How a Tariff War Became a Mexican Opportunity
The Mexican case turns on a distinction that national trade totals hide: exposure to a foreign shock depends on what a place was already able to sell. This chapter explains how U.S. tariffs on Chinese goods could open space for Mexican suppliers without any new tariff on Mexico, and how the study converts that possibility into a municipality-level measure. The measure is the bridge between a global event and the later claims about exports and jobs. It also tells us what the evidence cannot observe, including whether a particular factory actually moved.
From a Chinese tariff to a Mexican opening
A tariff is a charge that makes an imported product more expensive in the market where it is sold. When the United States raised tariffs on Chinese goods in 2018 and 2019, the immediate change was not that Mexican products became cheaper in an absolute sense. Their relative position improved. A U.S. buyer comparing a Chinese and a Mexican supplier now faced a larger policy-imposed cost on the Chinese option. If the two products were sufficiently substitutable, some demand could shift toward Mexico.
That shift is best understood as a change in the set of commercially viable choices, not as a mechanical transfer of factories. A Mexican producer still has to make the product, meet the buyer's specifications, deliver it reliably, and expand when an order arrives. A tariff can create room for a new supplier, but it cannot supply skilled workers, electricity, logistics, certification, or a relationship with a U.S. buyer. The opportunity becomes an export response only when those other conditions are present.
The paper describes two routes by which Mexico could benefit. An international company might move part of its production from China to Mexico, seeking a shorter supply chain or less exposure to the new tariff regime. Or a Mexican firm already producing the relevant good might expand its output and take market share from Chinese suppliers. Both routes are compatible with a nearshoring interpretation. Neither should be confused with transshipment, in which goods produced elsewhere are routed through Mexico with little change in Mexican production.
The buyer's decision is also more demanding than a simple price comparison. A U.S. manufacturer may need a component in a precise specification, at a predictable volume, under rules that determine origin and quality. A Mexican supplier that already makes a similar product may be able to respond quickly; another firm may face a long delay before it can certify a line or secure the necessary inputs. This is why an external shock can create sharply different local outcomes even when the policy is national. Relative prices open the door, but production systems determine who can walk through it.
The same logic explains why pre-existing capacity should not be treated as an accidental background condition. It is part of the mechanism linking a tariff to jobs. A locality with firms, workers, supplier relationships, and export routines has assets that do not appear in the tariff schedule but affect the speed and scale of substitution. Those assets can produce a genuine local dividend when demand shifts. They can also make the result difficult to reproduce in a place that lacks them.
The difference matters for the labor question. If Mexican customs records show more exports to the United States, that is evidence of a changed trade flow. It is not by itself evidence that Mexican workers produced all of the additional goods, that local firms invested in new capacity, or that the gains reached households broadly. The study therefore connects exports to municipal employment, labor income, and informality. That connection cannot identify every physical movement inside a supply chain, but it can test whether places more exposed to the tariff shock also experienced a distinct local adjustment.
Mexico's national position made such an adjustment plausible. Its proximity to the United States lowers some transport and coordination costs, and its trade integration gives firms an established route into the U.S. market. But “Mexico was well positioned” is still too broad to explain the result. The relevant product may have been made in one municipality and absent from another. A locality with an established manufacturing workforce and a history of exporting may respond quickly; a locality without that base may not respond at all, even if it is equally close to the national border or covered by the same trade agreement.
This is the role of pre-existing specialization. Before the trade war, each municipality already had an export basket: a mix of products sold to the United States. The U.S. tariff schedule also varied by product. A municipality whose basket was concentrated in goods targeted by the U.S. tariffs on China was positioned to receive a larger competitive shock than a municipality exporting products that were not targeted. The global event was common, but its local intensity was not.
This is a more precise claim than saying that some municipalities were simply “more globalized.” A place could export a great deal and still have relatively little exposure if its main products were outside the targeted lines. Another place could have a smaller overall export base but a larger exposure if its products were concentrated where the tariff gap widened. The measure therefore combines two facts: how much a municipality sold to the U.S. market before the shock, and how closely that basket matched the products affected by the U.S.–China dispute. It is the interaction, not either fact alone, that defines the opportunity being studied.
That point changes the question from “Did Mexico benefit?” to “Which Mexican places were most exposed to the product lines where Chinese competition weakened?” It also prevents a common category error. Proximity, trade agreements, and national manufacturing capacity are enabling conditions. They explain why Mexico could be a plausible alternative supplier. They do not determine the treatment intensity for a specific municipality. The export basket does that.
How exposure turns a global event into a local comparison
The study builds its exposure measure by weighting the U.S. tariff on each Chinese product by that product's share of a municipality's exports to the United States during a pre-war base period. In plain language, it asks: how much of this municipality's existing U.S.-bound business was concentrated in product lines that became more protected against Chinese competition? A product that represented a large share of a municipality's export basket receives a large weight; a product that represented a tiny share receives a small one.
The base-period weights are important because they are fixed before the main shock is evaluated. The measure does not first look at which municipalities expanded and then call them exposed. It starts with what they were already selling, applies the external tariff changes to that basket, and calculates a change in weighted exposure. Export outcomes are then compared between the pre-trade-war period of 2016–17 and the post-period extending through 2018–21, while labor outcomes use the available 2015 and 2020 population observations. The exact periods differ by outcome, so a coefficient always belongs to its own comparison.
A useful way to read the measure is as an intensity score, not as a new tax paid by Mexican firms. Suppose two municipalities sell different mixes of goods to the United States. If the first municipality is heavily concentrated in products on which U.S. tariffs on China rose, the same international policy change gives it a larger exposure score. If the second municipality sells products largely outside those tariff lines, its score changes less. Neither municipality needs to have chosen a new policy. Their difference comes from how the shock intersects with their inherited production structure.
The paper estimates the relationship between this exposure change and local outcomes with a weighted ordinary least squares model. The specification controls for changes in U.S. tariffs on Mexico, state fixed effects, and initial demographic and employment characteristics. Observations are weighted by initial manufacturing exports to the United States, so the estimates emphasize municipalities with larger initial export bases. These choices are not decorative statistical details. The Mexico-tariff control helps separate the China shock from changes in Mexico's own market access; state effects absorb some regional differences; initial characteristics address observable differences that could otherwise be related to later outcomes.
It helps to think of the controls as safeguards against particular confusions, not as a guarantee that the model has removed every source of bias. Holding state effects constant does not make a northern municipality identical to a southern one. Controlling for initial literacy, rural population, or employment composition does not reveal every firm decision or local policy. Weighting by initial exports also changes the practical question: the estimate is especially informative about places with meaningful manufacturing trade, rather than about the average municipality in the country. A careful reading therefore asks what comparison the model has made and which differences remain outside it.
The timing matters for the same reason. The exposure is built from earlier export patterns, while the outcomes are measured later. That ordering helps prevent the outcome from defining its own treatment, but it does not eliminate the possibility that high-exposure municipalities were already on a different trajectory. The paper addresses that concern with additional tests later in the analysis. At this stage, the important point is that the exposure measure supplies a plausible source of differential intensity; the identifying assumptions determine how strongly the resulting association can be interpreted.
The central comparison is still conditional. The design asks whether municipalities with greater predicted exposure changed more after the trade-war period, once these controls are included. For that comparison to support a causal interpretation, the more- and less-exposed places must not already have been moving along different outcome trends for unrelated reasons. This is the logic behind a parallel-trends assumption. It is not a claim that the municipalities were identical. It is a claim about what their trajectories would have looked like without the specific shock.
For example, if high-exposure municipalities had already been adding manufacturing jobs faster for several years, a later export increase might partly continue that earlier trend rather than result from the tariff change. The model can control for observed starting conditions, but the counterfactual trend is never directly visible. Evidence that the groups moved similarly before 2018 would make the post-shock divergence more persuasive; evidence of a prior divergence would weaken the interpretation. This is why the later event-study evidence matters, and why the exposure measure should be read together with the paper's design rather than as a standalone fact.
The weighting makes the mechanism visible. The tariff schedule supplies the external change; the pre-war export basket supplies the local weights; the resulting exposure measure supplies a comparable intensity across municipalities. The outcome comparison then asks whether that intensity predicts more export growth or different labor outcomes. The measure does not claim that tariffs were the only thing happening in Mexico. It creates a disciplined way to ask whether the pattern of change lines up with the product-specific shock.
The sample is equally important to the interpretation. The municipal trade data provide product-level detail for 411 municipalities, but the estimation sample contains 186 municipalities that exported manufactured goods to the United States during the base period. The paper reports that these municipalities accounted for about 47.5 percent of Mexico's manufacturing exports during 2016–21, about $180 billion in exports to the United States in 2021, and more than half of the country's population. That is substantial economic coverage, but it is not the same as a representative draw of all 2,457 municipalities.
Confidentiality rules remove small trade flows and firms that are the only or one of very few exporters in a locality. The municipalities that remain tend to have stronger labor-market outcomes, more manufacturing and services activity, and lower informality than the 2,271 municipalities outside the sample. This creates a meaningful but narrow inference. The study tells us about places already connected to manufacturing exports, exactly the places most able to respond to a sudden opening in the U.S. market. It does not show that a non-exporting rural municipality would receive the same opportunity.
The exposure measure also leaves several mechanisms unresolved. It does not record whether a multinational relocated a line from China to Mexico, whether an existing Mexican firm expanded, or whether a supply chain rerouted goods between Mexican municipalities. It is not a direct measure of foreign direct investment, plant construction, productivity, or wages. Nor does it capture every transaction that confidentiality rules exclude. The measure is valuable because it links a product-level external shock to a local comparison; it is limited because a proxy for exposure is not a full history of production.
That distinction will matter when the export results are read. If the more-exposed municipalities did not increase exports more than less-exposed municipalities, the proposed mechanism would lose force. If they did, the result would establish a market response, not yet its complete physical cause. The next question is therefore empirical and specific: how much more did these municipalities export to the United States, where were the gains concentrated, and did the same pattern appear in other destinations? Only after that link is established can the series ask what happened to local jobs.