World merchandise trade in the first half of 2025 came in materially better than the tariff arithmetic said it should. That has been read in some quarters as evidence that tariffs matter less than economists claimed. We think the more careful reading is that two effects have moved activity between periods, and both of them run out.
Source: WTO Global Trade Outlook and Statistics, April 2025, with the August interim estimate as subsequently reported by the WTO. Forecasts refer to the average of export and import volumes.
Effect one: importers bought early
When a tariff increase is announced with a lead time, the rational response for an importer with storage is to bring purchases forward. This is not a subtle effect and it is not new. It shows up as a surge in import volumes ahead of the effective date, followed by a corresponding hole once inventories are worked down.
The WTO has been explicit that the negative impact of tariffs in 2025 was smaller than predicted partly because new US tariffs were suspended until August, retaliation from other economies was limited, and there were numerous exemptions. Each of those is a timing or scope factor rather than a demand factor.
The important property of frontloading is that it borrows from the future at a one-to-one rate. A unit brought forward is a unit not bought later. Any forecast that extrapolates the strong half-year without netting off the inventory unwind is double counting.
Effect two: a technology cycle unrelated to trade policy
The second effect is a surge in demand for AI-enabling goods, concentrated in Asia and North America. This is genuine incremental demand rather than a timing shift, and it has been large enough to offset a meaningful part of the tariff drag in the aggregate figures.
It is also a different economic process wearing the same statistical clothing. Semiconductors, servers and networking equipment moving across borders because a capital cycle is running is not the same as consumer goods trade responding to relative prices. Aggregating them tells you the total, and conceals the fact that one component is cyclical capital expenditure and the other is structural demand facing a new tax.
What we would do with the number
Decompose it. For any commercial planning purpose, the useful series is merchandise trade excluding AI-related capital goods, and excluding whatever estimate of inventory build can be constructed from the customs data. That series is considerably weaker than the headline.
The forecast implication is straightforward. If frontloading unwinds and the technology cycle merely holds steady rather than accelerating, 2026 volumes will be much weaker than the aggregate 2025 outturn suggests, and the weakness will look sudden to anyone who was tracking the headline. The WTO has already cut its 2026 projection from 2.5 percent in April to 1.8 percent in its August interim estimate. We would not be surprised to see that cut again.
The uncertainty channel is separate and usually ignored
One further point. A tariff that might be imposed, removed or renegotiated is not the same as a tariff at its expected value. Firms making irreversible investments in supply chain assets discount for the variance, not just the mean. That shows up in investment data before it shows up in trade data, and it is a persistent effect rather than a cyclical one, because a shortened value chain reduces the trade intensity of any given level of output permanently.
Sources
- WTO, Global Trade Outlook and Statistics, April 2025 (PDF)
- WTO Global Trade Outlook update, October 2025 (PDF), for the subsequent revision