July's Goods Deficit Printed Twice: $118.8 Billion, Then $119.6 Billion

Two releases from the same building put a number on the same month. On August 27, 2026, the Census Bureau's U.S. Census Bureau, Monthly Advance Economic Indicators Report said that “The international trade deficit was $118.8 billion in July, up $17.4 billion from $101.4 billion in June.” Seven days later, on September 3, the joint Census Bureau and Bureau of Economic Analysis release U.S. Census Bureau, U.S. International Trade in Goods and Services (FT-900) put the July goods deficit at $119,592 million, or $119.6 billion.

Subtract and the gap is $792 million. A reader who has watched a few of these months will file that under revision and move on. That filing is half right, and the half that is wrong changes what the number means. Between August 27 and September 3, two separate operations were performed on the July figure. One added data. The other changed the accounting basis. The gap of $792 million is their net result, and the two operations do not point the same way.

This piece follows the $792 million to the end: which document performs which operation, where the split is published, and at what frequency.

Two Releases, One July, Two Deficits

Start with what each release claims to be measuring. The Advance Economic Indicators Report carries a standing footnote: “Data are presented on a Census basis. The information needed to convert to a balance of payment basis is not available.” That is the first operation named in advance. The advance report is not attempting a balance of payments figure. It is publishing the customs record, seasonally adjusted, and saying so.

One Table 1 footnote in the same release names the other operation, and it runs past the FT-900: “Current month statistics reflect nearly complete coverage; current month statistics reflecting complete coverage will be published in the upcoming U.S. International Trade in Goods and Services report (FT-900). Statistics for other months shown reflect complete coverage and appear in the most recently published FT-900.”

So the advance report tells a reader two things about its own $118.8 billion. It is a Census-basis number, and it is built on coverage that is not yet complete. The FT-900 seven days later differs on both counts at once. Its Exhibit 1 note reads “(1) Data are presented on a balance of payments (BOP) basis.” Its goods figures for July are exports of $201,035 million and imports of $320,627 million, which subtract to the printed balance of -$119,592 million.

The June pair is worth putting beside it. The advance report of August 27 carried June at $101.4 billion. The FT-900 of September 3 carried revised June at -$101,973 million, or $102.0 billion, from exports of $207,248 million and imports of $309,220 million, which subtract to -$101,972 before rounding. Gap: $573 million. And the month-over-month change differs too. On the Census basis as printed, the deficit widened by $17.4 billion. On the balance of payments basis, $119,592 million less $101,973 million is $17,619 million, or $17.6 billion.

What the Advance Report Says About Its Own Basis

The temptation is to treat the advance number as a preliminary version of the FT-900 number. It is not a version of it. It is a different accounting object that happens to be close in size.

The FT-900 notes state the relationship directly: “Goods on a Census basis are adjusted by BEA to a BOP basis to align the data with the concepts and definitions used to prepare the international and national economic accounts.” The same note explains why the adjustments exist rather than leaving it to inference. They are “necessary to supplement coverage of the Census data, to eliminate duplication of transactions recorded elsewhere in the international accounts, and to value transactions at market prices.”

Three purposes, and only the first sounds like more data. A transaction can be fully and correctly captured in the customs record and still need to be deducted, because the international accounts already record it elsewhere. That is the part of the $792 million that no amount of additional filing would produce.

The note also states how the adjustments are presented: “They include both additions to and deductions from goods on a Census basis and are presented in this release as net adjustments.” Net, in one figure, per side. And it states where the components live: “BEA also publishes more detailed quarterly and annual statistics for net adjustments in ITA Table 2.4. U.S. International Trade in Goods, Balance of Payments Adjustments.”

That last sentence is the frequency trap. The monthly release prints the net figure. The itemization appears quarterly and annually in a different table. A reader who wants to know which component moved in a single month is reading a document built to answer that question at a coarser frequency.

Exhibit 5 Is Where the Two Bases Sit Side by Side

The FT-900 does publish the reconciliation, and the place to find it is Exhibit 5, titled “Exhibit 5. U.S. Trade in Goods” and qualified by the line “In millions of dollars. Details may not equal totals due to seasonal adjustment and rounding. (R) - Revised.”

Its column structure is the whole argument of this piece in printed form. Under Balance there are two columns, Total Balance of Payments Basis and Total Census Basis. Under Exports there are three: Total Balance of Payments Basis, Net Adjustments, Total Census Basis. Under Imports, the same three. The exhibit therefore prints the Census-basis figure, the net adjustment, and the balance of payments figure for each side of the account, month by month, with a January-through-July cumulative row above the monthly rows.

Anyone asking which part of the $792 million is basis and which part is coverage can read the answer off that exhibit by comparing its Census-basis column against the advance report's earlier figure and its net adjustment column against zero. The reconciliation is not hidden and it is not deferred. It sits four exhibits past the headline table in the same PDF.

What the monthly exhibit does not break out is the composition of the net adjustment, and the note says where that goes instead: ITA Table 2.4, quarterly and annual. The distinction matters when a month moves unusually. A single monthly net adjustment figure cannot tell a reader whether nonmonetary gold moved or inland freight moved, and the release directs that question to a table with a longer publication cycle. The next chance to watch the coverage step by itself is already dated: the August advance report of September 30, 2026.

The Import Addition With No Export Mirror

Several adjustment items have a counterpart on the other side of the account. Fuel bought in U.S. ports by foreign carriers is an addition to exports; fuel bought in foreign ports by U.S. carriers is an addition to imports. Equipment repairs and developed motion picture film are deducted from both sides, and software is revalued at market value on both sides.

One item has no mirror, and it is a pure valuation rule rather than a coverage gap. The note reads: “Imports of goods from all countries should be valued at the customs value—the value at the foreign port of export including inland freight charges.” Then the exception: “However, the customs value for imports for certain Canadian and Mexican goods is the point of origin in Canada or Mexico.” And the remedy: “BEA makes an addition for the inland freight charges of transporting these goods to the U.S. border to make the value comparable to the customs value reported for imports from other countries.”

It is a geography problem. For a container arriving by sea, the customs value is struck at the foreign port. For a truckload leaving a plant in Ontario or Nuevo Leon, the recorded point can be the plant, not the border. That freight is inside a Rotterdam shipment's value and outside a Monterrey shipment's. BEA adds it back so the two are comparable. This is the same class of problem as the one in a prior look at what import price indexes exclude at the border: the boundary where a price is observed decides what the price contains.

A second import-side item has no mirror of its own, though the export list carries military grant-aid: “Imports by U.S. military agencies - This addition is made for purchases of goods abroad by U.S. military agencies, which are reported to BEA by the Department of Defense.” And one with no import counterpart on the export side: “Net exports of goods under merchanting - This addition is made to include the net value of the purchase and subsequent resale of goods abroad without the goods entering the United States.”

The adjustment set is asymmetric by design, because the two sides of the account carry different valuation conventions and different off-customs channels. An asymmetric adjustment applied to a deficit does not leave the deficit alone.

Where a $2,500 Threshold Enters the Same Arithmetic

One more item in the adjustment list is often read as a live monthly correction when it is not. The note lists, on both sides of the account, additions for “low-value (below reporting threshold) transactions” covering 1999 through 2009, phasing in a revised Census Bureau low-value methodology the note dates to goods on a Census basis beginning with statistics for 2010.

The threshold itself is a filing rule, not an estimation choice. Under Electronic Code of Federal Regulations, 15 CFR 30.37 Miscellaneous exemptions, electronic export information need not be filed for “Exports of commodities where the value of the commodities shipped from one USPPI to one ultimate consignee on a single exporting conveyance classified under an individual Schedule B number or HTSUSA commodity classification code is $2,500 or less.” The exemption “applies to individual Schedule B numbers or HTSUSA commodity classification codes regardless of the total shipment value,” but where a filer reports multiple items of one code it “only applies if the total value of exports for the Schedule B number or HTSUSA commodity classification code is $2,500 or less.” The section opens “Except as noted in § 30.2(a)(1)(iv)” and reserves the Census Bureau's authority to require reporting of normally exempt shipments.

The dates in the note are the point. The low-value adjustment is bounded at 1999-2009 and exists to phase in a revised Census methodology that took effect with 2010 statistics. For a July 2026 figure, the threshold's effect is inside the Census-basis estimate, not sitting in the BOP adjustment. A reader who attributes part of the $792 million to small shipments escaping the filing threshold is charging it to the wrong operation. The threshold matters, and it matters one step upstream.

Concrete Framework

The procedure that keeps the two operations apart, in the order the documents support:

1. Read the basis line before the number. The advance report states “Data are presented on a Census basis” and states that the conversion information is not available. FT-900 Exhibit 1 states “(1) Data are presented on a balance of payments (BOP) basis.” Two labels, two objects.

2. Take the difference, then refuse to name it. The $119.6 billion of September 3 less the $118.8 billion of August 27 is $792 million. That quantity is a sum of two operations with independent signs. Naming it a revision, or naming it an adjustment, is a guess until step 3.

3. Go to Exhibit 5, not to the headline table. Exhibit 5 prints Total Census Basis, Net Adjustments, and Total Balance of Payments Basis for exports and for imports. The Census-basis column against the advance figure gives the coverage move. The net adjustment column gives the basis move.

4. For composition, change frequency. The monthly exhibit gives net adjustments as one figure per side. ITA Table 2.4 gives components quarterly and annually. A monthly question about gold, military imports, or inland freight has to wait for a quarterly table, and the release says so.

5. Check whether the deficit and the change are on the same basis. The July widening was $17.4 billion as printed on the Census basis and $17,619 million on the balance of payments basis. Quoting one release's level with another release's change is how a $219 million discrepancy appears from nothing.

6. Keep the seasonal step separate from both. The unadjusted Census-basis July goods balance for the world was -$135,465.8 million. Seasonal adjustment is the largest single operation on the July number, larger than coverage and basis together. The FT-900 note puts it inside the basis step rather than ahead of it: “Adjustments that exhibit significant seasonal patterns are seasonally adjusted.” The same discipline applied to an earlier case where a benchmark revision landed on one month works here: name the operation, then measure it.

What to Watch Next Week

Two dated documents will settle the split for July, and both are already scheduled.

The August advance report arrives on September 30, 2026, per its own line: “The August 2026 Advance report is scheduled for release on September 30, 2026.” That release carries the prior month alongside the new one, so it will print a July Census-basis figure built on coverage that is no longer described as nearly complete. Compare that July figure against the $118.8 billion of August 27 and the coverage operation is measured on its own.

Then the FT-900 for August, scheduled by the Bureau of Economic Analysis, International Trade in Goods and Services for October 6, 2026. Its Exhibit 5 will carry July again, with the Census-basis column and the net adjustment column beside the balance of payments column. Read the July row there and the basis operation is measured on its own.

A reader who does both comparisons will have the $792 million decomposed without estimating anything. The habit to carry forward: when a headline moves between two releases from one agency, check whether the label above the number moved too, and look for the exhibit that prints both labels in adjacent columns.

What Would Invalidate This

Several things would break the account above, and they are worth stating with the same precision as the claim.

The $792 million and the $573 million are subtractions performed here, not figures printed by either agency. The advance report rounds to tenths of a billion, so both differences carry rounding of up to about fifty million dollars. If the underlying unrounded Census-basis figures were published, the gaps would shift.

The June comparison is the weaker of the two. The June Census-basis figure quoted here was published on August 27, before the FT-900 of September 3 revised June. The $573 million therefore mixes a basis difference with a revision, and it is used only as an order-of-magnitude reference, not as a measurement of either operation.

The claim that the two operations point in opposite directions rests on the sign of the net adjustment, which this piece has not read off Exhibit 5's numeric cells. Repeated machine extraction of that exhibit's figures from the release PDF returned inconsistent values, so those cells were set aside rather than quoted. If the July net adjustments turn out to narrow rather than widen the deficit, the direction of the argument changes while its structure stands: the gap would still be two operations, not one.

Finally, Exhibit 5 carries the qualifier “Details may not equal totals due to seasonal adjustment and rounding.” Its cumulative row and its monthly rows are not built to sum exactly, so a reader reconciling January through July against seven monthly rows should expect a residual that belongs to the seasonal step rather than to either operation above.