A tariff on an integrated neighbour is a different instrument

The standard mental model of a tariff is simple enough to be misleading. A foreign producer sells into your market; you levy a duty; the good becomes more expensive; domestic producers gain share. Whether you think that is good policy or bad, the mechanism is at least clean.

It stops being clean the moment the trading partner is an integrated neighbour, because the central assumption fails. The good is not foreign. It is partly yours, several times over.

The border-crossing problem

North American manufacturing, and automotive production in particular, is organised as a single production system that happens to have an international border running through it. A component can be cast on one side, machined on the other, assembled back on the first, and installed in a vehicle finished on the second. Each crossing is a customs event.

Under a tariff, each crossing is also a taxable event. The duty is not levied once on a finished foreign good; it is levied repeatedly on partially-completed work, much of which represents value added by domestic labour and domestic inputs. The effective rate on the finished vehicle bears very little resemblance to the headline rate, and it falls substantially on producers the policy was meant to protect.

This is why tariff schedules between integrated economies grow rules-of-origin machinery, drawback provisions and content thresholds. The complexity is not bureaucratic excess. It is an attempt to answer a question that has no clean answer in a system built on the assumption that it would never need to be asked: whose good is this?

What the trade data does and does not say

Two-way goods trade between the United States and Canada runs at roughly three quarters of a trillion dollars a year, and it moved recently.

US goods imports from Canada, as reported by each side

US dollars, billions, free-on-board valuation on both sides

Two things are visible there, and the second is the one people miss.

The flow fell about 7% year on year on the US measure, and about 8% on the Canadian one. That is a real contraction in a mature trading relationship, and it is the kind of move that shows up in the loan book of any bank lending to manufacturers, hauliers or agricultural exporters near the border long before it shows up in national statistics.

The second thing is the gap. Valued the same way on both sides, the United States records $382bn of imports from Canada in 2025 and Canada records $401bn of exports to the United States. Same goods, same year, nineteen billion dollars apart. This is ordinary, since the two customs authorities differ on transhipment treatment, on origin attribution, and on when a good is counted, but it is a useful reminder that trade figures are administrative records of two governments' paperwork, not measurements of a physical fact.

The valuation basis is a third lever on top of that. The United States publishes the same flow at $392bn on a landed-cost basis, which includes the freight and insurance of getting it there; set that against Canada's $401bn and the discrepancy halves without a single physical fact having changed. Anyone building an exposure model on one side's numbers is working with a specific bureaucratic construction of reality, and should say which one, on which basis.

The good is not foreign. It is partly yours, several times over, and the duty falls on every crossing.

Where it lands, for a lender

Tariff analysis usually stops at the macro level: effect on GDP, on inflation, on the currency. For an institution with a book, the interesting effects are considerably more local and arrive in a specific order.

Working capital before margins. The first thing a duty does to an exposed importer is consume cash. The duty is payable at the border, in advance of the sale that funds it. A borrower whose margins can absorb a tariff perfectly well may still breach a facility because the timing broke. This shows up as a line utilisation problem, not a credit-quality problem, and it is early.

Concentration through the supply chain, not the category. A commercial book with no obvious trade exposure can have a great deal of it. The equipment dealer, the logistics operator, the industrial landlord whose tenant is a supplier, the agricultural borrower selling into a processing plant across the border: none of these are classified as trade exposure, and all of them depend on the same flow. The correlating factor is the supply chain, and it will not appear in any reporting category.

Currency, quietly. A durable shift in the trade balance moves the exchange rate, which moves the competitive position of every borrower with cross-border revenue or cost, including the ones with no direct tariff exposure at all.

The asymmetry in the analysis

There is a structural bias in how this gets modelled, and it does not come from anyone's politics.

The direct effect of a duty on a named importer is calculable to a reasonable precision: a rate, a volume, a margin. The indirect effects, through the supply chain, the currency, retaliation, and the borrowers who are not classified as trade exposure at all, are not calculable to anything like that precision, and they are usually larger.

A model reports what it can compute. So the output tends to be confident about the small part and silent about the large one, and it looks rigorous precisely in proportion to how much it has left out. That is not a criticism of modelling; it is an argument for reading any tariff estimate alongside the question of what could not be put in it.