Essay

The Tariff Bill Has a Slow Third, and It Is Still Arriving

Twenty-six percent of the 2025 tariff increase ends up in consumer prices, and that is the whole bill, not a first installment, according to the team that cut the number into its three channels. Two thirds of it lands almost on impact. The last third moves at the speed of supply chains, nine to twelve months per tariff. The tariffs kept coming, February 2025 through a fresh round this July, so the slowest installments are still in transit. In May, nearly half the firms that pay tariffs told a Fed survey their price increases are not finished.

The question “who pays the tariff” usually gets answered with one number, because one number is what a headline can carry. Mary Amiti, Sebastian Heise, and David Weinstein kept the one number, about 26 percent of the tariff increase reaching consumer goods prices, and did the more useful thing underneath it. In what they call the first estimate to separate all three channels of pass-through (rather than assuming the indirect ones from an accounting table), they put each channel on its own clock. The direct channel is the sticker: an imported good gets taxed, its price moves almost immediately, and that is 64 percent of the whole effect. The indirect 36 percent is the quieter machine: a domestic manufacturer pays more for foreign steel, chips, and components, and with import competition weakened it takes a little margin too. That propagates invoice by invoice, contract renewal by contract renewal, and needs nine to twelve months to finish. Flash, then thunder: one strike, two arrival times. Worth the flag before leaning on it: this is a working paper posted in July, not yet peer-reviewed. And the number is a relative estimate: more-exposed goods against less, macro backdrop held fixed, goods prices only.

I recognize this move because it is the one I was trained to make before I ever touched a company. Take a claim too fuzzy to argue with, “tariffs raise prices,” and force it into something you can score: how much, through which channel, on what clock. That instinct is the same one that turned “which satellite orbit is better” into a hexagonal grid you could actually compute an answer on, the same one that turns a vague worry into a number with error bars. A fuzzy claim is unfalsifiable. A scored one has a due date.

Same economists, different instrument. Weeks before the paper posted, the same New York Fed group, Amiti and Heise on both bylines, published its May survey of regional firms: put down the price data and ask. Among importers, 70 percent of manufacturers and 40 percent of service firms had directly paid tariffs in the prior twelve months. Of those, only 20 percent of manufacturers and 30 percent of service firms report having fully passed the cost through. Forty-four and 47 percent say more increases are coming, mostly dated inside the next six months. That is aggregate self-report from one team running two instruments. Read it as the survey shadow of the lag the regressions measure. The authors flag that they cannot tell whether firms are answering for one round of tariffs or for the whole sequence; for a claim about overlapping clocks, that ambiguity is the point. The shadow says the slow third is still moving.

Now the objection that actually bites: most of the bill was never in the mail. The paper itself shows where the other 74 cents stop, and its ledger runs per dollar of duty actually paid. The consumer-goods basket is mostly domestic to begin with; the pre-retail import share is about a third. Foreign exporters barely cut prices, so American importers wrote the checks at the border. And at the retail stage the paper measures a 1 percent pre-retail cost increase surfacing as 0.56 on the tag, distribution margins eating the difference. (Customs also collects only about two-thirds of the statutory schedule, exemptions and rerouting doing the rest; score the same pass-through against the statutory dollar and 26 falls to roughly 17. That wedge sits upstream of the bill, not inside it.) Cavallo’s price tracker at Harvard (360,000 products across five retailers) watched retail prices level off by late February. He said it on the record: “Most of the pass-through has likely already occurred, assuming tariffs do not increase further.” The paper’s own footnote concedes its markup channel runs about half the size two of these same authors measured in 2018-19. The strongest published dissent, Mehrotra and Waugh at the Minneapolis Fed, finds price changes through December 2025 only weakly related to tariff exposure at all. Concede the shape of it. The missing 74 cents are a basket that is mostly domestic, margins that ate what crossed it, and a sliver the exporters gave up. The slow third is a claim about the 26 cents that do arrive.

The clock survives the concession. The lag structure is econometric, measured from realized prices, so the obvious pushback, that a plan is not a receipt, cuts against the survey and leaves the regressions standing. The indirect coefficients barely register at six months and turn significant only at nine to twelve; roughly half the slow third is in the door by month six. The survey’s own splits keep it honest, with 30 percent of tariff-paying manufacturers reporting they are done raising prices, recouped or not. Every tariff starts its own clock, and they arrived in waves: February and March 2025, the April baseline, steel and aluminum in June, copper and the country rates in August, Section 232 rounds on lumber, trucks, and semiconductors late in the year.

The clocks run both ways, too. November cut China by ten points, a cut the Fed Board reckons trimmed a few tenths off the cumulative goods-price increase. February cut India, and then the Supreme Court struck the IEEPA tariffs outright, swapping them for a lower stopgap surcharge. The surcharge died on schedule in July, and Section 301 duties took its place the same minute. The average statutory rate sits near 11 percent, about a third below its September peak, scheduled to reach 11.8 by December. July’s tariffs run their own nine to twelve months, into mid-2027; so do the cuts, in reverse.

And the other side of the ledger is real. The authors’ realized tally through February (offered as illustrative) ran about 2.2 points on the goods they track, a fifth of the consumption basket, with the indirect channels already contributing about 30 percent of it. The Fed Board’s statutory tracker calls pass-through “effectively complete” at five to nine months per round, against a dollar-for-dollar benchmark. Its own footnote closes the apparent fight: tariffed goods retail at about 2.7 times their border price, so dollar-for-dollar complete converts to roughly 37 on the proportional scale where this paper reads 26 and Cavallo’s shelves read 24. The conversions are loose and every band is wide; the paper’s own interval runs 14 to 44 percent. But three trackers land in one ballpark: the duty dollars arrived, and the percentage markup on top of them never did. A Dallas Fed note in May is literally titled “Effects of realized tariff changes on PCE prices peaked in first quarter 2026”. The money has even started flowing backward: since the February ruling, about 100 billion dollars of the struck IEEPA duties has been certified for refund, Apple alone above two billion. A firm made whole has less reason to keep raising prices, though the refunds land with whoever paid the duty, and the companies are so far split on sending any of it further down the chain. So the tail is thin and specific: the late waves’ slow third plus what July restarted, tenths of a point on goods prices, stretching into 2027.

The fast two thirds got counted in 2025, at the border and on the shelves. The slow third is the part a headline cannot carry, and it is why the paper is worth more than its number: it forced “tariffs raise prices” to name its channels and show its clock. Those clocks now run into mid-2027. The grade will not be read off a shelf price; it is read off the gap between exposed goods and sheltered ones, the only exam the paper claims it can sit. I intend to check.

Sources: