When Refunds Outrun Collections, the Same Customs Line Reads Two Ways

One line, two totals, and a 114 billion dollar gap between them

The Monthly Treasury Statement covering July 2026 carries customs duties twice, in two tables, at two very different values. Table 4, which splits receipts into gross collections, refunds, and net, reports fiscal-year-to-date gross customs receipts of 269,156,006,377.11 dollars. Table 3, which carries a single receipts column, reports the same fiscal-year-to-date customs line at 154,472,685,103.20 dollars. Neither figure is an error. They answer different questions. The distance between them is 114,683,321,273.91 dollars of refunds, which is 18.97 times the 6,044,842,995.86 dollars of refunds recorded across the same ten months of the prior fiscal year.

Which number a reader picks decides the headline. On the gross basis, customs duties rose 89.90 percent against the same ten months of the prior fiscal year, from 141,732,686,796.90 dollars — close enough to a doubling that the word gets used. On the net basis, the same ten months rose 13.84 percent, from 135,687,843,801.04 dollars. Same agency, same report, same month, same line item. One framing describes a structural break in federal revenue. The other describes an ordinary year.

This is not a revision story and not a data-quality story. Both values are published, current, and internally consistent: gross minus refunds equals net to the cent in the figures above. It is a definition story, and definition stories are the ones that survive longest without being caught, because every party to the argument can cite a real Treasury number.

Ten months of collections, month by month

The monthly detail is where the shape of the year becomes legible. All figures below are the published Table 4 values for the customs duties line, in billions of dollars, for each record date of fiscal year 2026 through July.

Month	Gross	Refunds	Net
Oct 2025	33.090	1.737	31.354
Nov 2025	31.733	0.977	30.756
Dec 2025	29.334	1.442	27.892
Jan 2026	29.545	1.803	27.742
Feb 2026	27.169	0.575	26.594
Mar 2026	24.017	1.862	22.155
Apr 2026	23.875	1.751	22.124
May 2026	21.930	21.972	-0.042
Jun 2026	23.628	49.183	-25.556
Jul 2026	24.835	33.381	-8.546
FYTD	269.156	114.683	154.473

Two facts fall out of the column of gross figures that the fiscal-year-to-date headline hides. First, gross collections peaked in the first month of the fiscal year and have declined since: July's 24.835 billion dollars is 24.95 percent below October's 33.090 billion. Second, July's gross month is 92.27 percent of the ten-month average of 26.916 billion dollars — that is, the most recent month is running below the average that produced the near-doubling.

The refunds column has a different shape entirely. Across October through April, refunds totalled 10.147 billion dollars, 5.10 percent of gross collections over those seven months. Across May through July they totalled 104.537 billion, which is 91.15 percent of all fiscal-year-to-date refunds concentrated in three months. Fiscal-year-to-date refunds are 42.61 percent of gross, against 4.26 percent in the prior year's comparable period. In level terms, refunds are 18.97 times the prior-year figure of 6,044,842,995.86 dollars.

Why a refund lands in a different month from the collection

The mechanical reason a monthly customs line can print negative is that collection and refund are separated in time by the customs entry process, and the statement records each in the month the cash moves. Duties are deposited as an estimate when goods enter. The final amount is fixed later, at liquidation, which eCFR, 19 CFR 159.1 defines as "the final computation or ascertainment of duties on entries for consumption or drawback entries." A refund can also be paid before liquidation: under eCFR, 19 CFR 173.4a, "whenever an importer of record declares or it is ascertained that excess duties, fees, charges, or exactions have been deposited or paid, the Center director may, prior to liquidation of an entry or reconciliation, take appropriate action to refund the deposit or payment of excess duties, fees, charges, or exactions." That authority runs to section 520(a)(4) of the Tariff Act of 1930, codified at 19 U.S.C. 1520(a)(4).

Small differences do not move at all. Under eCFR, 19 CFR 159.6, a difference of less than 20 dollars between duties assessed at liquidation and estimated duties deposited is disregarded and the entry is endorsed "as entered," with the regulation adding that in reliquidation cases a refund an importer requests will be refunded "even if less than $20" and that refunds resulting from court orders are processed regardless of amount. The netting rule in the same section applies the 20 dollar minimum after duties, fees, and internal revenue taxes are combined. The practical consequence is a pipeline with a long and uneven lag: a large population of entries can sit unliquidated for months and then resolve together.

What the statement does not do is explain the cause. The MTS tables carry an amount and a month; they do not carry a reason code. A refund concentration in May, June, and July 2026 is consistent with several mechanically different stories — a wave of liquidations reaching the same window, pre-liquidation refunds under the 1520(a)(4) authority, or refunds compelled by adjudication — and these tables cannot distinguish them. Saying which one applies would require entry-level or agency-level documentation that this dataset does not contain. That is the honest stopping point.

What run-rate arithmetic can and cannot support

With ten of twelve months reported, the temptation is to annualise. It can be done, but only as a set of scenarios with the disagreement between them stated, not as a forecast. On the gross basis, three defensible extensions of the remaining two months give: 318.826 billion dollars if August and September each repeat July's 24.835 billion; 322.987 billion if each runs at the ten-month average of 26.916 billion; and 316.084 billion if each runs at the May-to-July average of 23.464 billion. The spread across those three is about 6.9 billion, which is small — the gross series is stable enough that the choice of pace barely matters.

The net basis is the opposite. If August and September each repeat July's net of negative 8.546 billion, fiscal-year net customs receipts land near 137.380 billion — below the 154.473 billion already booked through July. If the last three months' average net of negative 11.381 billion repeats, the landing is near 131.710 billion. If refunds stop entirely and the last three months' gross pace continues, the landing is near 201.401 billion. The spread there is roughly 70 billion dollars, and none of the three is more probable than the others on the evidence in these tables. That is the real finding: gross customs receipts for fiscal 2026 are close to determined, and net customs receipts are not determined at all.

None of these are ceilings, and it is worth being explicit about why. A run-rate extension assumes the last observed month is representative of the next two. For gross collections that assumption is weak but bounded, because the monthly series has moved in a 21.930 to 33.090 billion range all year. For refunds it is not bounded in any direction the data supports: the monthly refund figure ranged from 0.575 billion to 49.183 billion within the same ten months, a spread of roughly 85 times. Extending a refund pace forward is extending a queue-clearing process whose queue length is not published. The correct posture is a range with the mechanism named, not a point estimate.

One further caution on the source. The Table 3 record for July 2026 also carries a budget-estimate field reading 406,059.00, in a table whose other amount fields are stated to the cent. The unit convention for that column could not be confirmed against a published field definition, so no comparison against it appears above. A figure whose scale is uncertain is worth nothing in a ratio.

Where this framework does not apply

The gross-versus-net distinction is the right lens for the July 2026 statement. It is not always the right lens, and there are conditions under which leaning on it produces a worse reading than ignoring it.

First, when refunds are small and stable, the two bases converge and the distinction is noise. Through April 2026 refunds ran at 5.10 percent of gross, and in the prior year's comparable ten months at 4.26 percent. At those levels a reader who tracked only the gross series would have been wrong by a few percent, not by a factor. The definition trap opens only when the refund share moves, and it moved this year.

Second, the net basis is not automatically the more conservative one. Net is the figure that flows to the deficit arithmetic, which argues for using it. But net in a month absorbs refunds of duties collected in earlier fiscal years as well as this one, so a net monthly series is not a clean measure of current-period trade taxation either. Both series are contaminated for the purpose most readers actually have in mind, which is measuring what current import flows are paying.

Third, neither series can be converted into an effective tariff rate here. That requires an import-value denominator from a primary trade-statistics source, which is not in these tables, and dividing customs receipts by an import figure drawn from a different agency with a different period convention would manufacture precision that does not exist. That calculation is out of scope until the denominator is sourced properly.

Fourth, a small internal detail worth carrying: the ten published monthly gross figures sum to 269,156,006,376.46 dollars against a published fiscal-year-to-date gross of 269,156,006,377.11, a residual of 65 cents. It is immaterial, and it is also a reminder that the fiscal-year-to-date column is a published aggregate rather than a re-summation of the monthly rows. Do not treat the two as interchangeable at high precision.

What to Watch Next Week

The next Monthly Treasury Statement, covering August 2026, is the single highest-value observation available. Three things in it settle questions this one leaves open.

• The August refund figure. If it falls back toward the 0.575 to 1.862 billion band that held from October to April, the May-to-July concentration was a discrete clearing event and the gross series remains the better trend measure. If it stays above 20 billion, refunds have become a running feature and the net series has to be the headline.
• The August gross figure against 24.835 billion. A fourth consecutive month in the 21.9 to 24.9 billion band would confirm that gross collections plateaued well below the October peak, which weakens any characterisation of the year as an accelerating revenue trend.
• Whether the fiscal-year-to-date net line in Table 3 falls below the prior-year comparable. It stood 13.84 percent above through July, a cushion of 18.785 billion dollars over the prior-year figure of 135.688 billion. That cushion does not disappear inside this fiscal year at July's pace: two more months at July's net of negative 8.546 billion leave the line at 137.380 billion, still 1.692 billion above the prior-year comparable, and erasing the cushion outright takes 2.20 months of that pace, which is three monthly statements rather than the two the fiscal year has left. So the thing to read in August is the size of the cushion, not a crossing. If August's net comes in worse than negative 9.392 billion, the cushion falls below one month of that pace and a crossing early in fiscal 2027 becomes the live question.

The U.S. Treasury Fiscal Data, Monthly Treasury Statement dataset publishes both tables through an API as soon as the statement is released, and the Treasury Fiscal Data API, MTS Table 4 endpoint returns the gross, refund, and net columns together, which the printed presentation of the statement does not put side by side as conveniently. Background on what the statement covers is at the Bureau of the Fiscal Service, Monthly Treasury Statement page, which states that it "summarizes the financial activities of the federal government and off-budget federal entities and conforms to the Budget of the U.S. Government."

Concrete Framework

A checklist for any receipts line where collections and refunds are reported separately.

1. Establish which basis the number you are reading is on before quoting it. For customs duties, Table 4 column one is gross, column three is net, and Table 3 carries net only. If a figure is quoted without a basis, assume nothing.
2. Compute the refund share of gross for the current period and the prior comparable period. Here: 42.61 percent against 4.26 percent. A share that moved by more than a few points means the two bases will tell different stories, and the divergence is the story.
3. Check whether the growth rate survives a basis change. Here it does not: 89.90 percent gross against 13.84 percent net. Any claim that does not survive this test should be stated with the basis attached, permanently.
4. Look at the monthly series, not only the year-to-date aggregate. A year-to-date total that includes an early peak can keep reading as growth long after the monthly series has turned. Here the peak was the first month of the fiscal year.
5. Locate the procedural lag before extrapolating. If refunds are governed by a completion event such as liquidation, the refund month is not the collection month, and a run rate built on refunds is a run rate on an unpublished queue. Name the queue or drop the extrapolation.
6. Confirm units for every field you put in a ratio. A field whose scale is not documented does not go into a denominator, however tempting the comparison.
7. Re-derive aggregates from components once. If they do not tie, note the residual and state which one you are using. Do not silently switch between them.

The general principle underneath all seven: when an institution publishes the same item twice, it is usually not a mistake, and the difference between the two publications is normally where the substance is. The instinct to pick the larger number and move on is the one that gets punished. In this case the larger number is a real number, the smaller number is a real number, and the 114.683 billion dollars between them is the only part of the year that is genuinely new.