Why Some Mexican Municipalities Captured the U.S. Market
The exposure measure only matters if it predicts a different export response. This chapter reads the World Bank study's export evidence at two levels: individual products and Mexican municipalities. The central task is to keep three claims separate—a marginal relationship, a model-based translation of the observed shock, and the geographic and destination pattern. That separation shows both how strongly the tariff opportunity reached the U.S. market and why the result should not be read as a uniform Mexican export boom.
Two estimates answer two different questions
The first estimate is at the product level. The study asks whether Mexican exports to the United States grew more in product lines where U.S. tariffs on Chinese goods increased more. In the preferred specification, a 1 percent increase in the average ad valorem U.S. tariff on Chinese manufactured products is associated with nearly a 2.5 percent increase in Mexico's exports of the corresponding products to the United States.
An ad valorem tariff is a percentage of a product's value, so the estimate compares proportional changes rather than dollar changes. The result is also an elasticity: it describes how responsive Mexican exports were, on average, to a change in the tariff measure. It does not say that every product gained 2.5 percent, or that a product outside the affected manufacturing sample would respond in the same way. The specification includes changes in U.S. tariffs on Mexico, sector fixed effects, and pre-existing export trends. The point of those controls is to ask whether the China-specific tariff change adds explanatory power beyond broader product and market trends.
This product-level result is useful because it keeps the first link close to the policy shock. The tariff schedule varies by product, and the export data are also organized by product. If Chinese competition became less favorable in a particular line and Mexican exports of that line rose in the U.S. market, the pattern is consistent with substitution toward Mexico. It still does not reveal whether an existing Mexican factory expanded, a new operation arrived, or a supply chain changed in some other way. But it establishes that the opportunity was visible in the products where the competitive gap changed.
The second estimate aggregates that logic within municipalities. The preferred municipal specification finds that a 1 percent increase in a municipality's exposure to U.S. tariffs on Chinese goods is associated with a 4.3 percent increase in its manufacturing exports to the United States. This is not a 4.3-percentage-point increase in exports. It is a proportional response to a proportional change in the weighted exposure measure. The unit of exposure is a municipality's product basket, and the outcome is the change in that municipality's U.S.-bound manufacturing exports.
The two estimates are related but not interchangeable. The product estimate asks how individual product lines respond to tariff changes. The municipal estimate asks how places with different mixtures of those product lines respond. The second can be larger because the municipality-level measure captures the combined shift across a basket and connects it to a local export total. It also carries the additional risks of aggregation: a municipality can change its mix of products, and a local total can conceal which firms or product lines account for the movement.
This is why an elasticity should be read as a relationship between proportional changes, not as a forecast for a particular factory. A municipality with a small initial export base and a high exposure score is not automatically expected to add the same number of dollars as a much larger municipality. Likewise, a large percentage response can represent very different employment or investment consequences depending on the starting scale and the way firms organize production. The paper's estimates are informative about responsiveness; they do not remove the need to ask what the starting conditions were.
The paper then applies the municipal coefficient to the actual change in exposure during the period. That calculation implies a 41 percent increase in manufacturing exports to the United States among the municipalities in the estimation sample, equivalent to about $68 billion. Expressed against Mexico's total manufacturing exports, including activity not captured in the municipal dataset, the paper reports an equivalent share of 15.6 percent.
Those figures are important, but they answer a different question from the 4.3 percent marginal estimate. The 4.3 percent is the estimated response to a one-percent increase in exposure. The 41 percent and $68 billion are translations obtained by applying that estimated relationship to the observed change in exposure for the sample. The 15.6 percent compares the translated amount with a broader national export denominator. None of the three should be treated as a directly observed counterfactual in which researchers can see exactly what Mexico would have exported without the trade war.
The translation depends on assumptions that deserve to remain attached to it. It treats the estimated relationship as sufficiently linear over the relevant change and assumes homogeneous treatment effects when extending the coefficient across the sample. It also depends on the sample's coverage and on the paper's comparison periods. A model-based calculation can be a useful way to express magnitude, but it is not the same kind of evidence as a customs record showing a shipment.
That distinction is especially important in public discussions of nearshoring. A headline that Mexico gained $68 billion can suggest a single national event, while the underlying estimate describes the manufacturing exports of a selected set of municipalities and then translates a regression coefficient into a dollar amount. The calculation helps convey scale; it does not erase selection, uncertainty, or the possibility that the response varied across firms and places.
The denominator changes the meaning again. The sample-wide translation concerns manufacturing exports to the United States among municipalities with the necessary data and prior export activity. The national-share comparison uses Mexico's total manufacturing exports, including activity outside that municipal dataset. Moving between those denominators is useful for context, but it can make a local estimate sound more universal than it is. A reader should retain the chain: estimated marginal response, applied to an observed exposure change, then compared with a broader total.
The visual distinction is the point. The product-level and municipal relationships reinforce each other, but the model-based total is a further step. Keeping the steps visible makes it possible to acknowledge a large estimated export response without presenting an extrapolation as an observed national fact.
The geography and destination tests narrow the story
The export gains were concentrated rather than evenly distributed. The paper identifies northern and south-central states as especially exposed to the U.S. tariffs on Chinese goods. This geography is consistent with the role of existing manufacturing and U.S.-oriented production networks, but the exposure measure supplies the more precise explanation: these places were more concentrated in the affected product lines.
Geography matters here because it is a summary of several underlying relationships. Northern municipalities may benefit from proximity and established cross-border production connections. South-central municipalities can also be exposed when their manufacturing base matches the tariff-targeted products, even though they are not simply interchangeable with border locations. A map of high exposure is therefore not a map of “the best places in Mexico.” It is a map of where the external tariff schedule overlapped most strongly with the export structures that were already present.
This is also why the result should not be interpreted as proof that location alone caused the export gain. The same state can contain municipalities with different products, firms, and labor pools. Conversely, two municipalities in different regions can face similar exposure if their export baskets look alike. The geographic pattern is valuable because it shows concentration, but the product-basket mechanism explains why the concentration appears.
There is a useful distinction between a place being near a market and a place being ready for a particular market opportunity. Distance can make delivery easier, but it does not tell us whether a municipality makes the relevant good or has the workers and firms to scale it. The exposure measure captures readiness only indirectly, through prior exports. That makes the regional pattern economically meaningful: it identifies where the global shock met an existing local specialization.
The regional result should therefore be treated as a distributional fact as well as a production fact. A national export gain can be concentrated in a few corridors and still have a large aggregate value. But the workers, services, and public institutions in those corridors experience the adjustment differently from people in places that do not receive new orders. The geographic concentration is the first reason the later labor analysis must remain municipal rather than relying on Mexico-wide averages.
The destination test provides another way to distinguish a U.S.-market response from a general export upswing. If municipalities more exposed to the tariffs on Chinese goods were simply located in industries experiencing a worldwide surge in demand, their exports to other destinations might be expected to rise as well. The study therefore examines manufacturing exports from those municipalities to the rest of the world.
After the controls, the relationship between tariff exposure and exports to non-U.S. destinations is not statistically significant. The result is consistent with a specific change in access to the U.S. market: exports rose where the U.S.–China tariff gap created an opening, but the same exposure did not produce a comparable, broad-based increase in shipments to every other destination. This makes the story more focused than “targeted sectors boomed everywhere.”
A non-significant estimate is not proof that no municipality increased exports elsewhere, and it is not proof that demand outside the United States was irrelevant. The paper reports large standard errors in the non-U.S. specifications, so the test is better understood as evidence against a clearly detectable general export effect than as evidence of zero change in every other market. The appropriate conclusion is comparative: the observed signal is strongest in U.S.-bound exports.
That is the difference between absence of a statistically detectable pattern and evidence of no activity. The destination test asks whether the exposure measure predicts a systematic change across the municipalities and specifications examined. It does not inspect every firm, product contract, or destination. Its value is that it makes a broad alternative explanation less persuasive: if the affected sectors were simply riding a worldwide wave, the same municipal exposure should have been more visible outside the United States as well.
The destination result also helps separate opportunity from capacity. A municipality may have the ability to produce a targeted good, but the measured shock was generated by a change in the U.S. competitive environment. If exports to the United States respond while exports to the rest of the world do not show a comparable response, the market access channel becomes more plausible. It does not prove a single firm-level pathway, but it connects the location-level estimate to the particular market where Chinese goods became more expensive.
Taken together, the evidence establishes a first link in the larger chain. Product lines exposed to higher U.S. tariffs on China saw larger Mexican exports to the United States. Municipalities whose pre-war baskets were concentrated in those lines saw an even stronger local export response. The response was concentrated geographically and was not mirrored by a statistically significant increase in exports to other destinations.
What this evidence does not yet tell us is how the additional export activity was produced or distributed within each municipality. It cannot by itself distinguish a new Mexican plant from an expansion by an existing firm, and it does not show whether workers received higher pay, more formal jobs, or simply more opportunities to enter employment. It also does not settle every form of trade deflection: a local non-U.S. export test addresses destination patterns, not every possible movement of goods inside a national supply chain.
That is why export growth is best treated as the gateway to the labor question, not its answer. The same additional shipment can be produced through overtime, new hires, migration, productivity improvements, capital investment, or a combination of them. Those pathways have different consequences for average pay and job quality. The municipal export result tells us where to look for the adjustment; the labor data tell us which margin moved after the opportunity arrived.
Those limits do not make the export results weak. They define their proper role. The results show that the weighted exposure measure was associated with a distinct U.S.-market response, and they make a local labor-market question unavoidable. If more-exposed municipalities exported more, did the gains stop at customs records, or did they alter the number and quality of jobs? The answer requires separating total labor income from average earnings and following the response across workers, sectors, and formal status.