In 2020, several members of the United States Congress wrote a joint letter to the US International Trade Commission in support of a petition for anti-dumping duties filed by mattress manufacturers. The mattress manufacturers had requested new trade barriers to shield them from foreign competition. The letter noted that these manufacturers had a long history in the US, securing jobs for over a century—a common enough argument. But it also noted that any benefit to the mattress industry had a much broader reach: firms in other industries, producing textiles, foams, and innersprings, would also benefit from anti-dumping duties for mattress manufacturers, because mattress manufacturing draws on an extensive, predominantly domestic, supply chain.Footnote 1
Similar examples can be found in other contexts. In 2017, German auto manufacturer BMW pushed for changes in trade policy, pointing out that even suppliers that don’t themselves export benefit indirectly from new export opportunities for BMW—asking for, and defending, policies in line with BMW’s interests.Footnote 2 In 2013, US Senator Sherrod Brown noted that investments in Ohio’s auto industry indirectly created benefits for firms across Ohio throughout the supply chain, and called for favorable policies for auto manufacturers.Footnote 3
Such claims don’t sit easily with existing frameworks for understanding trade politics or the political influence of firms. Economic ties among firms remain largely outside their purview. The literature on trade politics, in particular, emphasizes cleavages across classes,Footnote 4 industries,Footnote 5 firms,Footnote 6 or occupations,Footnote 7 not ties among firms across industry boundaries. Indeed, much of the current literature focuses on individual firms, ignoring the economic and institutional context in which firms operate. Similarly, firm-specific attributes, not economic ties among firms, feature prominently in the literature on corporate power and the political influence of firms.
Yet, firms don’t operate in isolation. As these examples illustrate, they are embedded in production networks, composed of customer–supplier relationships, within the domestic economy; and firms and policymakers invoke these networks to justify particularistic policies. Even where previous work considers networks among firms, it does not account for these examples. A related body of literature observes that networks among firms—including customer–supplier relationships—predict similarities in political behavior,Footnote 8 rather than considering how firms use these networks to justify and receive privileged treatment. Work on multinational investment comes closest, showing that in weak institutional environments, foreign firms with ties to domestic firms are less concerned about expropriation.Footnote 9 For the most part, however, prior work has emphasized other types of networks—for example, networks created by global value chains,Footnote 10 interlocks among corporate boards,Footnote 11 shared membership in business associations,Footnote 12 social ties among firm owners,Footnote 13 or shareholder chains.Footnote 14
Mirroring this emphasis on networks among firms as an important aspect of politics, we contend that the examples we mentioned are part of a broader pattern. We argue, and demonstrate empirically, that differences in their embeddedness in production networks translate into differences in the political influence of firms. Connections to other firms extend a firm’s economic and political footprint. This enables firms to assemble a broader coalition, and to portray privileged treatment as more than a purely particularistic policy. A firm’s position within the domestic economy emerges as a source of political influence: firms are more likely to gain privileged treatment from their government if other firms also benefit indirectly.
We situate this argument in the context of trade politics. Which firms drive trade policy shapes a country’s integration into global markets and the ability to sustain that integration, as well as the distribution of power and wealth within societies. Understanding which firms succeed in gaining protection has long been a central question in international political economy, and it has been a prominent application to understand the political influence of firms since at least Schattschneider.Footnote 15 The resurgence of protectionist trade policy, and industrial policy more broadly, in recent years underscores the sustained relevance of this question.Footnote 16
Specifically, we examine decisions on the imposition of US anti-dumping duties as a manifestation of firm influence in politics. Anti-dumping duties are imposed temporarily to protect domestic firms from foreign competition. They have attractive features for our purposes. They substantially exceed pre-existing tariff rates, underscoring their relevance to firms.Footnote 17 And they are imposed in response to specific requests by firms, requiring an unambiguous decision by the government and allowing us to identify firms that voiced protectionist demands. While implemented through an administrative process and ruled on according to their legal and economic merits, decisions on anti-dumping petitions remain subject to political contestation and influence.Footnote 18 Differences in firms’ political influence should therefore be reflected in the success rates of anti-dumping petitions.
Combining data on US anti-dumping petitions with original measures of customer–supplier relationships, we show that a firm’s position within the domestic production network shapes the success rates of its anti-dumping petitions. Firms are more likely to obtain privileged treatment from their government, and have their anti-dumping petitions approved, for products tied into domestic production networks. This effect exists in addition to, and is distinct from, standard predictors identified in the literature, such as firm and industry size, geographic location in swing states, asset mobility, and industry-specific attributes.
The upside of our research design is that we can evaluate government responses to explicit requests by firms. This upside comes with an inherent trade-off: firms must have made such requests in the first place. While we cannot eliminate the resulting selection problem, we show that such selection would have to be implausibly strong to account for the relationship we report. Several additional pieces of evidence suggest a causal relationship. First, we exploit that trade flows have shifted throughout our sample period, in part due to exogenous changes in trade policy uncertainty, resulting in differential increases in imports across inputs. As the domestic footprint of an industry declines, so does its political clout.Footnote 19 Second, tracing sourcing decisions to exchange-rate movements at the level of an industry’s suppliers, we corroborate our results with instrumental variable estimates. Third, numerous firms have submitted anti-dumping petitions for different products. We present evidence that the success rate of these petitions increases for products that draw on a larger domestic production network, even across products from the same firm—which, together with industry-level attributes, captures most of the explanations given in the literature.
To elaborate on the role of politics, we offer evidence from written briefs submitted by members of Congress during anti-dumping investigations. Despite being short, often barely a page long, these briefs repeatedly invoke the indirect benefits to suppliers throughout the economy. Analyzing hundreds of these briefs, we demonstrate that firms drawing on a larger domestic production network receive support from more members of Congress, consistent with our argument that these are benefiting broader constituencies.
We also provide evidence for one specific mechanism. Building on recent work highlighting the role of swing states in anti-dumping petitions,Footnote 20 we show that firms drawing on a larger domestic production network are more likely to be of indirect relevance to swing states. This indirect exposure of swing states, which potentially increases political pressure on decision makers in anti-dumping petitions, explains part of the associations we report. This result establishes that the effects of a policy can build a politically relevant constituency through a firm’s production network.
Our account identifies a distinct source of corporate power: firms embedded in domestic production networks enjoy outsized political influence. While we focus on anti-dumping petitions as one manifestation of such influence, the argument readily extends to other issue areas. It also provides a different rationale for why large firms and industries might be particularly influential in politics.
We highlight two further contributions. First, we offer an initial step toward incorporating domestic production networks into explanations of trade politics. Our account speaks to several prominent themes in this literature: the role of individual firms, including multinationals, in driving trade and trade policy;Footnote 21 the expansion of global production networks and the resulting political backlash;Footnote 22 and the political consequences of the geographic concentration of industries.Footnote 23 The nuances of domestic production networks have received scant attention in these strands of literature. Yet production networks tie firms together across industries and connect smaller firms to global markets through their customers and suppliers. They are the counterpart to global production networks, which upend domestic production networks and erode an important political constituency. And they extend the geographic footprint of industries indirectly, such that the effects of policies are not limited to the directly targeted industries. We thus offer a starting point for developing a different perspective on trade politics: taking into account firm-to-firm linkages promises new implications for understanding political coalitions, the drivers of trade policy, the trade-offs of multinational production, and—as we outline in the conclusion—the role of geography and institutions in politics.
Second, a related body of literature, drawing on contributions in sociology and organization theory, examines the sources of corporate unity in politics.Footnote 24 This work documents that network ties among firms are reflected in similarities in political behavior.Footnote 25 Like Dreiling and Darves, who highlight how board interlocks and membership in business associations predict joint support for trade liberalization among firms,Footnote 26 we note the fruitfulness of considering economic ties among firms in explanations of trade politics. Shifting the focus to differences in the political influence of firms, we contribute in several ways to this literature. We show that networks among firms confer a form of (structural) political power that is unevenly distributed across firms within these networks; we emphasize a dimension of networks—a firm’s relevance to upstream suppliers—not considered in this literature; and we provide novel empirical measures of embeddedness in production networks, relating them to the success of firms in securing protectionist trade policy.
Firms, Production Networks, and Anti-dumping Petitions
Trade policy has long been used to understand questions of firm influence.Footnote 27 We focus on a specific type of trade policy: anti-dumping duties.Footnote 28 Dumping occurs when foreign firms sell products below cost or below the price in their home market. Anti-dumping duties are short-term, targeted instruments. For firms, they present an opportunity to obtain protection from foreign competition outside the standard policymaking process.
The process for the filing and investigation of petitions varies by country. In the US, anti-dumping duties are imposed in response to petitions by firms (and other petitioners, such as labor unions). Petitioners need to define the imported product, identify a comparable product in US classifications, document that the imported product affects a sufficiently large share of the US industry, and establish that the imported product causes or threatens to cause material injury to the US industry. These petitions reflect protectionist demands. They require an unambiguous decision by the government: whether to impose anti-dumping duties. Unlike in many other cases of policymaking, it is therefore possible to observe both whether a firm made policy demands and how the government responded.
Petitions involve a fair amount of documentation. Typically, they are prepared with the support of law firms. At the same time, as one of these law firms highlights, “the standard for initiation is low.”Footnote 29 Petitions only have to include the information required in petitions “to the extent reasonably available to the petitioner.”Footnote 30 Thus some firms view anti-dumping petitions as part of a broader toolkit of business strategyFootnote 31 —so much so that government agencies voice concerns about firms requesting anti-dumping duties without much merit.Footnote 32 Still, anti-dumping petitions are not a random sample of protectionist demands. Plausibly, they are based on substantiated concerns, and firms incorporate the odds of a successful outcome in the decision of whether to file a petition. In the empirical section, we address the resulting selection problem.
Petitions are filed simultaneously with the Department of Commerce and the International Trade Commission (ITC). Broadly speaking, Commerce determines whether a foreign firm likely sells its products below fair market value; the ITC determines whether the imported product is causing or threatening to cause material injury to domestic firms. Both determinations need to be affirmative for the imposition of anti-dumping duties.
Investigations are split into a preliminary phase (with a lower evidentiary burden) and a final investigation. The ITC, headed by six commissioners, reviews documents, hears witnesses, and requests written information from petitioners, other domestic firms, and the firms allegedly dumping. The investigation by the ITC ends in a vote by the commissioners; tied votes are considered affirmative. If the ITC commissioners vote affirmatively and Commerce determines that products are likely being sold below fair market value, Commerce issues an anti-dumping order, imposing duties on the products under investigation.
Many of the decisions in this rule-based process allow discretion. For example, Commerce may declare a “particular market situation” in the exporting country, arguing that market distortions justify an adjustment to the prices of imported products or production costs. US courts have repeatedly overturned decisions invoking particular market situations, suggesting that Commerce took advantage of this discretion to approve anti-dumping duties that otherwise would have been declined.Footnote 33 In response, in 2024, Commerce introduced rules expanding its ability to invoke particular market situations.Footnote 34
Investigations require determining a “like product”—a product essentially the same as the imported product—in US classifications. Commerce retains discretion over how this determination is made. This decision is not a mere technicality: it shapes which data are collected to assess fair market values and which industries are considered to assess injury to US firms. The time period for benchmarking also offers discretion. While domestic production is usually assessed over a twelve-month period, investigators can decide which twelve-month period to consider, may consider various time periods, and may use either value or volume.Footnote 35
In assessing whether imported products threaten to cause injury to domestic firms, the ITC can exclude some firms from consideration, molding to some extent the group of firms it considers in this assessment. The ITC can also limit its assessment to specific geographic regions where affected firms are concentrated. Engaging in, as a court ruling put it, such “arbitrary or free handed sculpting of regional markets”Footnote 36 has allowed the ITC to find material injury where otherwise a case would have been declined.
The interpretation of the assembled evidence is, likewise, not clear-cut. This is reflected in voting outcomes on petitions. The commissioners’ vote at the ITC is unanimous in about 44 percent of decisions. In about 25 percent, at least two commissioners dissented.Footnote 37
The investigation process allows for discretion and differences in interpretation. It also allows politics to enter. Both institutions involved in these investigations have close ties to the president and the Senate. The Department of Commerce is headed by the secretary of commerce. The secretary (with the assistant secretary for enforcement and compliance as their delegate) is ultimately responsible for the decisions of the department. Political considerations should be expected to factor in here: the secretary serves at the pleasure of the president and requires Senate confirmation.
The role of politics is potentially more questionable in decisions by ITC commissioners: the ITC is nominally an independent and bipartisan agency. Yet its six commissioners are nominated by the president and confirmed by the Senate. Based on observable indicators, the ITC is about as independent as the Securities and Exchange Commission, or SEC.Footnote 38 As a high-profile agency, the SEC typically is considered subject to political influence: firms with ties to senators are less likely to face enforcement actions by the SEC.Footnote 39
The broader literature on agency independence suggests specific mechanisms through which politics matters.Footnote 40 First, Congress ultimately controls agency resources. The ITC is a small agency, with a budget of $122 million in 2022 and a staff of about 400. Like other agencies, it faces resource constraints. In its 2025 budget justification, the ITC noted the need for more funding and the pains of being short-staffed, in particular in its offices related to anti-dumping investigations.Footnote 41 The six commissioners, one of whom serves as chair, rely on this budget for their daily work. Sustaining the support of the president and key members of Congress therefore remains important. Decisions that stray from their interests may undermine that support.
Second, the threat of (uncomfortable) congressional hearings, which potentially harm future career prospects,Footnote 42 constrains the ITC commissioners. And implicitly, as in the context of central banks, the independence of agencies always remains conditional: it might get revoked if an agency fails to deliver on key goals of congressional majorities and the president.Footnote 43 This presents a significant concern for the ITC, which in official communications prides itself on its independence.
Third, commissioners have their own views on what US trade policy should look like. And the nomination process ensures that the views of commissioners align (at least partially) with those of the president and the Senate. Moreover, the role of chair rotates every two years. Because the president designates the chair, voting against the interests of the president has career repercussions. The chair enjoys not just visibility but also influence. For example, in 2022, Jason Kearns, at the time chair of the ITC and a proponent of more protectionist trade policy, increased the influence of commissioners over methodologies and data sources with a new directive. The goal was to move policy advice given to Congress and the president in a more protectionist direction, with a focus on domestic manufacturing and supply chain resilience. In 2024, chair David S. Johanson rescinded the directive.Footnote 44
Previous studies offer evidence that politics shapes decisions on anti-dumping petitions. Members of Congress frequently intervene by providing testimony in ITC hearings. Their involvement appears to matter for investigation outcomes.Footnote 45 Consistent with the appointment-and-confirmation process, the interests of the president and the Senate in particular are reflected in the outcomes of these investigations.Footnote 46 Finally, attributes that render firms politically influential, such as asset specificityFootnote 47 and industry size,Footnote 48 shape the success rates of petitions, indicating political considerations.
Production Networks and Political Influence
We highlight a different attribute that renders firms politically influential. We contend that firms with linkages throughout the domestic economy enjoy privileged treatment, because benefits to such firms create indirect benefits to other constituents. In the context of anti-dumping petitions, we expect that firms are more likely to see their petitions approved if protectionist trade policy creates indirect benefits for domestic suppliers.
Figure 1 illustrates the core of our argument in two stylized scenarios. In the left panel, production by firm A absorbs a small share of the output of other firms (1, 2, and 3). Higher tariffs on firm A’s products help it remain in business. But these benefits are largely confined to firm A. Few other firms, and as a consequence few other constituencies, voters, or policymakers, benefit from the higher tariffs.
Illustration of a firm of low importance to suppliers (left) and of high importance to suppliers (right)
Note: Square boxes (firms 1, 2, and 3) are suppliers; the amount of filling shows what share of each supplier’s output is absorbed by firm A. In both panels, firm A is downstream from the other firms, but its relevance to the other firms differs in the two situations.
In the right panel, firm A draws on a large share of the other firms’ output. If firm A goes out of business, they will lose an important customer. Thus protectionist policy for firm A indirectly benefits these other firms: it allows firm A to keep sourcing from them and alleviates cost pressures firm A might otherwise pass on.
We note several related aspects of this illustration. First, it is about domestic suppliers, not foreign suppliers. Our account thus presents a counterpoint to the literature on global production networks. Second, we expect that firm A is politically more powerful if it is connected to multiple suppliers upstream because this creates indirect benefits for more constituents. Third, because we are interested in whether firm A is important to upstream suppliers, we emphasize the share of upstream suppliers’ output rather than whether the suppliers are large themselves, or A’s total domestic sourcing. Finally, in our account, firm A is not influential because it is large.Footnote 49 Larger firms can absorb larger shares of upstream suppliers’ output, but we provide a theoretically and empirically different explanation for why they are politically powerful.
In short, we argue that the extent of the filled-in area reflects firm influence. As we will detail in the following, we contend that firm A has more influence in politics if it draws on a larger share of output across suppliers, because this (1) results in a broader and more diverse political coalition, (2) increases the geographic footprint of the firm, and (3) allows firms and allied policymakers to counter a narrative of particularistic policymaking.Footnote 50
Coalition size and breadth. The larger the indirect benefits are to suppliers, the broader becomes the overall constituency. For example, the major automakers in Tennessee employ about 12,000 people directly; suppliers to these automakers employ another 82,000.Footnote 51 This suggests strength in numbers. In hearings and written briefs, the effects on both employees and firms feature prominently. As expected from collective-action arguments, the effects on employees tend to be referenced by others (such as firms and policy-makers), whereas firms regularly provide their own submissions. We cannot ascertain how key policymakers in the legislature and executive—and, as a result, decision makers at the ITC and the Department of Commerce—weigh these different constituents, though prior work suggests that both factor into the calculus of policymakers to some extent.Footnote 52
The indirect effects not only broaden the constituency but also diversify it across industries and occupations. Occupations, in turn, differ on multiple dimensions, including the education, income, gender, and race of employees. We illustrate this in Figure 2 for the mattress manufacturing industry mentioned in the introduction. Columns along the horizontal axis correspond to distinct occupations. In the top row, we depict the composition of the mattress manufacturing industry across occupations, using the industry-occupation matrices from the Bureau of Labor Statistics. Darker colors indicate that the respective occupation accounts for a larger share of industry employment. We also depict the composition of four major industries that are suppliers to the mattress manufacturing industry: urethane foams, springs and wires, fabric mills, and millwork.
Occupational makeup of mattress manufacturing and its major supplying industries
Note: Occupations on the horizontal axis. Darker colors indicate that an occupation accounts for a larger share of industry employment.
The occupational composition differs markedly across these industries: these supplying industries employ many occupations that are barely relevant, or not relevant at all, to the mattress manufacturing industry. Protectionist trade policy supporting mattress manufacturing therefore affects a broad range of occupations only indirectly, and still others mostly indirectly. These occupations differ on several observable dimensions, including wages and education, as well as gender and racial composition. The indirect constituency behind mattress manufacturing is, therefore, tapping into various socioeconomic groups. Each of these, in turn, might heighten the political relevance to policymakers beyond what the mattress manufacturing industry alone might achieve.
These attributes matter not only for which citizens are affected. The occupational makeup matters for how industries are perceived in politics, and consequently, for the political effectiveness of these industries. Industries themselves also differ in politically relevant attributes. We highlight two: unionization rate and firm size. Unions might become important political allies in anti-dumping petitions. Unionization rates in the US are fairly low, but vary considerably across industries: they are twice as high in fabric milling than in mattress manufacturing,Footnote 53 even though fabric mills are, in terms of the occupational profile, the closest major supplier to mattress manufacturing. A firm with a larger domestic production network is more likely to tap into suppliers that are unionized, thus mobilizing support from unions and from policymakers who support unions. Firm size, likewise, varies across industries and in particular across customers and suppliers: on average, suppliers are smaller and younger than their customers.Footnote 54 This allows large firms to point out that policy benefits accrue to smaller firms as well, pre-empting pushback against particularistic policy that solely benefits large firms. Where firms source from large firms, it extends the indirect constituency to potentially powerful political interests.
Similarly, production networks extend the indirect effects across industries in different sectors of the economy. In mattress manufacturing, for example, even the major suppliers are from very different parts of the industry spectrum, as reflected in their NAICS codes. More minor suppliers extend well outside manufacturing. The indirect constituency therefore adds diversity—which, again, might be valuable in itself, but also might rope in constituents with political influence.
Suppliers may get actively involved on behalf of a firm.Footnote 55 This special-interest channel is plausibly an important component of how domestic production networks become politically relevant. But such active involvement is not necessary. Policy-makers allied with a petitioning firm may be able to mobilize other policymakers whose constituents would benefit indirectly. Similarly, citizens and employees are frequently referenced in investigation hearings and political commentary, but they do not have to recognize this link explicitly.Footnote 56 As we will discuss, this broader constituency is an important asset in political contests. The direct beneficiaries can invoke these indirect benefits to highlight the broader appeal, across a more diverse set of constituents, of policies that benefit them. Indeed, in US politics, more diverse coalitions are politically more successful.Footnote 57
Economic geography. Taking the indirect effects into account extends the diversity of a coalition on another important dimension: geography.Footnote 58 To return to the example of automakers in Tennessee, the major automakers are located in just five counties, but suppliers to these automakers are located in eighty out of ninety-five counties,Footnote 59 as well as in other states.
Figure 3 demonstrates the extent of these indirect effects for anti-dumping petitions filed in 2015. We consider a county as directly involved if at least 250 employees work there in an industry associated with an anti-dumping petition. We consider a county as indirectly involved if at least 250 employees work there in industries that supply industries associated with an anti-dumping petition (we weight industry employment by the share of industry output supplied to an industry).
US counties where at least 250 employees are affected directly (light green), directly and indirectly (dark green), or only indirectly (blue) by anti-dumping petitions filed in 2015
Notes: Authors’ calculations, combining data on anti-dumping disputes from Bown Reference Bown2011, product-industry concordance from Schott Reference Schott2008, data on employment by county from the County Business Patterns (Eckert et al. Reference Eckert, Fort, Schott and Yang2020), and data on input linkages from the 2012 Bureau of Economic Analysis Input-Output Accounts (see the next section for details).
Many counties are affected both directly and indirectly, depicted in dark green. This helps legislators build broader and more diverse coalitions within their districts, consistent with points we mentioned earlier. Additionally, many counties, depicted in blue, are affected only indirectly, extending the geographic footprint of the constituency behind anti-dumping petitions.
In the US, politically influential industries tend to be geographically concentrated, which facilitates collective action, but politically dispersed, which facilitates coalition building.Footnote 60 Production networks achieve this balance in a striking way: while the industries themselves might be concentrated, the effects on other industries are politically dispersed.
This effect is particularly relevant in the context of anti-dumping petitions. Representation by influential members of Congress tends to increase the success rate of anti-dumping petitions.Footnote 61 Similarly, industries located in states relevant to senators on the Finance Committee and industries relevant to swing states tend to have higher success rates in their anti-dumping petitions.Footnote 62 For firms tied into production networks, it is more likely that an anti-dumping petition gains such political relevance indirectly.
Narrative. Firms can justify particularistic demands with language that emphasizes broader, indirect benefits. Policy that benefits other firms indirectly can be depicted as being more than just special-interest politics: there is less special-interest politics involved where tariffs benefit firms and workers across industries and across space. Of course, tariffs still have overall costs for society, including downstream consumers. In the political calculus over whether to impose anti-dumping duties, then, the benefits of protectionist policy for other firms are an important counterweight to such opposition, both politically and economically.
Emphasizing the benefits to other domestic constituents allows firms and policy-makers to frame the debate more broadly in the context of fairness. This is consistent with the general perception of anti-dumping duties as an instrument for restoring fair trade. Spinning a narrative about providing fair treatment to domestic firms in response to unfair competition by foreign firms is already politically appealing.Footnote 63 The indirect benefits for suppliers complement this narrative with a narrative that highlights broadly shared, as opposed to concentrated, benefits across domestic firms—including, as noted earlier, firms of different sizes and in different industries.
This discussion emphasizes domestic suppliers. Foreign suppliers add little to the extent or diversity of a firm’s domestic constituency. Sourcing from foreign affiliates of US firms upstream might be a middle ground in this distinction—upstream firms still benefit from tariffs protecting their customers, but the link to the domestic economy becomes more tenuous. Indeed, the previous arguments suggest it would be difficult for an upstream supplier that offshored production to become an effective ally in anti-dumping petitions, which are about unfair foreign competition threatening US-produced goods. More broadly, however, our emphasis on domestic suppliers suggests systematic differences across firms, depending on how tied they are into global production networks; sourcing inputs from abroad upends domestic production networks. A related point emerges from the literature on the globalization backlash, which notes the increasing pushback against multinationals, in particular in response to offshoring. We suggest a different mechanism: by offshoring suppliers, multinationals are eroding their domestic constituency, which reduces their political clout.
How do commissioners learn about these indirect benefits to suppliers? The investigation process provides several opportunities. Through hearings and submissions during the investigation process, firms can relay information directly to the Department of Commerce and the ITC commissioners. They can also relay information to policymakers, who can include it in briefs and public statements, and possibly in closed-door meetings. Relaying such information has at least some credibility and is not available to every firm. In preliminary investigation hearings, speakers are reminded that false or misleading statements carry potential penalties. And in final investigation hearings, all parties testify under oath, opening witnesses up to perjury charges for false or misleading statements.
Anecdotal evidence suggests that firms and policymakers indeed convey such information. For example, in a 2021 letter to the ITC, Brian Higgins, a member of Congress, wrote that one of the petitioning firms was not only located in his district, providing “highly skilled, good paying union jobs”—but also that the company “is an important customer of our nation’s steel industry as it uses flat rolled American steel as an input product.”Footnote 64
Nor do firms hesitate to mention that their suppliers depend on them. In 2001, in testimony before the ITC, a representative for LaBarge Pipe and Steel first pointed out that his company maintained a domestic supply chain, despite the cost advantages of moving to overseas suppliers, and then noted that “unless the United States imposes dumping duties to restore fairness to the marketplace for large diameter line pipe, in order for us to stay in business we will have to abandon our traditional relationship with domestic suppliers. In fact, if this dumping were allowed to continue we think one or two of our domestic suppliers would have gone out of business.”Footnote 65
Similarly, in a 2017 petition on paper products, with the North Pacific Paper Company (NORPAC) as one of the petitioners, the CEO of the company and its vice president of manufacturing both mentioned the indirect benefits to suppliers in their testimony. The CEO stated that his company “employs about 360 people directly and many more indirectly who provide a host of goods and services.”Footnote 66 And the vice president added, “You’re dealing with people and jobs. And not just the jobs of the people that are working in the mill, but the suppliers to the mill, anybody that’s affiliated with an operation. There’s just a lot of customers and employees and suppliers that are impacted on that.”Footnote 67 Some of the customers—mostly from the newspaper industry—voiced opposition to anti-dumping duties, but no similar concerns were raised by suppliers, who stood to benefit from NORPAC gaining protection.
These anecdotes are also notable for another reason: the definition of material injury in the investigation process is limited to the industry producing the product in question. Downstream customers and upstream suppliers are explicitly not considered. The US government has long resisted the formal incorporation of these or other “public interest” considerations.Footnote 68 The discussion of indirect effects therefore cannot be explained with a legal or economic rationale.
Yet policymakers, firms, and the legal representatives of firms repeatedly bring them up, despite a considerable opportunity cost. Hearings assign strict time limits to statements by interested parties. Arguments about indirect benefits crowd out other arguments. Similarly, in the example from the introduction, the letter on behalf of mattress manufacturers is barely a page long. The reference to US suppliers who stand to benefit takes up valuable space, at the expense of other arguments that could have been put forward; clearly, this reference was perceived to hold at least some sway in Congress’s decision on whether to impose anti-dumping duties.
We conclude with our main hypothesis:
H1: Firms are more likely to have anti-dumping petitions approved for goods whose production absorbs a larger share of suppliers’ output.
Empirical Evidence
We compile and combine two data sets. The first contains data on petitions for anti-dumping duties filed by US firms.Footnote 69 We create product-, industry-, and firm-level identifiers to link this data set with other data sources. The second data set contains several variants of our main predictor. It delivers original measures of linkages between firms at the industry level, which we derive from US input-output tables and link to anti-dumping petitions.
Anti-Dumping Petitions
We compile data on US anti-dumping petitions and their decisions between 1992 and 2019 from the Global Anti-Dumping Database.Footnote 70 It provides information on which domestic firms acted as petitioners, identified by firm name; which products were concerned, identified by up to ten-digit HTS codes; and the date and outcome of the ruling. We maintain separate entries for each country listed in a petition because rulings may vary across countries, and refer to this as a petition in the following. For each petition, we have multiple observations if several firms acted as petitioner or several products were included in the petition. For example, investigation 731-TA-1271 relates to three petitioners, DAK Americas LLC, M&G Chemicals, and Nan Ya Plastics, and involves two products, for a total of six observations.
Our outcome variable, whether a petition is successful, is binary. If the investigation ends with an affirmative or partially affirmative ruling, we consider it a successful petition, coded 1. Petitions with a negative ruling, either in the preliminary phase or in the final phase, are coded 0. We omit withdrawn and terminated cases because we cannot ascertain the outcome of any private bargaining taking place. This affects about 5 percent of all petitions. As we report in the online appendix, the estimates remain statistically significant regardless of how we allocate successes and failures across unobserved outcomes.
Our data set comprises 918 petitions, which cover about 2,000 distinct products and 700 petitioners; 58 percent of the petitions are successful. Figure 4 displays the percentage of successful petitions per year, along with the number of petitions filed. We observe an upward trend in success rates over time. Though petitions are not more frequent in presidential election years (darker color), they are more likely to be successful then.
Success rate (bars) and number (line) of US anti-dumping petitions, 1992–2019
Notes: Success rate is the share of successful petitions among all petitions. Presidential election years are highlighted with a darker color. Authors’ calculation, based on Bown Reference Bown2011.
We link to other data sets in two ways. First, we match each HTS product code to the corresponding NAICS industry code, using the concordance from Schott.Footnote 71 We thereby associate each product listed in an anti-dumping petition with a six-digit industry code. If multiple product codes match onto the same industry, we drop these duplicates.
This retains a high level of detail. For example, within the Manufacturing sector, we have the three-digit Electrical Equipment, Appliance, and Component Manufacturing subsector. Within this subsector, we can distinguish between six-digit codes representing Household Cooking Appliance Manufacturing, Household Refrigerator and Home Freezer Manufacturing, and Household Laundry Equipment Manufacturing. In some of our empirical specifications, we take advantage of differences within subsectors and control for differences across them through fixed effects.
Second, we match the petitioner data with a firm-level data set, Orbis, from Bureau van Dijk. We start by standardizing firm names. For example, we remove “co,” “corp,” “llc,” and special characters. We then match the data with Orbis, first with exact matches, then with fuzzy matches based on phonetic similarity and Jaro-Winkler similarity (a measure of string distance). Finally, we inspect the data manually to resolve inconsistencies and to obtain unique Orbis identifiers. This allows matching the data with other variables, including the NAICS industry code derived from the firm (rather than the product). We rely on the firm identifier and the firm-derived industry code to merge in control variables for firm- and industry-level attributes.
Embeddedness in Production Networks
Our second data set collects three measures capturing to what extent an industry is important to suppliers upstream. The three measures are based on the US Input-Output Accounts published by the Bureau of Economic Analysis. We draw on the Use Tables, which indicate the purchases of different commodities by each US industry. Detailed input-output tables are produced in benchmark years, ending with 2 and 7. We use the input-output tables from 1997 through 2017. The 1997 tables are the first to be based on NAICS, while the 2017 tables are the last to be within the time span of our data set on anti-dumping petitions. Short of firm-level data, which exist for only a few countries and are not publicly available,Footnote 72 these are the most detailed data available for capturing relationships between firms.Footnote 73 Overall, from the five vintages of the Use Tables, we obtain information on over 600,000 customer–supplier relationships between industries.
By using input-output tables from several years, our measure captures cross-industry differences as well as changes in input-output relationships over time. The latter are at least in part driven by exogenous changes in production technologies. For example, components needed by the automotive industry have changed with the rise of electric vehicles and electronic systems. The drawback of using measures from several years is that some of the differences we capture are not based on substantive changes in production technology but on accounting changes in industry classifications: consolidations and disaggregations. In the online appendix, we show that our results are robust to using data from a single year across the entire sample.
Inputs. According to the discussion in the previous section, our first measure captures to what extent firms upstream rely on an industry downstream as a customer, such that benefits to firms in industry
$i$
also benefit other firms that supply industry
$i$
. Denoting with
${m_{ij}}$
the value of inputs
$j$
of industry
$i$
, and with
${q_j}$
the total output of
$j$
, and (for simplicity) ignoring time subscripts in all of the following, we define
$${\sigma _{i,j}} = {{{m_{i,j}}} \over {{q_j}}}.$$
where
${\sigma _{i,j}}$
is the share of
$j$
’s total production that is supplied to industry
$i$
. Alternatively, we can conceive of
${\sigma _{i,j}}$
as industry
$i$
’s absorption of
$j$
’s output. The larger
${\sigma _j}$
is, the more indirect benefits
$j$
receives from policies that allow
$i$
to maintain or increase production.
Using this definition, we calculate the sum of industry
$i$
’s input shares across
$j$
:
$${\eta _i} = \mathop \sum \nolimits_{j = 1}^N {\sigma _{i,j}}.$$
We emphasize three attributes of our approach. First, the Use Tables capture purchases by domestic firms. Purchases by foreign firms are not included, which matches our theory. However, our measure does capture purchases from foreign firms that are sourced through imports. This is justified to the extent that total demand for a product, regardless of whether this demand is satisfied from domestic or foreign suppliers, sustains a higher price for that product. At the same time, if an industry largely sources products from abroad, it can hardly claim to provide benefits to upstream domestic suppliers. Later on, we provide a variant that strips out imports.
Second, the size of an industry enters our measure indirectly. Larger industries can absorb larger shares of other industries’ total output, but industry size is not by construction related to the number of supplying industries or the distribution of inputs across supplying industries.
Third, the size of supplying industries enters indirectly as well. Industries that source a large share of smaller industries’ output have, everything else equal, larger values of
$\eta $
than industries that source the same total amount of intermediates from a single industry with the same total output. This is consistent with our theory: everything else equal, our measure is larger for industries that are important to many smaller suppliers than for industries that are important to a small number of large suppliers, even if total purchases and the joint output of upstream suppliers are identical.
Total domestic inputs. The construction of
$\eta $
is based on immediate suppliers. We did not model these suppliers also having suppliers, extending the reach of an industry throughout the economy. On the one hand, this is an attractive feature. That other firms benefit indirectly becomes less and less plausible as these other firms become further removed in the production process. On the other hand, these more indirect relationships remain important, at least to some extent, because they capture the total reach of an industry.
To incorporate these higher-order relationships, we turn to the Total Requirements Tables. These provide coefficients
${c_{i,j}}$
, which capture how many dollars of inputs industry
$i$
needs from industry
$j$
to produce one dollar of output. This coefficient captures all input relations for industry
$i$
, including inputs for inputs, and so on. To construct a measure analogously to our core measure
$\eta $
, we obtain the total output of each industry,
${q_j}$
, and match it with the Total Requirements Table, to calculate
$$\eta _i^{\rm T} = \mathop \sum \nolimits_{j = 1}^N {c_{i,j}}{{{q_i}} \over {{q_j}}}.$$
Imports. The Use Tables are based on purchases by domestic firms. A portion of these purchases stems from foreign suppliers. The Use Tables do not distinguish between domestic and foreign sourcing, which leads us to overestimate the importance of industries to domestic suppliers.Footnote 74
This presents two potential problems for our analyses. First, because industries differ in their reliance on tradable inputs,
$\eta $
will vary in the resulting measurement bias across industries. Second, because the importance of imported inputs changes over time,
$\eta $
will vary in the resulting measurement bias across time. The entry into force of NAFTA in the second half of the 1990s and the extension of permanent normal trade relations to China in the early 2000s were associated with dramatic increases in imported inputs by US firms.Footnote
75
Indeed, Quinn and Liu show that the sudden influx of imports from China, often called the “China shock,” was more akin to a “multinational corporation shock,” driven by increased sourcing by US multinationals.Footnote
76
At the same time, these changes in trade patterns offer opportunities for our research design because they imply changes in the reliance on domestic suppliers that are driven by changes in trade policy—and some of these changes, such as the binding of tariff rates, are plausibly exogenous to domestic political factors at the level of individual products: the change in trade flows was not a consequence of differential cuts in tariff rates across specific products (which would plausibly be related to the political power of US firms) but a reduction of uncertainty in US trade policy as a consequence of reducing the gap between applied and bound tariffs.Footnote 77
To account for imports of inputs, we combine data from the input-output tables with product-level data on US imports. To hold constant changes in production technology and industry classifications, we use data from the 2012 version of the input-output tables. This ensures that the driver of differences over time in our measure is changes in trade flows. We then concord these data to BEA input-output codes and match them with the 2012 Use Table to adjust for imported inputs; we provide a detailed discussion of this process, including the allocation of imports across industries, in the online appendix.
With
$m_{i,j}^{\rm{*}}$
denoting industry
$i$
’s purchases of imported inputs
$j$
, we analogously calculate
$$\sigma _{i,j}^* = {{{m_{i,j}} - m_{i,j}^*} \over {{q_j}}} = {\sigma _{i,j}} - {{m_{i,j}^*} \over {{q_j}}},$$
resulting in our third measure,
$$\eta _i^{\rm M} = \mathop \sum \nolimits_{j = 1}^N \sigma _{i,j}^*.$$
Descriptives. We use the concordances provided by the BEA to match these measures from the input-output format to NAICS six-digit industry codes; where BEA codes match multiple NAICS codes, we distribute these accordingly, using (where available) data on output shares by industry. We match the data to different NAICS vintages using concordances from the US Census Bureau. From there, we obtain for each product listed in an anti-dumping petition the three measures just described. We consolidate NAICS codes to the 2012 benchmark year throughout. The complete set of measures, across NAICS codes, is available from the authors.
Figure 5 displays the values of
$\eta $
,
${\eta ^{\rm{T}}}$
, and
${\eta ^{\rm{M}}}$
across NAICS industries for a single year, 2012, in the top, middle, and bottom panel, respectively. The figures include the same industry codes represented in our petition data. The measures vary considerably across industries and from each other. Figure 6 charts how
${\eta ^{\rm{M}}}$
varies over time for a subset of industries. Some industries remained relatively stable; others replaced a large and increasing share of inputs with foreign suppliers; and other industries experienced considerable swings over the sample period.
Distribution of input share (
$\eta $
), total input share (
${\eta ^{\rm{T}}}$
), and imported input share (
${\eta ^{\rm{M}}}$
) in 2012, across NAICS six-digit industries associated with anti-dumping petitions (calculated from US input-output tables)
Imported input share (
${\eta ^M}$
) over time for select industries (calculated from US input-output tables)
Results
Each observation in our final data set corresponds to a petition-firm-product: for each petition, we have an observation for every firm and every product (identified by the corresponding NAICS industry code) listed in the petition, with separate entries for each country listed in a petition. We maintain this structure for two reasons. First, it lets us incorporate control variables at the firm and industry level that feature prominently in the literature. We can thus differentiate our argument from established arguments about firm and industry size, mobility, and location. Second, it allows research designs with firm fixed effects, exploiting product-level differences, as well as industry-level instrumental variables.
Because our outcome variable, success in an anti-dumping petition, is binary, we estimate logit models with robust standard errors. We cluster standard errors on the petition, so as not to overcount the amount of information in our data set. We present results with alternative clustering structures, including clustering on the industry, in the online appendix. There we also report results from weighted regression models, and from models that collapse the data by the petition.
We include several control variables in our base model. First, given the trends in the outcome, we include year, year squared, year cubed, and a dummy variable for presidential election years. Second, larger industries might be politically more powerful and absorb larger shares of output from upstream industries. We include logged industry output, obtained from the BEA. Third, to adjust for differences in the legal merit of cases, we control for the percentage change in imports of the products mentioned in the petition. This represents the most plausible observable measure of whether a product was in fact dumped. Fourth, we include a control variable for steel products, which feature prominently among anti-dumping petitions in the public discourse and in political relevance. Fifth, we account for whether a firm made campaign contributions in the two years before or after filing a petition, a main alternative channel through which firms secure preferential policies;Footnote 78 the data are from the Database on Ideology, Money in Politics, and Elections.Footnote 79 Finally, we include a control variable for nonmarket economies, where the rules for establishing and determining the occurrence of dumping differ markedly. We obtain the list of nonmarket economies, as determined by the US Department of Commerce, from the Federal Register for every year in our data set.
In a variant of the base model, we include additional variables that capture firm- and industry-level attributes. First, multinational corporations play a prominent role in US trade politics,Footnote 80 including in anti-dumping petitions.Footnote 81 We therefore include a dummy variable to indicate whether a firm is a multinational corporation, or part of one. To identify such corporations, we draw on the ownership data from Orbis and code a firm as multinational if at any point either it or its ultimate domestic owner owns a subsidiary in a foreign country. We describe results with different variants of this measure, including some accounting for vertical integration, later on. Second, firms listed on a stock exchange gain political power through their widespread public ownership.Footnote 82 We include whether a firm is listed on a stock exchange, meaning either it or its ultimate domestic owner was (per Orbis) listed on a stock exchange at any time. Third, because more mobile firms are more likely to receive favorable treatment, we include the log-transformed current stock of fixed assets at the subsector level, from the BEA. Fourth, following prior work on protectionist demands, we include the real exchange rate, weighted by import partners, at the subsector level.Footnote 83 Together with changes in imports, this variable captures the most prominent economic determinants of the filing of anti-dumping petititions. Finally, we include fixed effects by NAICS three-digit code.
Table 1 presents results for the baseline model in odd columns and for the expanded model in even columns.Footnote
84
Results for
$\eta $
are reported in columns 1 and 2, for
${\eta ^{\rm{T}}}$
in columns 3 and 4, and for
${\eta ^{\rm{M}}}$
in columns 5 and 6. The relationship between our variants of
$\eta $
and petition success is positive and statistically significant in all models. The effect sizes are considerable. In the baseline model (column 1), moving from the twenty-fifth to the seventy-fifth percentile on
$\eta $
results approximately in a 30 percent increase, from 62 percent to 82 percent, in the probability that an anti-dumping petition is approved: firms are considerably more likely to have petitions approved if benefits to their products extend to suppliers upstream. We obtain similar substantive results in the other models. The difference in coefficient size across the three measures is primarily driven by the different scales of the variables.
Success of anti-dumping petitions: base models
Note: Logit models with robust standard errors, clustered by petition;
$p$
-values in parentheses.
These results suggest a distinct source of corporate political power based on a firm’s domestic production network. The results for
${\eta ^{\rm{M}}}$
, in columns 5 and 6, lend themselves to a noteworthy interpretation: as firms increase their global sourcing,
${\eta ^{\rm{M}}}$
declines, and this decline translates into a decline in political power. Firms thus face a trade-off between business and political goals. The globalization of supply chains undermines the domestic constituency of multinational corporations, resulting in a loss of political power.
Robustness: Control Variables and Model Specification
In the online appendix, we present results from several robustness checks.
Clustering. We clustered standard errors on the petition. Our variants of
$\eta $
are calculated at the level of NAICS industries. We therefore cluster standard errors instead on industries. We also consider two-way clustering on petitions and industries, and clustering on firms.
Weights. Within each petition, our data set includes varying numbers of observations as a function of the products listed in the petition, the extent to which the products map onto different NAICS codes, and the number of firms listed in the petition. This creates uneven numbers of observations across cases, such that our coefficient estimates give considerably more weight to cases with many observations, which might also deflate
$p$
-values. To offset this, we estimate weighted logistic regression models, with weights corresponding to the number of observations in each case.Footnote
85
Related-party trade. Our third measure,
${\eta ^{\rm{M}}}$
, accounts for imports of inputs, to isolate relevance to domestic suppliers. A substantial portion of US imports is related-party trade: imports sourced from US-owned firms abroad. In such cases, sourcing products from abroad still has benefits to US firms.
${\eta ^{\rm{M}}}$
stripped out these related-party imports, as part of overall imports. To account for related-party trade and retain it as part of our measure, we obtain data on related-party imports by industry from the Census Bureau. We then reconstruct our measure, adjusting only for imports sourced through arms-length transactions.
Case merits. If cases with higher values of
$\eta $
are cases with more observable evidence of dumping, success rates would be higher, but not for the reasons we identified. To adjust for observable evidence of dumping, we included changes in imports and exchange rate misalignments as control variables. Drawing on sensitivity analysis,Footnote
86
we report in the online appendix that an omitted variable would have to be implausibly strong relative to these observed indicators of dumping to invalidate the reported association.
To further probe the legal merit of cases, we turn to disputes filed with the World Trade Organization (WTO) Dispute Settlement Body.Footnote 87 Almost 20 percent of US decisions on anti-dumping petitions are challenged at the WTO.Footnote 88 Disputes at the WTO are relatively rare and typically not filed frivolously, at least not against wealthy countries such as the US. This indicates that other governments mostly challenge anti-dumping decisions they perceive as lacking legal merit. If differences in legal merit explain the association we reported, we would observe a negative correlation between our measures and a decision being challenged.
We draw on Schott and Jung,Footnote 89 as well as the one-page summaries of disputes provided by the WTO, to create two variables: whether the US decision was challenged; and whether this challenge was successful.Footnote 90 In the online appendix, we report that cases with higher values of our measures are not any less likely (in fact, more likely) to be challenged, suggesting that differences in the merits of cases are not explaining the pattern we reported.
Product characteristics. One justification for tariffs is the protection of industries further downstream. Some evidence suggests that governments are more sensitive to the protectionist interests of downstream industries producing final consumer goods.Footnote 91 To differentiate this explanation from ours, we include a measure of upstreamness.Footnote 92
Firm-level and industry-level characteristics. We control for additional firm- and industry-level features. These include firm and industry size, measured by the log number of employees; the capital–labor ratio;Footnote 93 the number of counties in which an industry has more than 250 employees;Footnote 94 value added as a share of GDP and an industry’s contribution to GDP growth, to distinguish our measure from industries that are of economic importance; and total imports of industry inputs, to isolate differences in the extent of domestic sourcing for any given amount of foreign sourcing.
Lobbying and campaign contributions. In addition to campaign contributions, we include lobbying data from LobbyViewFootnote 95 to control for whether a firm lobbied during the sample period.
Vertical integration. The input-output tables are based on establishments. Some of these might be within the boundaries of a firm, reflecting vertical integration, which promises efficiencies for firms when it comes to mobilizing political action among policymakers. To account for differences in vertical integration, we combine data from Orbis on ownership structures with data from the input-output tables. From the input-output tables, we identify all supplier industries at the level of NAICS industries. We then match each firm in the anti-dumping data set with its subsidiaries, derived from the Orbis data. For each firm, we calculate the number of subsidiaries in industries that are considered upstream in the production process, based on the input-output tables. A second dimension of vertical integration is whether a firm has a subsidiary in the target market. We identify from Orbis whether a firm (or a parent in the petitioner’s corporate family) had a subsidiary in the target market.
Exporting markets. Anti-dumping petitions are filed against firms from a wide range of countries. This suggests confounding if country characteristics correlate with our key measures. In the online appendix, we report that the results are robust to including exporting-market fixed effects.
One salient distinction is that between market and nonmarket economies.Footnote 96 Anti-dumping petitions directed against firms from nonmarket economies follow a distinct process: there is more leeway in establishing the case for dumping, because firms and regulators have more flexibility in identifying counterfactuals and market prices. As a consequence, anti-dumping petitions directed against nonmarket economies have higher success rates. Together with the more flexible process for establishing legal merits, this might reduce the role of politics in these decisions.
In the online appendix, we verify that the results are not contingent on including nonmarket economies in the sample. Also, consistent with this argument, the effects are strongest in cases against market economies. In cases against nonmarket economies, the effects of our key variables shrink in size and lose statistical significance. Moreover, several of the control variables change sign in the sample of nonmarket economies, suggesting distinct political dynamics.
Political representation. We consider several variables that account for differences in political representation. First, we control for the average partisan representation of an industry, calculated as the share of electoral districts with more than 250 employees in which an industry is represented in the US House of Representatives by a member of the Democratic Party. Second, previous work has shown that industry location in swing states confers political advantages.Footnote 97 We define swing states as those states that were won by a vote margin of 5 percent or less in the past presidential election. We then create industry- and firm-level measures. For industry-level measures, we create a variable representing whether an industry involved in an anti-dumping petition employs more than 250 people in a swing state;Footnote 98 and we calculate total employment in swing states by an industry involved in an anti-dumping petition and include the resulting (log-transformed) variable. For the firm-level measure, we identify whether a petitioner is located in a swing state using information from the Global Anti-Dumping Database, company directories (including Orbis and Dun and Bradstreet), and information from anti-dumping petitions and press releases.
Identification: Fixed Effects and Instrumental Variables
Firm fixed effects. Several firms in our sample submitted multiple anti-dumping petitions. This allows us to include firm fixed effects: we can hold constant firm-specific attributes and evaluate petition success rates across different products. To the extent these do not change within the time span of our panel, the firm fixed effects account for several factors potentially associated with a firm’s political influence, such as geographic location, company size and employment, integration into global markets, and political connections.
To allow including firms without changes on the outcome variable, we estimate linear probability models. Even in this demanding specification, anti-dumping petitions are more likely to be approved for products important to firms upstream (Table 2, columns 1–6). The results are correctly signed in all models, and in the extended models retain their statistical significance at the 5 percent level for all variants of
$\eta $
.
Success of anti-dumping petitions: firm fixed effects, two-stage least squares
Notes: Columns 1 to 6: linear probability models with robust standard errors, clustered by petition. Columns 7 to 8: 2SLS with robust standard errors, clustered by petition.
$p$
-values in parentheses. Adjusted 95% interval for
${\eta ^{\rm{M}}}$
based on Lee et al. Reference Lee, McCrary, Moreira and Porter2022:
$\left[ { - .17,.44} \right]$
in column 7,
$\left[ {.12,.56} \right]$
in column 8.
Instrumental-variable models. As a second way to address concerns over omitted-variable bias, we pursue an instrumental-variable strategy. We use that currency fluctuations shape sourcing decisions: as the US dollar appreciates, it becomes cheaper to source from abroad.Footnote
99
We calculate our instrument in two steps. First, following Broz and Werfel, we combine bilateral US imports with bilateral real exchange rates to calculate the trade-weighted real exchange rate at the industry level. Second, we combine this measure with data from the input-output tables to calculate the average real exchange rate at the level of an industry’s suppliers. The instrument is a weighted sum of the trade-weighted real exchange rate, with weights corresponding to each industry’s share in industry
$i$
’s total inputs.Footnote
100
The instrument is plausibly exogenous: while an industry’s exchange rate is correlated with filing anti-dumping petitions,Footnote 101 the trade-weighted real exchange rate in upstream industries is not evidently correlated with the success rate of anti-dumping petitions. Similarly, the input shares used in the construction of the instrument are not evidently correlated with the success rate of anti-dumping petitions in an industry. When we control for industry fixed effects and an industry’s own exchange rate in our extended model, we can further rule out that correlation across industry-specific exchange rates creates a pathway from the instrument to the outcome.
The instrument is also plausibly relevant: as the US dollar appreciates for an industry’s supplying industries, sourcing inputs should become more attractive, and
${\eta ^{\rm{M}}}$
should decline as a consequence. Empirically, this is borne out in the first stage, which indicates a negative and, with
$F$
-statistics well above 10, sufficiently strong correlation between the exchange rate of suppliers and
${\eta ^{\rm{M}}}$
.
In the base model, we obtain relatively large standard errors, resulting in a correctly signed but statistically insignificant coefficient estimate for
${\eta ^{\rm{M}}}$
. This coefficient regains statistical significance at the 5 percent level when we include industry fixed effects and the extended set of control variables (column 8).Footnote
102
We also obtain statistically significant effects when using robust 95 percent confidence intervals.Footnote
103
These estimates support a causal interpretation: changes in the extent of the domestic production network that can be traced to exchange-rate movements are reflected in a lower success rate of anti-dumping petitions. We obtain similar results in the reduced-form regressions: as the US dollar appreciates at the level of an industry’s suppliers, the success rate of its anti-dumping petitions declines.
Identification: Sample Selection Bias
Our research design exploits that anti-dumping petitions represent explicit demands by firms. But this entails a selection problem: our sample includes only cases where firms have made such demands. While we cannot eliminate this selection problem, several pieces of evidence indicate that it is unlikely to be accounting for the results.
First, selection bias is a concern only to the extent that the decision to file a petition is conditionally correlated with our measures. This is the case if firms file petitions based on the legal merits of a case (which, beyond control variables, remain unobserved) and their political clout (as captured by our measures). And in this case petitions important to suppliers should on average have weaker legal merits and lower success rates. This would introduce downward bias, reinforcing our interpretation. We discussed tentative evidence for such a pattern earlier: affirmative decisions on products important to suppliers are more likely to be overturned at the WTO.
Second, we adjust for the two arguably most important observable correlates: changes in imports and real exchange rate misalignments. In the online appendix, we also include import volumes as an alternative measure of import penetration.
The first of these arguments relies on a specific type of selection bias; likewise, adjusting for observables does not rule out unobserved sources of selection bias. For example, firms might get nudged by suppliers toward filing anti-dumping petitions, and suppliers might be able to provide crucial information in the process, increasing the success rate of petitions.
Selection problems are a form of omitted-variable bias.Footnote 104 Consequently, we can assess to what extent they are a concern by assessing under what conditions an omitted variable would explain the observed correlation. Following Cinelli and Hazlett,Footnote 105 we use a benchmarking approach. Changes in imports and real exchange rate misalignments are established predictors of the initiation and success rate of anti-dumping petitions. We assess how much stronger than these variables an omitted variable would have to be to invalidate our results. As we report in the online appendix, it would have to be more than ten times stronger as a predictor of the success rate of anti-dumping petitions. It therefore appears implausible that selection problems, and more generally omitted-variable bias, account for the associations we report.
Supplementary Evidence: Congressional Support
These patterns might capture a mostly economic rationale: governments protect firms that create positive spillovers. While this is part of our argument, we go beyond it and contend that these spillovers are politically relevant. In this and the subsequent section, we offer more evidence for this political dimension.
If firms with larger domestic production networks enjoy broader political support, we should observe more members of Congress willing to get involved on their behalf. We obtain all written briefs submitted by members of Congress as part of anti-dumping investigations between 2009 and 2020. We access these from the Electronic Document Information System of the International Trade Commission. We ascertain whether each brief supports or opposes a petition and retain only briefs in support. We then count the signatories for each petition at each stage of the investigation. For petitions that had submissions at several stages, we take the value from the earliest stage. We drop briefs signed by congressional caucuses that had an unusually high number of signatories. On average, a petition receives briefs with about five signatories in total.
We estimate negative binomial regression models, including the same control variables as previously. The results are reported in the online appendix. Petitions on products more relevant to suppliers receive more support from members of Congress. The results are statistically significant and substantively large: compared to a product at the twenty-fifth percentile of
$\eta $
, a product at the seventy-fifth percentile receives twenty additional signatures. We highlight two other noteworthy results. First, members of Congress are less likely to get involved for products that see large import surges. This provides further incidental evidence that their involvement is at least partially politically motivated. Second, members of Congress shy away from supporting petitions on more upstream products. This is consistent with more lobbying competition among firms. Members of Congress appear unwilling to get caught in the middle of these clashes.
Supplementary Evidence: Indirect Exposure of Swing States
Ties to upstream suppliers extend a firm’s geographic footprint. They might do so in politically relevant ways. Building on the argument of Bown and colleagues that firms from swing states are more successful in anti-dumping petitions,Footnote 106 our framework suggests that firms tied into larger production networks have more success in part because they have ties to upstream suppliers in swing states.
To assess this, we first identify the indirect exposure of swing states. Again, we define swing states as those that were won with a vote margin of 5 percent or less in the previous presidential election. We then calculate the weighted employment of an industry’s suppliers in swing states, combining the share of upstream industry output absorbed by an industry as weights with data on employment by the upstream industry from the County Business Patterns.Footnote 107 The resulting measure captures to what extent an industry creates indirect exposure of employees in swing states.
The indirect exposure of swing states is on the causal path from our measures to the outcome. To assess what proportion of the effects is accounted for by this mechanism, we perform mediation analyses.Footnote 108 As we show in the online appendix, all three variables measuring ties to upstream suppliers are strong predictors of the indirect exposure of swing states: firms that draw on a larger domestic production network are more likely to create spillovers into swing states. Depending on the model specification, between 25 and 35 percent of the effects we reported are accounted for by the indirect exposure of swing states. We find no evidence that our mechanism is mediated by direct employment in swing states. These results provide evidence for a specific mechanism through which ties to suppliers upstream become politically relevant.
Conclusion
We introduced a novel explanation for differences in the political influence of firms, emphasizing connections between firms through production networks. Combining data from anti-dumping petitions in the US with original measures of an industry’s position in the domestic production network, we provided empirical evidence for our main argument: the larger the indirect benefits are for firms upstream, the more likely a firm is to see its anti-dumping petition approved.
We complement a vibrant literature on global production networks.Footnote 109 Multinational firms, which pair a credible exit threat with economic size, have long been considered particularly powerful, driving the most recent wave of globalization politically and economically.Footnote 110 At the same time, offshoring has led to pushback against multinationals.Footnote 111 This literature identifies a key trade-off for firms between remaining competitive in economic markets and facing opposition in political markets. Our work suggests a different mechanism underlying some of this trade-off. By offshoring, firms are not only fostering opposition from those who lost their jobs. They are also eroding a domestic political constituency, which previously equipped them with unique political advantages. In the same vein, recent trends of onshoring might fundamentally reconfigure the distribution of political influence across firms as a function of their ability to find suppliers in the domestic market.
While it was not a focus of our analysis, we note a striking empirical pattern: the politics around anti-dumping petitions appear to work differently for cases against market and nonmarket economies. These results provide a backdrop for interpreting the contemporary protectionist turn in US trade policy, which includes an array of policies directed against established market economies, such as Canada, Mexico, and the European Union. Whether these policies indicate a break with the differential politics around market and nonmarket economies remains an open question. However, domestic production networks play an important role in this protectionist turn. Firms with large domestic production networks might find their governments’ ears, while also being less exposed to policy uncertainty and disruptions in global production networks, giving them advantages over their competitors.
How might our argument and findings travel outside the US? We have highlighted several aspects that are not unique to the political system of the US, but also not ubiquitous. To the extent that regulators are more politically insulated elsewhere, we would expect a smaller role for politics. At the same time, the process to investigate anti-dumping petitions is not too dissimilar across countries. For example, Egerod and Justesen document that a prominent predictor of political relevance—asset mobility—shapes the imposition of anti-dumping duties across countries with different political systems and development levels.Footnote 112 Whether patterns similar to the ones we have reported extend to other countries remains, ultimately, an empirical question.
On a different level, our account has novel implications for understanding cross-country differences in firm influence and policymaking. A long-standing literature examines the role of domestic institutions in shaping trade policy choices.Footnote 113 Whether support for a policy is broad or narrow plays a key role here, as does the associated question of economic geography.Footnote 114 Some institutions, such as plurality rule, place a premium on interests that are concentrated; others, such as proportional representation, place a premium on interests that are dispersed.Footnote 115 This literature has produced ambiguous results for trade policy outcomes because not just the costs but also the gains of trade liberalization tend to be concentrated.Footnote 116
We provide a different perspective on the features of narrow versus broad interests. The location of a firm or an industry might be geographically concentrated. But the effects of a policy targeted at this firm or industry can be geographically dispersed. The mechanism we highlight broadens interests considerably because it creates more diverse coalitions across space, across industries, and across occupations. Domestic production networks therefore may play an important role in determining whether a policy affects narrow or broad interests.
This opens up new avenues for understanding differences in policy outcomes across and within countries. Within countries, Busch and Reinhardt highlight the distinction between the economic and political concentration of an industry.Footnote 117 The distinction between the concentration of an industry and the concentration of the effects of a policy targeted at that industry suggests the potential fruitfulness of analogous analyses. Across countries, we offer a distinct interpretation of what makes interests broad. Links to other firms should be most important under institutions placing a premium on broad interests. Extending our argument to different institutional contexts and different policy outcomes thus promises both theoretical and empirical innovations for future research.
Acknowledgments
For generous comments, we thank Stephen Chaudoin, Ben Graham, Ye June Jung, In Song Kim, Lukas Linsi, Amy Pond, Dennis Quinn, Randy Stone, Rachel Wellhausen, and participants in the annual meetings of the American Political Science Association, the European Political Science Association, the International Political Economy Society, the Midwest Political Science Association, and the Bureaucracies in International Relations Conference at the University of Southern California. The IO editors and two anonymous reviewers provided extraordinarily constructive feedback. Michael Stelzig and Jonas Geus provided outstanding research assistance.
Funding
This research was funded by the European Research Council (Grant Agreement No. 101041658, Project PINPOINT), funded by the European Union. Views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.
Open access
