One of the many ways conventional economic theory hinders our discussions of trade is that it gets us thinking about goods “produced” in one country and “consumed” in another. Mexicans grow tomatoes, drill oil, sew shirts, and assemble cars; Americans eat, burn, wear, and drive them.

Most trade in the real world does not look like this. What you have, rather, are commodity chains, where different parts of the production process take place in different countries. In most cross-border transactions, the buyers are not consumers, or even distributors, but producers who use the imported goods as inputs. And in many cases, the relevant transactions are not arm’s-length market exchanges, but transfers within a single corporate structure. Even the final purchasers may not be consumers: in general, investment goods and exports have higher imported content than consumption goods do.

Case in point: US-Mexico trade. With Donald Trump’s proposed 20 percent tariff, I was curious what US imports from Mexico actually look like. Here’s what the Census says:

As we can see, consumer goods account for only about a quarter of US imports from Mexico. Given that a large fraction of the service imports are tourism, the total share of consumption in US imports from Mexico will be a bit higher, between 30 and 35 percent. (But presumably tourism would not be affected by a tariff.)

The remainder is divided about evenly between industrial inputs (raw materials plus intermediate goods like cloth, steel, auto parts, etc.) and investment goods. Machinery and equipment, including computers, account for an impressive 25 percent of Mexican exports to the US. Petroleum products, despite the widespread perception of Mexico as an oil exporter, account for less than 5 percent.

OK, so why does this matter?

Well, it’s enough, to begin with, that most of us have a distorted idea of what “trade” involves. It’s always dangerous to talk about something at a high level of abstraction without a clear sense of the concrete reality involved — even if, in a given case, the abstract description works fine.

But in this case I don’t think it works fine. I think our model — of one country producing and the other consuming — misleads us in some important ways about the likely impact of something like Trump’s tariff.

First of all, the fact that trade is normally part of a longer commodity chain helps explain why trade flows are often insensitive to changes in relative prices. Notice, for instance, the $50 billion auto parts imported from Mexico — about one-seventh of total Mexican exports to the US. Some of these parts may be generic, but most presumably represent investment by the parent company in a specialized supply chain. There’s little or no short-run possibility of substituting components from elsewhere in response to changes in relative prices.

And insofar as the importer and exporter are part of the same corporate structure, the relevant price is an administered one that, in the short run, depends more to do on internal accounting practices than with exchange rates, tariffs, or other macro phenomena.

In the long run, of course, this kind of intra-corporate trade is responsive to relative prices — production wouldn’t be located in Mexico to begin with if costs weren’t lower there. But for intra-corporate trade, the long-run response to price difference will mainly come on the supply side, not the demand side.

In other words, if there were a large, persistent rise in prices in Mexico relative to the US, that might well eventually reduce Mexican exports, but the main way that would happen is firms disinvesting in production capacity there — not expenditure switching by consumers in response to higher prices. This kind of trade is the excluded category in orthodox trade theory — it doesn’t respond rapidly to changes in prices, but neither does it reflect any fundamental differences in natural resources or other “endowments” between countries.

The second reason the composition of trade matters is when we look at the distributional impact. If Mexican exports were just corn tortillas, as some people seem to imagine, it would be relatively easy to answer “who pays” for a tariff. You’d just estimate the price elasticities of supply and demand and do the math. (OK, maybe not that easy.)

But with a high proportion of intermediate and investment goods, it’s much trickier. Especially since there are profits collected at a number of points along the commodity chain, so an increase in the price of Mexican imports at the border is not necessarily passed on to ultimate consumers. Some fraction will presumably come out of the various rents along the way. Even the broad claim that it must ultimately be Americans who pay doesn’t hold, since a large fraction of imports are inputs for export industries.

The third reason follows directly. Insofar as the final users of imports are exporters, tariffs and other relative-price changes will have less of an effect on the trade balance. In the old days of import substitution industrialization, people took this problem seriously — they recognized that the effective rate of protection for a given industry might be quite different from the statutory rate, depending on how dependent the industry was on imported inputs. In this case, if a large fraction of Mexican imports are destined for US export industries — and they are — then a tariff on Mexican goods will improve US competitiveness less than the textbook analysis would predict.

Finally, the disproportionately large share of intermediate and investment goods in international  trade should factor into how we think about trade in general. The more I study this stuff, the more I get the sense of international trade and finance as a world unto itself — sitting on top of, dependent on, the rest of the economy, but irrelevant to most of the routine activity of extracting human labor to meet human needs. Imports are purchased to make exports, which will be purchased to make more exports to somewhere else.

An exaggeration? Yes, but maybe not an extreme one. Somewhere in Civilization and Capitalism, Fernand Braudel describes the early modern world as an archipelago of towns scattered around the margins of an interior world — whether in France or India — that remained focused on immediate, local needs. The boundary regions were more connected to each other than to their own hinterlands perhaps only a few miles away. Mutatis mutandis (and there’s a lot of mutatis!) I think something like this applies today. Traders and producers for trade are mostly much more integrated with each other than with the rest of us. Your t-shirt is a valid counterexample, but not necessarily a representative one.

In summary: most US imports from Mexico are intermediate and investment goods, not consumer goods. And a tariff on Mexican goods is more likely to raise costs for US businesses — including for US exporters — than to lead people to substitute American-made goods for Mexican ones.