Nineteen months
of moving ground
The United States raised the tariff rate implied by its policies from 2.4% to roughly 28% and back to 11% in nineteen months. The more consequential number is three — the number of times the legal foundation beneath those tariffs was replaced, twice under court compulsion.
● filled markers are published figures · dashed and unmarked stretches are interpolated between them.
Read as trade policy, this was a campaign to reshore manufacturing and close the deficit. On those terms the evidence to date does not support it. Read as a constitutional experiment — how far a president can tax imports without Congress — it produced a definitive answer in February 2026, and the rebuild since then tells you more about the next decade than the tariff rate does.
Every figure carries a source tag you can open. Every chart states its source, vintage, unit, denominator and whether values are observed, modelled or interpolated. Where the evidence supports only a co-movement rather than a causal estimate, the text says so.
The sequence
Sixty-three actions across nineteen months, each tagged by the statute it rests on and by source. Filter by phase, authority or target; select any entry for the analysis.
Day one. Directs Commerce, Treasury and USTR to review trade deficits, currency practices and the USMCA, and to identify authorities for unilateral action. No tariff yet — this is the scoping document for everything that follows.
First use of IEEPA in history to impose tariffs. The statute had been a sanctions tool for 48 years. The stated basis is narcotics and migration, not trade — a framing chosen because it needs no investigation, no hearing and no finding of injury.
China counters with duties on US coal, LNG, crude, farm machinery and pickup trucks, plus an antitrust probe into Google and export controls on tungsten and tellurium. The template for the year: narrow, calibrated, and aimed at politically sensitive US exports.
The 2018 steel and aluminium tariffs are reinstated in full: aluminium up from 10% to 25%, all country exclusions, quotas and product carve-outs cancelled. Steel must be melted and poured, aluminium smelted and cast, in the US to escape. This is the layer that will still be standing in 2026.
Orders a bespoke tariff rate for every trading partner, computed from tariffs, VAT, non-tariff barriers, exchange rates and bilateral balances. Triggers a wave of diplomatic outreach — and two months of guessing at a formula that turns out to be the bilateral goods deficit divided by imports.
Canada answers with 25% on C$30bn of US goods and a further C$125bn list on deck. Ontario threatens a surcharge on electricity exports to three US states. The integrated auto corridor — where parts cross the border repeatedly — becomes a tariff multiplier.
Two days after imposition, a large share of Mexican and Canadian trade is exempted. The first instance of a pattern that defines the whole programme: announce broad, retreat to narrow, keep the headline.
The EU readies €26bn of countermeasures on bourbon, motorcycles and jeans. Canada adds C$29.8bn. Q1 US GDP prints −0.5% as importers front-run the tariffs with record inventory builds.
A novel construct: tariffs on third countries for their commercial relations with a fourth, applied at the Secretary of State's discretion with no published list.
Covers passenger vehicles and light trucks, with parts to follow. The Budget Lab models average vehicle prices rising 13.5%, about $6,400 on a new car.
A trade-deficit national emergency underwrites a universal 10% tariff plus individualised rates up to 50%. Executive Order 14256 simultaneously ends the $800 de minimis exemption for China and Hong Kong. In a single afternoon the average US tariff rate is set to rise more than tenfold.
25% on finished vehicles. Announced offsets and stacking relief follow within four weeks after US automakers warn the measure hurts them more than their foreign competitors.
The floor applies to nearly every origin. Canada and Mexico are excluded — they are already covered by the fentanyl orders.
Equities, the dollar and Treasuries sell off together, a combination that signals foreign capital leaving rather than a flight to safety. The pause is announced mid-session; the S&P 500 gains 9.5%, its best day since 2008. China alone is escalated — to 125% reciprocal, 145% all-in. The implied average tariff rate touches roughly 28%, the highest since the 1930s.
A quiet Friday-night CBP notice removes a very large block of consumer electronics from the reciprocal tariffs — the single biggest carve-out of the programme, and an implicit concession that the China rate was unpayable.
Beijing states it will ignore further US increases, and shifts to non-tariff instruments: rare-earth export licensing, entity listings and regulatory probes. The pivot from tariffs to chokepoints is the strategically decisive move of the year.
Both are procedurally slower routes that require investigation and findings — and both become the backbone of the 2026 tariff structure.
Aimed at Chinese-owned plants in Cambodia, Malaysia, Thailand and Vietnam. Cambodia's rate reflects non-cooperation with the investigation rather than an economic finding.
The exemption preserves the North American supply chain but sharply disadvantages European and Asian parts makers.
Britain holds the 10% floor, keeps a 25% steel rate when the world goes to 50%, and secures a 100,000-vehicle quota. The template for what a deal means here: relief from the worst case, not a return to the status quo.
A 90-day standdown reached after the first 145%-tariffed cargoes arrive at Los Angeles with volumes halved and retailers warning of visible shortages. The escalation ceiling has been found.
A unanimous three-judge panel in V.O.S. Selections holds that IEEPA does not authorise these tariffs and that the declared emergencies bear no rational relation to the measures. Stayed on appeal — the tariffs keep running, but every dollar collected from here is contingent.
The UK alone stays at 25% under its May deal. Downstream fabricators, whose employment base is many times that of the mills, absorb the input cost.
Refrigerators, washing machines and dishwashers become derivative steel products. The derivative mechanism lets Commerce widen coverage without a new investigation — an expansion valve used repeatedly thereafter.
Congress legislates the $800 exemption out of existence with a two-year runway. The administration pre-empts its own statute six weeks later.
Rates for some two dozen partners are communicated by form letter posted to social media, with deadlines pushed to 1 August. Trade policy delivered as correspondence.
The rate is inclusive of MFN duties rather than additive — goods already dutiable above 15% pay no reciprocal tariff at all. That structural detail is worth more than the headline number.
Brussels shelves its €93bn retaliation list and pledges energy purchases and investment. European industry gets predictability at a permanently higher tariff floor than existed in 2024.
A 40% emergency tariff stacked on the 10% floor, expressly tied to a foreign criminal trial. The US runs a goods surplus with Brazil — the clearest case in the programme of tariffs as political instrument.
Cathode, ores, concentrates and scrap are excluded, sparing most Chilean supply. COMEX copper, which had run to a record premium on the announcement, gives up roughly a fifth of its value in a day.
An extra 25% penalty on top of the 25% reciprocal rate, effective 27 August. Textiles, gems and shrimp bear the brunt; pharmaceuticals are spared. Trade policy annexed to sanctions policy.
After two postponements, rates land for more than 60 partners. Syria 41%, Laos and Myanmar 40%, Switzerland 39%. A 40% penalty applies to goods found transshipped to disguise origin. The statutory average rate reaches 18.6% — the highest since 1933.
Construction materials, furniture and machinery. Duty applies to the metal content by value, requiring importers to decompose their own bills of materials for customs.
The appellate court upholds the CIT on the same day the de minimis exemption closes globally by executive order. Two years of revenue are now openly contingent on the Supreme Court.
Announced for 1 October, with an exemption for any firm breaking ground on US plants — which most large manufacturers already were. The tariff functions more as an investment-announcement mechanism than a revenue measure.
Proclamation 10976, signed 29 September and published 6 October, sets 10% on softwood timber and lumber and 25% on upholstered wooden products, kitchen cabinets and vanities, rising to 30% and 50% respectively from 1 January 2027 absent a negotiated exception. Falls hardest on Canadian producers already carrying anti-dumping duties, and on a US housing market where affordability is the binding constraint.
Licensing is claimed over any product anywhere containing Chinese-origin rare earths above a threshold — a mirror of US semiconductor rules. China refines over 80% of global supply. Washington threatens an additional 100% tariff from 1 November.
The fentanyl tariff halves to 10%, the 24% reciprocal tranche stays suspended to 10 November 2026, port fees are suspended both ways, and China pauses the October export controls for a year and resumes soybean purchases. The all-in China rate falls from about 57% to 47%.
The bench presses hard on whether the power to regulate importation includes the power to tax, and on the major questions doctrine. The market reads the argument as unfavourable to the government.
Products the US cannot grow at scale are exempted, after grocery inflation becomes the dominant political story. An admission that a tariff on a good with no domestic substitute is simply a consumption tax.
In exchange for higher NHS spending on new medicines — the first significant increase in two decades. Tariff relief traded for foreign health-budget commitments.
Signed on 14 January, published at 91 FR 2443 on 20 January, and applying to goods entered on or after 15 January. Three dates for one action, one day apart at the narrowest — the smallest instance in this report of the distinction the timeline now tracks explicitly. It follows a nine-month §232 investigation whose report reached the President on 22 December 2025. Deliberately narrow: high-end accelerators including the NVIDIA H200 and AMD MI325X, with exemptions for data-centre buildout, repairs, R&D, startups, non-data-centre consumer and civil industrial uses, and the public sector. The proclamation notes the US fully manufactures only about 10% of the chips it needs. The White House called it a phase-one action.
Announced in a social-media post against Denmark, Norway, Sweden, France, Germany, the United Kingdom, the Netherlands and Finland, conditioned on the sale of Greenland. The eight issued a joint statement of solidarity; the EU paused ratification of the July 2025 trade framework. No executive order, proclamation or Federal Register notice was ever issued and no legal authority was ever cited — the entry is filed under IEEPA only because that was the vehicle then in use. Withdrawn on 21 January after the President and the NATO Secretary-General announced a framework at Davos. Four days from threat to retraction, and the cleanest case in the programme of an announcement that moved diplomacy without ever becoming an instrument. The further reading that the threat channel had by 2026 stopped moving markets is the author's judgement, not a sourced finding.
Announced by social-media post after a call between the two leaders, and confirmed by Prime Minister Modi in his own post. Two separate duties are involved: the 25% reciprocal rate set by the executive order of 31 July 2025, said to fall to 18%, and the additional 25% Russian-oil penalty of 6 August 2025, to be removed outright. India commits to winding down Russian crude outside existing contracts and to buying US energy. No instrument issued that day, and the White House had published no formal statement. Only one half of the announcement ever became law.
The order terminates the additional 25% duty imposed by EO 14329 for goods entered on or after 12:01 a.m. on 7 February, strikes HTSUS headings 9903.01.84 through 9903.01.89, and directs CBP to refund duties already collected; CBP implemented it on 9 February. The same day, a US–India joint statement formally confirms intent to apply an 18% reciprocal rate under EO 14257 — a step up from the 2 February social-media post, but still a statement of intent rather than an amending instrument. No separate proclamation or executive order implementing the 18% figure has been located, and contemporaneous reporting notes CBP guidance kept the prior 25% reciprocal rate in effect in the interim. The Supreme Court voided the entire IEEPA schedule fourteen days later, before the gap was ever closed. The largest negotiated de-escalation of the programme was half signed, half stated, and then moot.
H.J.Res. 72 passes 219–211 on Roll Call 65, terminating the national emergency declared in EO 14193 that underpins the Canada tariffs. Bacon, Fitzpatrick, Hurd, Kiley, Massie and Newhouse cross the aisle; one Democrat votes against. The procedural bar on privileged resolutions had lapsed at the end of January and an attempt to extend it to July failed the previous night. The margin is short of veto-proof and a veto is promised. Congress begins to reclaim the question nine days before the Court answers it.
Roberts writes that the Framers gave the tariff power to Congress alone, and that when Congress delegates it, it does so clearly and with constraints — it did neither here. The power to regulate is not the power to tax. Kagan, Sotomayor and Jackson concur on plain text without reaching major questions; Thomas, Alito and Kavanaugh dissent. Every reciprocal and fentanyl tariff is void from inception. About $166bn was collected under IEEPA, owed back to more than 330,000 businesses, and CBO expects most of it to be refunded. Earlier estimates of $142bn and $150bn were made while collections data were still incomplete. The opinion is silent on how refunds are to be made.
A balance-of-payments statute never used since its 1974 enactment, capped at 15% and 150 days without congressional extension. Cited: a $1.2tn goods deficit, the first negative primary income balance on record, and a net international investment position of −90% of GDP. Executive Order 14389, signed the same day, ends collection of the IEEPA duties. The surcharge runs from 12:01 on 24 February to 24 July, with a USMCA carve-out and product exceptions closely tracking the IEEPA lists — but a single flat rate for every country, and none of the negotiated country schedules. Between the Court's opinion and its replacement instrument lies one afternoon.
A Truth Social post the day after the Supreme Court ruling raises the surcharge to the statutory maximum: “raising the 10% Worldwide Tariff... to the fully allowed, and legally tested, 15% level.” Proclamation 11012 was never amended, and CBP's implementing guidance of 23 February instructs the trade to report 10% under HTSUS 9903.03.01 from 24 February. Where the announcement and the instrument disagree, this report follows the instrument — which is the whole reason the distinction is tracked. The gap is now visible to every importer.
$1.6tn in lost primary revenue plus $0.4tn in debt service — and that excludes refunds. Twenty-four states sue over §122 the same day.
Targets structural excess capacity and production in manufacturing across China, the EU, Singapore, Switzerland, Norway, Indonesia, Malaysia, Cambodia, Thailand, Korea, Vietnam, Taiwan, Bangladesh, Mexico, Japan and India. Hearings opened 5 May 2026. Still undetermined.
A separate action filed the following day, alleging failure to impose or effectively enforce prohibitions on imports made with forced labour. Unlike §122, §301 tariffs have no rate cap and no expiry. This is the one that becomes the replacement regime.
50% on articles wholly of steel, aluminium or copper; 25% on derivatives; 15% on metal-intensive industrial and grid equipment through 2027; 10% where foreign-made goods use US metal; nothing below 15% metal content. Coverage widens while the marginal rate on complex goods falls.
Processing begins on repayment of unlawfully collected IEEPA duties plus interest. About $166bn was collected under IEEPA; CBO expects most of it to be refunded.
A divided CIT panel finds current conditions are not the large and serious balance-of-payments deficits the statute requires. Relief is limited to the named plaintiffs, so everyone else keeps paying. The government appeals; the Federal Circuit grants a temporary stay on 12 May.
May 2026 sees about $22bn collected against about $22bn refunded — a net outflow of $42m, the first negative month. June is far larger: $23.6bn collected against $49.2bn refunded, a net outflow of $25.6bn.
Proposed remedy: 10% or 12.5% by origin, covering partners that account for 99.4% of US imports. A near-universal tariff assembled from sixty separate country findings.
Three proclamations signed the same day and published on 23 July, each now linked directly: 11046 on alcoholic beverages, 11047 on dairy, and 11048 on motor vehicles. Each imposes 50%, the statutory maximum. The sector names explain the finding, not the coverage — every annex runs far wider than its title. Hockey equipment and assorted paper and wood sit in the alcohol annex; the motor-vehicle annex, the broadest of the three, reaches from cement and plywood to clothing, tableware and floor coverings, and pointedly excludes actual motor vehicles, which stay under §232. The cited discrimination: Canada's dairy TRQ allocation favouring CETA (EU) access over USMCA access, provincial liquor-board practices, and — for the motor-vehicle proclamation — Canada's own retaliatory Surtax Order on U.S. vehicles, which cut US motor-vehicle exports to Canada by roughly 22% over the following year. Section 338 requires collection to begin thirty days after proclamation, which is how 20 July produced a 19 August date with no slack in it, overrides USMCA preference, and has no expiry.
Rebuilt on a durable statute, and the contrast with what it replaced is the whole point. The IEEPA measure was a 40% duty declared by executive order in a single afternoon. This one ran the full statutory course: initiation on 15 July 2025 (90 FR 34069), an actionability determination and proposed remedy on 1 June 2026 (91 FR 33854), a comment period, a public hearing on 6 July, and a final Notice of Action on 20 July taking effect at 12:01 on 22 July. Twelve months, three Federal Register notices, and a record. Exemptions cover beef, orange juice, aircraft and parts, energy and pharmaceuticals; a limited in-transit carve-out ran to 29 July.
USTR's Notice of Action in the 60-economy forced-labour investigation takes effect the same minute §122 lapses. Ten percent for 17 partners including Canada, Mexico, India and the UK; 12.5% for the remaining 38 including China, Brazil, Vietnam and Russia; the EU and Taiwan capped at 10% net of MFN, Japan and Switzerland at 12.5% net of MFN. Exemptions for coffee, beef, orange juice, aircraft, energy and pharmaceutical inputs. For most importers the landed cost barely moves — the statute underneath it has changed completely. MFN, §232 and §301 duties begin applying to postal shipments the same day.
About $100bn of the $166bn in IEEPA duties has been refunded. July saw $33.4bn refunded against $24.8bn collected on Treasury figures, which CBO rounds to $36bn against $26bn. The full-year customs shortfall against the February baseline is around $250bn, even crediting the replacement §301 tariffs.
A proclamation amends Annex II of all three underlying instruments, deleting 19 August and inserting 22 August, on advice that Canada had expressed a commitment to remove the measures at issue. Because 19 August was the statutory floor rather than a discretionary schedule, the extra window had to be created by a fresh presidential act rather than by simply waiting. A last opening for a broader deal, granted on the eve of the effective date.
The Section 338 duties take effect on roughly US$20bn of Canadian goods, about 5% of US imports from Canada, applying regardless of USMCA origin. The same day, in a nationally broadcast address, the Prime Minister announces he has suspended trade negotiations and recalled Canada’s negotiating team, and states plainly that Canada will match the new US tariffs dollar for dollar, with details to follow. The Department of Finance confirms those details three days later: new countermeasures on C$27.6bn of US goods, concentrated in steel and aluminum, dairy, appliances, agricultural equipment, pulp and paper, plastics and electronics, with existing steel and aluminum surtaxes rising from 25% to 50% to match. These take effect 8 September 2026 — not yet in force as of this report. Five days old as of this report, and the most active front in the trade war.
Three tariff rates, not one
These measures are routinely conflated and they answer different questions. Mixing them is the most common error in tariff commentary, so they are defined and plotted separately here.
Duty implied by policy on the books, holding 2024 import shares fixed. The right measure of the burden on a buyer who does not change behaviour.
Duties actually collected divided by total goods imports. Lower because of exemptions, USMCA preference, Chapter 98 provisions and re-sourcing.
The statutory rate after buyers shift origin and product mix. The right measure for revenue and macro modelling, not for consumer welfare.
A single monthly point hides the largest discontinuity of the period. Resolved to event dates, the statutory rate fell almost eight points in one morning and was largely rebuilt within four days on a different statute. The immediate pre- and post-SCOTUS break uses the 20 February 2026 Budget Lab vintage; the implemented §122 regime uses the first later Budget Lab estimate that reflects the 10% surcharge actually collected. Each card names its vintage.
With the reciprocal and fentanyl tariffs vacated and nothing yet put in their place, the pre-substitution rate falls to 9.1% — about one third of the 2025 peak, and still the highest since 1946 excluding 2025. Post-substitution, 8.0%. TBL's model assumed no replacement; the administration proved that assumption wrong within four days.
Gross, refunded, net
The revenue story cannot be told with a single net line. Gross collections held up through 2026 — it is the refunds ordered by the Supreme Court that turned customs duties into an outflow.
May 2026 was the first negative month, though only barely: about $22bn collected against about $22bn refunded, a net outflow of $42m. June was the real break — $23.6bn collected against $49.2bn refunded, a net outflow of $25.6bn — the largest monthly reversal in the period reviewed here. July ran $24.8bn against $33.4bn on Treasury's figures, which CBO rounds to $26bn against $36bn. Roughly $100bn of the $166bn owed has now been returned, and CBO puts the projected full-year customs shortfall at about $250bn against its own February baseline — even after crediting the replacement §301 tariffs.
Who was hit
Peak IEEPA-era general rate compared with today's broad forced-labor §301 baseline. Other §232, §301, §338 and AD/CVD duties may stack on top and are excluded from the green bar.
Apr 2025 peak; 47% all-in under the Busan truce. Legacy §301 lists (7.5–100% by product) were never at issue in the ruling and survive untouched. The 12.5% general rate shown here is the forced-labor action, not the truce rate.
Diversion, or reshoring?
A falling bilateral deficit with China is the programme's headline result. Whether it represents production moving to the United States or suppliers moving to third countries is a different question, and the two have opposite policy implications.
Down by roughly half. Imports from China fell 29% across 2025 and a further 35% year on year in January 2026.
Taiwan overtook China in Q1 2026, driven by AI servers and advanced semiconductors rather than by any tariff-induced reshoring.
The fastest-growing top-ten supplier — textiles, electronics assembly, footwear and furniture, much of it Chinese-owned capacity relocated rather than created.
The headline success of the programme. But US exports to China fell 35.2% over the same period, so the gap narrowed partly by shrinking trade in both directions.
Investment can lead employment by months or years, so this is not a decisive test on its own. But if substantial reshoring had already translated into production at scale, stronger signals would normally be expected by now, and they are not evident.
The leading indicator for future factory jobs turned down and kept falling. By mid-2026 manufacturing had lost its place as the largest non-residential construction market, overtaken by power.
The manufacturing-specific index, not total industrial production. It stood at 98.4 (2017=100) in July 2026, up 1.2% on a year earlier — and still below its 2017 annual-average level after nine years. Manufacturing output grew 0.2% in July after 0.3% in June.
July 2026, up 0.1pp on the month and 2.2 percentage points below its 1972–2025 average of 78.2%. Installed manufacturing capacity is not being fully used.
Across 2025, with declines in every month after Liberation Day. Within manufacturing, primary metals added jobs while machinery, computers and transportation equipment posted the steepest losses.
Year to date through June 2026 against the same period of 2025, on the Census Value of Construction Put in Place. At the June annualised rate, power had overtaken manufacturing as the largest non-residential construction category — the first time since 2021. Nominal spending, not inflation-adjusted.
Earlier readings from the same series: down 9.6% year on year at October 2025, and at its lowest level in nearly three years in December 2025.
The four indicators do not point the same way, and that is the finding. Manufacturing production is modestly higher — up 1.2% year on year in July 2026, though the index at 98.4 remains below its 2017 annual-average level — while employment and factory construction have both fallen. Aggregate manufacturing capacity does not appear to be the binding constraint: utilisation sits 2.2 percentage points below its long-run average. That is consistent with weak demand, or with capacity being mismatched across industries, rather than with a generalised shortage of factory capacity. Output can also rise on productivity and mix without the plant-building and hiring that reshoring would require — and the composition supports that reading, with business-equipment production up 6.6% year on year while consumer-goods production fell 1.8%. The honest verdict is not that nothing happened; it is that no broad resurgence is evident.
The Budget Lab's own model is the fairest statement of the case for the policy, and it is a qualified one: in the long run US manufacturing output expands by 1.2% under the post-SCOTUS tariff set — but that expansion is more than crowded out elsewhere, with construction output contracting 2.4% and agriculture more than 1%, leaving the economy persistently 0.1% smaller. Sectoral reallocation, not net growth.
The observed data are meanwhile consistent with substitution rather than reshoring. Direct imports from China fell by roughly half while Vietnam, Taiwan and Mexico absorbed the volume, and neither US manufacturing employment nor manufacturing construction rose. CSIS notes that much of the import decline is more apparent than real, since China expanded trade and investment links through South-East Asia. Federal Reserve research from 2024 had already found that the largest US suppliers were themselves increasing their reliance on Chinese inputs — meaning measured exposure can fall while actual exposure does not. None of this is a causal estimate; it is a set of co-movements that the reshoring hypothesis does not fit well.
The currency and the base
Two questions earlier editions could not answer. A tariff shock is only as large as the currency lets it be, and only as lucrative as the import base it lands on.
Standard theory says a tariff should strengthen the imposing country's currency, which would claw back part of the price increase. Many forecasters expected exactly that in 2025. It did not happen. The broad dollar index reached its record high of about 130.0 in January 2025 and weakened materially through the remainder of the year, falling roughly 8% on a trade-weighted basis and about 10% against the majors. It has recovered only marginally since.
So dollar-denominated import prices rose on two counts at once: the duty, and a weakening currency. This is consistent with the low foreign-incidence estimates rather than in tension with them. One qualification matters: it does not undercut the New York Fed result. That specification uses country-and-date fixed effects designed to separate tariff-specific price movements from general country-level ones, so a broad currency move is already absorbed. The exchange rate is context for the aggregate burden, not a correction to the incidence estimate.
Real goods imports in chained 2017 dollars, indexed to January 2025. Goods only — services are not dutiable and have no place in a tariff-base measure. Every month is a published observation from a single FT-900 vintage; nothing here is interpolated.
Real goods imports rose about 5% above their January 2025 level by March as buyers front-ran the announced tariffs, then fell to 79% of that level by October — a swing of twenty-six index points inside eight months. Eighteen months on they remain about 12% below where they started.
The pattern is consistent with tariffs reducing import demand, but this chart alone does not identify the tariff's causal contribution. The reversal of front-running, inventory normalisation, domestic demand, product mix, the exchange rate and global growth all moved in the same window. What can be said without qualification is that the taxable base was materially smaller through the tariff period than before it — which matters for the revenue arithmetic, since duty collected is rate times base. It also complicates the deficit story: a narrower goods gap that comes from importing less in real terms is not the same achievement as exporting more.
January 2025 against June 2026, real, same vintage. The aggregate hides a composition shift: capital goods imports rose 27% on the AI build-out while consumer goods fell 25% and industrial supplies nearly halved. This is where the contraction landed, not where tariff liability landed — establishing the latter needs HTS-level duty data, which this report does not have.
Who paid, and what it cost
Incidence is two separate questions asked at two separate points. Conflating them is how the claim that foreign exporters bear tariffs survives contact with the data.
Who bears the duty at the point of entry — does the foreign exporter cut its price, or does the US importer pay the full tariff on top of an unchanged price? This is a question about terms of trade.
Amiti, Flanagan, Heise and Weinstein at the New York Fed measure this directly on import data through November 2025: US firms and consumers bore 94% of the burden over January–August, and 86% by November, as foreign exporters began absorbing slightly more. In November a 10% tariff was associated with a 1.4% fall in foreign export prices — 86% pass-through to US import prices. NBER and the Kiel Institute reach comparable domestic shares of 94% and 96%. The range shown is the spread across the period and across those studies, not a confidence interval.
Given that the cost landed domestically, where inside the US economy did it settle — in company margins, or in retail prices? This is a question about pass-through, and it is separate from Stage 1.
The striking figure is the 51% absorbed in domestic business margins rather than passed through. Margin absorption is one important mechanism helping explain why consumer pass-through was smaller than the headline rate — alongside exemptions, origin substitution, inventory timing and incomplete pass-through generally. On this evidence a substantial share of the cost did not vanish so much as move from the till to the income statement.
A one-time shift in the level of prices, reached over time as tariffs pass through. It is not an additional 0.7% of inflation every year — once the level has adjusted, the contribution to the annual rate falls back toward zero.
A recurring annual purchasing-power cost for the average household while the tariffs remain in place. Related to the figure on the left but not the same concept: this is a flow, that is a level.
The two views point in opposite directions and both are correct. In dollars the top decile pays about four times the bottom, because it consumes more. As a share of post-tax income the bottom decile pays nearly three times the top — 1.1% against 0.4% — because it spends a far larger fraction of what it earns. A tariff is a consumption tax, and consumption taxes are regressive by construction.
Only these two points are published. The eight intermediate deciles are deliberately not shown: consumption shares are not linear in income, so interpolating between the endpoints would manufacture a smooth curve the source does not support.
Reduced sharply once IEEPA fell; apparel is no longer a top-affected category unless §122-style breadth returns.
Same. The 2026 regime is concentrated in industrial goods rather than clothing.
Still among the most affected. §232 autos survived the ruling untouched.
The single most affected category in 2026 — 50% headline duty on metal-intensive goods.
Second most affected in 2026, largely through metal content and derivative coverage.
Among the few manufacturing subsectors to add jobs in 2025. This is the intended effect of the §232 metals tariffs.
Employment also rose, though Cato attributes much of this to falling interest rates and demand from other industries rather than to the tariffs themselves.
Among the steepest job losses of 2025. The most tariff-cost-exposed manufacturing subsector in Equitable Growth's input-cost analysis.
Steep losses. Import share of inputs is 16.3%, and derivative metal coverage widened repeatedly through 2025.
Steep losses despite 25% protection on finished vehicles — input costs and retaliation ran against the tariff's intended direction.
This is the structural asymmetry of metal tariffs and it is not new. Downstream users employ far more people than the mills they buy from — in aluminium alone, downstream fabrication accounts for roughly 78% of employment across the core categories, and up to 97% on a broader definition. A policy that raises input costs to protect the upstream 22% is trading a large employment base for a small one.
The March 2025 package (25% on C$29.8bn, plus C$14bn more from the steel/aluminum countermeasures) was removed 1 September 2025 as Canada matched US CUSMA exemptions, keeping only steel, aluminum and auto surtaxes. A separate, new measure follows the August 2026 §338 tariffs: Ottawa's dollar-for-dollar response on C$27.6bn of US goods, with steel and aluminum surtaxes rising from 25% to 50%. Scheduled for 12:01 a.m. 8 September 2026 — not yet in force as of this report.
Retaliatory tariffs and the October 2025 rare-earth controls are suspended to late 2026. Soybean purchases resumed. The leverage was export controls, not tariffs.
Targeted bourbon, motorcycles, jeans and farm produce from Republican-leaning states. Shelved under the July 2025 framework and repeatedly extended — suspended by choice, not lapsed by law.
Several partners opened WTO cases in May 2025, but the Appellate Body has been non-functional since 2019, so no binding ruling is available.
The balance sheet
What did the United States economically receive in exchange? Both columns, with a confidence rating on each line and no forced net figure.
Real money, and the clearest success of the programme while it lasted.
Current-law estimate under the rebuilt §232/§301 structure.
Concentrated and visible, but small relative to the downstream base.
Market access, investment pledges and purchase commitments from the UK, EU, Japan, Korea and ASEAN partners. Their legal footing was undermined when the underlying tariff was voided.
Genuine de-risking on the direct measure. Indirect exposure through third countries is largely unmeasured.
To more than 330,000 businesses, plus interest.
About 60% below plan, even after crediting the replacement §301 tariffs.
Equivalent to roughly $1,100 per household per year at current-law policy.
A regressive tax by construction, since lower-income households consume a larger share of income.
Far larger employment base than the protected subsectors.
Modelled effect of the 2025 package; the 2026 regime is smaller, so this overstates the current drag.
Canada C$30bn, the EU €21bn activated of a €93bn list, China's soybean halt. Exposure concentrated in farm states.
The two columns are not denominated in the same units and are not measured to the same standard. Revenue and refunds are accounting facts. Price effects and GDP drag are model outputs that depend on assumptions about Federal Reserve behaviour and substitution elasticities. Employment and investment movements are observed co-movements during a period that also contained a rate cycle, an AI capital-expenditure boom and a benchmark revision to the payroll data. Forcing a single net figure would imply a precision none of these components has. The honest summary is narrower: the revenue was real while it lasted and roughly $100bn of it has now been returned; the consumer price effect was smaller than most 2025 forecasts because firms absorbed it in margin; and while some output indicators improved, the employment, construction and capacity-use data do not show a broad-based reshoring resurgence so far.
Measured against its own goals
Assessed on the objectives the administration stated. Select any line for the underlying evidence and its limits.
Customs duties reached $195bn in FY2025, up 150% on FY2024, and $264bn across calendar 2025. The Supreme Court then made the $166bn collected under IEEPA refundable, of which about $100bn has been returned. Net customs collections turned negative in May 2026 and stayed negative through July. CBO now puts FY2026 customs at about $250bn — some 60% — below its February projection, and projects a full-year FY2026 deficit of $2.1tn against the $1.9tn it forecast in February. What survives is still material: the Budget Lab scores current-law tariffs at roughly $1.9tn over ten years.
Manufacturing employment fell in every month after Liberation Day and ended 2025 down 68,000, with primary metals adding jobs while machinery, computers and transportation equipment posted the steepest losses. Manufacturing production is modestly higher — up 1.2% year on year in July 2026, on an index still below its 2017 annual-average level — but factory construction kept falling, down 22.1% year to date through June 2026, and manufacturing capacity utilisation remains 2.2 percentage points below its long-run average. Aggregate installed capacity therefore does not appear to be the binding constraint; weak demand or sectoral capacity mismatch may help explain the gap. These are co-movements during the tariff period, not identified causal estimates — the same window contained a rate cycle and an AI capital-expenditure boom. The Budget Lab separately projects that the remaining post-SCOTUS tariffs reduce payroll employment by about 550,000 and raise unemployment 0.3pp by the end of 2026.
The goods deficit set a record in 2025 and the total deficit narrowed by just $2.1bn against 2024, entirely through services — the goods gap widened by $25.5bn. The bilateral gap with China did fall sharply, from $297bn to $203bn and a further 43% in early 2026, but US exports to China fell 35.2% over the same period. The Tax Foundation's structural point stands: the trade balance reflects the gap between national saving and investment, which tariffs do not directly change.
At the border, the New York Fed puts the foreign-exporter share at 6–14% for the first eleven months of 2025; NBER and Kiel put the domestic share at 94% and 96%. Domestically, Goldman finds 51% absorbed in US business margins and 37% passed to consumers. On the available evidence a tariff is overwhelmingly a tax on the importing country, which is what the standard model predicts.
The UK, EU, Japan, Korea, Vietnam, Indonesia, Malaysia and others signed frameworks conceding market access, investment pledges and purchase commitments in exchange for capped rates. Two qualifications. The caps were relief from threatened rates, not reductions from 2024 levels. And when the Supreme Court voided the tariff that bought them, every agreement was left resting on a rate that no longer legally existed — USTR asserted the deals still bind, but several partners publicly questioned which rate now applied.
The central legal claim failed. IEEPA is unavailable for tariffs. Section 122 is capped at 15% and 150 days and was itself held unlawful at first instance. What remains — §232, §301, §338 — is more procedurally constrained than IEEPA, but the authorities differ: §232 and §301 generally require investigative records, findings, hearings and comment periods, while §338 operates through presidential findings with a statutory 30-day effective-date floor rather than a notice-and-comment process. USTR needed four and a half months to convert the forced-labour §301 investigation into collected duties. The speed that defined 2025 is structurally unavailable.
The forward calendar
Separated into what is in force, what is scheduled, what is merely under investigation, and what is a scenario. These are not the same thing and should not be modelled as though they were.
Preferential rates for firms with onshoring agreements; full exemption for those also entering pricing agreements. Seventeen manufacturers are already named as qualifying.
The suspended 24% reciprocal tranche was an IEEPA measure and died with the ruling — but China's rare-earth control pause, the soybean commitments and the port-fee suspension all run to this date and must be renewed.
The larger of the two March investigations, and the main upside risk beyond the current-law path. Uncapped, open-ended §301 tariffs on China, the EU, Japan, Korea, Mexico, India and others would follow an affirmative finding. No determination has been issued, so this is not in any current-law projection.
Possibly en banc, with a further Supreme Court appeal available. The surcharge has already expired, so the stakes are refunds and precedent rather than prospective collection.
Plus large batteries, cast iron fittings, plastic piping, industrial chemicals and grid equipment reported as under consideration.
The Budget Lab describes this as current law including several scheduled increases between now and December — not as a conditional forecast. Labelled here exactly as the source labels it. Any affirmative overcapacity determination would sit on top of this path, not inside it.
The rate has found a floor, not a ceiling. Eleven percent is roughly four times the 2024 level, and most of what remains is now carried by authorities without an automatic near-term expiry. The volatility of 2025 came from an instrument that could be deployed overnight; §232 and §301 cannot be, which cuts both ways — slower to impose, and considerably harder to unwind.
The constraint has moved from politics to procedure, unevenly. Sections 232 and 301 generally require an investigation, a finding and a comment period — USTR needed four and a half months to convert the forced-labour investigation into collected duties. Section 338 is different: it operates through presidential findings under its own statutory framework, with a 30-day effective-date floor rather than a notice-and-comment process. Anyone modelling future actions should watch Federal Register initiations for §232/§301, and presidential proclamations directly for §338.
Canada is the live risk. Section 338 has no time limit, overrides USMCA, has never been tested in court, and took effect four days ago into a negotiation that has already collapsed. Ottawa has committed to matching dollar for dollar. If the overcapacity determination lands on the EU and Japan while that is running, the second half of 2026 will look less like a de-escalation than a re-founding.
The fiscal hole is the underrated story. CBO projects a full-year FY2026 deficit of $2.1tn, about $200bn above its February forecast, with customs collections running roughly 60% below plan. Through the first ten months the deficit stood at $1.8tn, $169bn wider than the same period a year earlier. These are larger, more certain and more immediate than any of the trade effects in this report — and they run directly against the case that tariffs would pay down debt.
Sources and method
Claim-level tags throughout the report link here. Open any tag to reach the underlying document.
Rate definitions. Three measures are used and never mixed: statutory pre-substitution, realized effective duty, and post-substitution modelled. Each chart states which it is showing and what the denominator is.
Observed versus interpolated. Filled markers on line charts are published figures. Unmarked points are interpolated between anchors and carry no independent authority. Where a component of a decomposition is the author's estimate rather than a published figure, the chart's basis note says so.
Causal language. Employment, investment and bankruptcy movements are reported as co-movements during the tariff period, not as causal effects, unless the cited study was designed to identify causation. The period also contained a rate cycle, an AI capital-expenditure boom and a payroll benchmark revision.
Policy status. Status follows the underlying legal position and the source's own definition. Current-law scheduled increases are labelled Scheduled; unresolved investigations are labelled Pending; assumptions introduced by this report rather than by a source are labelled Scenario.
Timeline dates. Each entry is dated to the event named in its own headline — a proclamation signed, a deal announced, an investigation opened, a tariff taking effect. Where the legal effective date differs from that action date, a second date marked eff. sits beneath the first and both appear in the entry text. This is not housekeeping. Announcement, signature, publication and effect are four different dates, and the space between them is where a good deal of this programme actually lived: a 15% surcharge that was announced but never collected, an 18% rate announced by post and later formally confirmed in a joint statement but never implemented through a tariff-amending instrument, a chip tariff signed on one day and owed on the next.
Source stage. A §301 or §232 action passes through initiation, determination and final action, and each stage is a different document proving a different thing. An initiation notice does not establish a tariff. For actions covered by the latest source-stage audit, where the report cites a tariff that is in force, the tag resolves to the final instrument; where it cites an investigation, it resolves to the initiation notice. Several older Section 232 entries are disclosed exceptions to this, named below. The Brazil entry names all three notices because the contrast with the executive order it replaced is the point of the entry. The major current-law tariff actions reviewed in the latest source-stage audit resolve to their operative instruments: Proclamation 11012 and the CBP guidance under it for §122, Proclamations 11046, 11047 and 11048 for §338, Executive Order 14384 for the Indian surcharge, Proclamation 11002 for chips, Proclamation 10976 for lumber, Proclamation 11020 for pharmaceuticals. Several older Section 232 timeline entries — the initial steel and aluminium actions, the metals-doubling and appliance-derivative expansions, and the copper action — still resolve to a dated chronology source rather than their own proclamations, and remain candidates for primary-document replacement in a future revision. Where the legal commentary that formerly stood in for those documents is still worth reading, it appears as a second tag beside the instrument rather than in place of it.
URL stability. Fixed-vintage statistical sources are pointed at dated archives where one exists — the FT-900 tag resolves to the BEA archive of the June 2026 release rather than the "current release" path. Two do not: the Census construction release and the Federal Reserve G.17 both resolve to "current" URLs that will move when the next month publishes. The vintage is named in each chart's metadata, so the figure remains identifiable even after the link advances.
Source tiers. Every tag carries the institution and year and links to a specific document. Most resolve to a named primary or agency document; a minority resolve to a dated chronology entry or an event-specific press report, and those carry a dashed border. No timeline entry resolves to a general tertiary summary, and the registry holds no key that nothing in the report cites. All 63 timeline records have undergone event-level source review across several revisions, and each fresh round of primary-document checking has continued to surface corrections — a signing date read as an effective date, commentary standing in for the instrument it described, a source dated after the event it was supporting. The claim made here is that every figure links to a document a reader can open and check, not that the file is closed. Tier reflects provenance, not credibility: legal commentary on a court ruling is tier 5 however good it is, because it is not the issuing court. Two load-bearing figures still rest on non-primary sources and are flagged as such: the Goldman Sachs incidence decomposition — a tertiary report of a research note that is not itself public — and the trade-share figures reported through secondary aggregators rather than pulled from Census tables directly.
Vintage discipline. Budget Lab has published under several vintages and revised both its model and its Section 122 assumption. The February break panel uses the 20 February 2026 report throughout and discloses the 16.0% correction issued the next day rather than blending the two. The decile panel uses the 8 April 2026 report for both endpoints. The headline rate, price level and household cost use the 24 August 2026 vintage. Where those vintages are not comparable, the chart says so.
Known gaps. The report still lacks an HTS-to-end-use tariff exposure mapping, a directly measured authority-level decomposition built from customs microdata, state-level retaliation exposure, and the original Goldman Sachs research note behind the domestic incidence split. The broad dollar index is the only remaining reconstructed series and is drawn dashed for that reason. Everything else in the report is a published observation, an explicitly labelled model output, a disclosed interpolation, or a disclosed author estimate.