carb Qly Line Pipe 2004 Live Swine 2004 Pet Resin 2004 *Silicon Metal 2004 Chlorinated Isocyanurates 2004 Purified Carboxymethylcellulose 2004 Polyvinyl Alcohol 2004 * denotes a steel investigation 30