$120.7bn More Collected, $2.1bn More Kept (September 2026): The Tariff Refund Ate 98% of the Customs Windfall

In the fiscal year through August, the United States collected $292.5bn of gross customs duties against $171.9bn a year earlier — a 70.2% increase. Over the same eleven months it refunded $125.2bn, against $6.7bn the year before. Net customs revenue came in at $167.3bn versus $165.2bn: an increase of 1.3%, which means 98.2% of the extra money collected went back out of the door inside the same fiscal year. The refund is also not over — the daily statements show $15.6bn of Customs and Border Protection withdrawals in the first two weeks of September alone, against a normal month of roughly $1.4bn.

What the August Treasury statement actually says

The Monthly Treasury Statement reports receipts by source in a format that separates three things most coverage collapses into one: gross receipts, refunds, and net receipts. For customs duties in the fiscal year to 31 August 2026, those three numbers are $292.533bn, $125.225bn and $167.308bn. For the same eleven months of fiscal 2025 they were $171.865bn, $6.674bn and $165.191bn.

Source: US Department of the Treasury, Monthly Treasury Statement, Table 4, August 2026.

Read the third row against the first. A 70% rise in gross collections produced a 1.3% rise in what the government kept. The arithmetic is not subtle: of the $120.7bn of additional gross duty collected, $118.6bn was returned through the refund line in the same period.

The reason is a court decision. On 20 February 2026, in Learning Resources, Inc. v. Trump, the Supreme Court held 6–3 that the International Emergency Economic Powers Act "does not authorize the President to impose tariffs." The duties collected under that statute had to come back. The Court of International Trade subsequently ordered CBP to return them, and CBP built a dedicated processing system — CAPE, the Consolidated Administration and Processing of Entries — which opened in April. CBP told a court in an August filing that roughly $100bn of a pool of about $166bn had been paid as of 31 July.

The three months the customs line ran backwards

Monthly data makes the shape of it clear. These are the net figures — gross collections minus refunds — for the fiscal year to date.

Source: Monthly Treasury Statement, Table 4. Figures rounded to $0.1bn.

Three consecutive months in which one of the federal government's larger revenue lines contributed nothing or less than nothing. June alone returned $49.2bn — more than the entire customs-duty take of a typical pre-2025 fiscal half-year.

Gross collections fell about a fifth — they did not disappear

One thing the net figure can hide is that tariffs are still being collected in size. The invalidated statute was replaced by duties resting on other authorities, so the flow continued at a lower rate rather than stopping. Customs deposits in the Daily Treasury Statement ran $31.7bn in September 2025, $30.4bn in December 2025, $28.6bn in February 2026 and $25.0bn in August 2026 — down roughly 21% from a year earlier, not down to nothing. The rate ladder that replaced the struck-down duties was covered in the Section 122 and Section 301 handover in July.

That matters for how you read the year. The ongoing tariff take is a real and continuing revenue stream of roughly $290bn a year gross. The refund is a one-off retrospective repayment of money collected under an authority a court removed. They are separate flows that happen to share a line in the accounts, and conflating them produces either the claim that tariffs raise no money or the claim that they raised $292bn — both wrong.

The refund did not end in August

The August statement's $10.5bn of refunds, down from $33.4bn in July, looks like a programme running out of road. The daily data says otherwise.

Through 14 September, withdrawals by CBP totalled $15.566bn for the month to date against customs deposits of $2.880bn, with $6.334bn going out on 14 September alone. For scale, the same withdrawal line ran $1.166bn in September 2025, $1.808bn in December 2025 and $1.368bn in February 2026. September is therefore already running at roughly eleven times a normal month's pace, and above the whole of August, with more than half the month still to go.

Two caveats worth stating plainly. The daily withdrawal line covers all CBP disbursements, not refunds alone, so a small part of it is ordinary agency spending — visible in that $1.2bn to $1.8bn baseline. And customs deposits are lumpy within a month because periodic monthly statement payments cluster in the middle of it, so a mid-month deposit figure understates the month.

Neither caveat changes the signal. An agency whose disbursements normally run under $2bn a month has paid out $15.6bn in fourteen days.

How a refund becomes an issuance problem

The mechanism is unglamorous and entirely about cash.

The federal accounts show the pressure the refund landed into. The fiscal-year deficit through August was $1.966tn, on receipts of $4.845tn and outlays of $6.811tn, against $1.775tn in the same eleven months a year earlier. Net interest for the period crossed a trillion dollars at $1.017tn, up from $933.5bn; gross interest on Treasury debt securities was $1.267tn against $1.124tn, an increase of $143.1bn or 12.7%. Total public debt outstanding stood at $40.054tn on 14 September.

Set against a $1.97tn deficit, $125bn is not the dominant term. But it is roughly 6% of the year's borrowing need, it was unplanned, and it arrived concentrated in three months rather than spread across twelve. On the Treasury's daily par yield curve, the 10-year closed at 5.00% on 15 September — the first close at or above 5% in 2026 — having bottomed at 3.97% on 27 February, the week after the Supreme Court ruling, and started the year at 4.19%. The curve on 15 September read 4.67% at two years, 5.00% at ten, 5.40% at twenty and 5.36% at thirty.

That sequence is not proof of anything on its own. A great deal happened between February and September, including an energy move and a repricing of the policy path, and the mechanics of the long-end move itself were examined in the August term-premium piece. The honest claim is narrower and still worth having: an unbudgeted six-percent addition to a year's financing need is exactly the kind of thing that shows up in the compensation investors demand for holding duration, and it happened during the seven months in which that compensation rose.

What this does and does not say about the dollar

The currency channel here is the interest-rate factor, and it cuts both ways depending on which leg of the yield you think is moving.

If a higher 10-year reflects a higher expected path of policy rates, the rate differential against the other majors widens and that has historically supported the dollar — the ordinary carry mechanism the USD page tracks. If instead the yield is rising because investors want more compensation for fiscal and supply risk at unchanged expected policy, the same yield rise can sit alongside a weaker currency, because the extra yield is a risk premium rather than a return premium. Those two stories produce identical 10-year prints and opposite currency implications, which is why a single number is never the read.

The refund is a small input to that question rather than the answer to it. What it does is remove an argument: anyone pointing at a 70% rise in gross customs collections as a fiscal improvement has to reconcile that with a net figure 1.3% higher than last year's. The improvement, on the government's own accounting, was about two billion dollars.

What would change the picture

Four things, in rough order of how soon they resolve.

The Federal Circuit appeal is the first. The Department of Justice is contesting the Court of International Trade's authority to order refunds for importers who did not themselves sue, and roughly $11.4bn of finally liquidated entries — about 6.9% of the pool — cannot be processed while it is pending. A government win shrinks the remaining outflow; a loss releases it.

The second is the September Monthly Treasury Statement, which closes fiscal 2026 and will show whether the daily data's September acceleration carried through the month. On the current fourteen-day pace the customs line could print negative again in the fiscal year's final month.

The third is the corporate side of the same money. The refunds are being paid to importers, not to the households that ultimately bore the duties in retail prices, and the earnings consequences of that were traced in the Q2 refund-quarter analysis and in Walmart's $2.9bn receipt. A refund is a transfer, and where it lands determines what it does to demand.

The fourth is policy. The Federal Open Market Committee's decision arrives at 2 p.m. Eastern on 16 September with the target range at 3.50–3.75%, into a market that has repriced sharply in recent weeks; the three central-bank decisions of this week were mapped in the rate-decision preview, and what a 5% 10-year does to a borrower rather than a trader in the mortgage-spread piece. None of that changes the refund arithmetic. It changes the cost of financing it.

For how this site reads fiscal and rate data across the eight majors, see what the meter measures.

Educational macro context only — not investment advice.