Nondurable Orders and Nondurable Shipments: The Same $324,224 Million Appears in Both Tables

The U.S. Census Bureau, Monthly Full Report on Manufacturers' Shipments, Inventories and Orders for July 2026 opens with the sentence that trading desks price off: "New orders for manufactured goods in July, up following two consecutive monthly decreases, increased $5.8 billion or 0.9 percent to $663.6 billion, the U.S. Census Bureau reported today." Released September 2, it reads as a demand signal. Manufacturers booked more forward work in July than in June.

Open the two tables underneath that sentence and roughly half of the $663.6 billion is not a measurement of forward work. It is the shipments figure under a different column heading. This is not a quality complaint; it is arithmetic Census publishes in the open, in a formula on its own website, visible to the dollar if you set Table 1 beside Table 2.

The Identity Census Prints on Its Own Site

The U.S. Census Bureau, About the Manufacturers' Unfilled Orders (M3UFO) Survey page states the production rule without ornament: "NEW ORDERS (current) = SHIPMENTS (current) + [UNFILLED ORDERS (current) - UNFILLED ORDERS (prior)]".

New orders, in the monthly manufacturing report, is a derived series: not a count of commitments that arrived in the mail, but what falls out of two other reported quantities. The U.S. Census Bureau, Manufacturers' Shipments, Inventories, & Orders Definitions page states the same relationship from the backlog side, and the hedge in the first word matters: "Generally, unfilled orders at the end of an accounting period are equal to unfilled orders at the beginning of the period, plus new orders net of cancellations received during the period, less net sales."

The definitions page then adds the sentence that decides this whole article, qualifier included: "Orders data are not collected from industries that have mostly immediate deliveries." The reporting instructions repeat it with a routing note: "Orders data are not collected from industries that have immediate deliveries. See Part G to determine if the M3 survey request orders data for your industry category."

Feed a zero backlog through the identity and it collapses. Zero this month and zero last month makes the bracketed term zero, and new orders equals shipments. Not approximately. Identically, by construction.

Table 3 Has No Nondurable Rows

The proof is not buried. It is a missing block of rows.

The U.S. Census Bureau, Table 3, Value of Manufacturers' Unfilled Orders for Industry Groups lists seven durable major groups: primary metals, fabricated metal products, machinery, computers and electronic products, electrical equipment and appliances and components, transportation equipment, and furniture and related products. It lists no food products, chemical products, petroleum and coal products, or any other nondurable group. Nondurable goods industries do not appear in Table 3 at all.

The consequence shows in the top line. For July 2026, seasonally adjusted, all manufacturing industries unfilled orders is $1,600,284 million. Durable goods industries unfilled orders is also $1,600,284 million. The two are the same number because the nondurable contribution is zero. The release changes nouns after its lead sentence, though not for that reason. The lead reports "New orders for manufactured goods"; every revision paragraph that follows is a durable goods paragraph, shipments and inventories included, and the unfilled orders one reads "Unfilled orders for manufactured durable goods in July, up twenty-four of the last twenty-five months, increased $9.9 billion or 0.6 percent to $1,600.3 billion, unchanged from the previously published increase." Table 3 does print an all-manufacturing unfilled orders line. It is the durable line over again.

Now put U.S. Census Bureau, Table 1, Value of Manufacturers' Shipments for Industry Groups next to U.S. Census Bureau, Table 2, Value of Manufacturers' New Orders for Industry Groups. Nondurable goods shipments, seasonally adjusted: $324,224 million in July 2026, $322,011 million in June, $324,994 million in May. Nondurable goods new orders, seasonally adjusted: $324,224 million in July, $322,011 million in June, $324,994 million in May.

Three consecutive months, matching to the last dollar. Not rounding. The nondurable new orders series is the nondurable shipments series.

Scale it. Total new orders in July 2026 was $663,616 million, of which durable goods was $339,392 million and nondurable goods was $324,224 million. That puts 48.9 percent of the headline aggregate in the category where the orders label carries no separate measurement. Total shipments was $658,833 million on the same basis: $334,609 million durable plus the identical $324,224 million nondurable.

Forty-Five Categories Without a Backlog

It would be convenient if the split ran along the durable and nondurable line. It does not, and the survey documents where else it falls.

The M3 survey runs 92 industry categories combined into 65 publication levels. The unfilled orders survey that benchmarks the backlog covers 47 of them, across eight major groups: primary metals, fabricated metal products, machinery, computers and electronic products, electrical equipment and appliances and components, transportation equipment, furniture and related products, and miscellaneous products. That leaves 45 categories outside the backlog survey, and not all are nondurable.

Table 7 of the full report, which carries the unfilled orders to shipments ratio, names the durable-side exclusions in a footnote that has to be read to its end: "Excludes the following industries with no unfilled orders: Wood products; nonmetallic mineral products; medical equipment and supplies; office supplies; other miscellaneous manufacturing; cutlery and handtools; farm and garden machinery; computer storage and peripheral devices; audio and video equipment; manufacturing and reproducing magnetic and optical media; batteries; automobiles; light trucks and utility vehicles; heavy duty trucks; and miscellaneous furniture."

Automobiles. Light trucks and utility vehicles. Heavy duty trucks. Three of the fifteen named industries are the vehicle assembly lines themselves. So when the July release says "Transportation equipment, up following two consecutive monthly decreases, led the increase, $2.6 billion or 2.3 percent to $116.2 billion," the vehicle assembly portion of that increase is a shipments increase wearing an orders label. The backlog inside transportation equipment is enormous, $1,006,294 million in July against $1,001,732 million in June, but it sits in aircraft and related equipment, not in assembly plants. A reader who treats the whole line as forward demand is aggregating two kinds of number.

The true share of the $663,616 million that is definitionally shipments is therefore above 48.9 percent. Census does not publish the exact figure, because vehicle assembly is not broken out from motor vehicle parts. The aggregate it does publish, motor vehicles and parts, shows advance new orders of $73,877 million against shipments of $73,750 million in July, a gap of $127 million arising on the parts and bodies side, where some backlog exists.

Where the Identity Fails to Close

Run the identity on the durable aggregate and it does not balance, which is instructive.

Durable goods shipments in July 2026 were $334,609 million. Unfilled orders rose from $1,590,419 million in June to $1,600,284 million in July, a change of $9,865 million, which the release rounds to $9.9 billion. Shipments plus the change in backlog gives $344,474 million. Published durable goods new orders was $339,392 million. The residual is $5,082 million.

The footnotes explain it. Table 2 carries the note "Data on new orders are not available for the semiconductor industry. Estimates and percent changes for new orders exclude semiconductor industry data," and Table 3 carries the parallel note for unfilled orders. Table 1 states the mirror image as an inclusion: "Estimates and percent changes of shipments for the semiconductor industry are included in computers and electronic products, and all other applicable aggregate totals." Shipments includes semiconductor manufacturing; the orders side drops it. So the $5,082 million residual is an implied semiconductor shipments figure that the identity cannot absorb. That number is computed here, not published by Census, and should be treated as a residual rather than an estimate.

The practical point is firmer. Anyone reconciling new orders against shipments and backlog here finds a gap on the durable side and none on the nondurable side, and it is definitional, not an error. The same failure appeared in the natural gas storage series, where net change and implied flow part company at the reclassification line. An identity that closes in the methodology document need not close on the published aggregates, because the aggregates have different coverage.

What the Series Is Advertised to Do

The clearance paperwork states the purpose in plain terms. The Federal Register, Agency Information Collection Activities; Submission to OMB; the Manufacturers' Shipments, Inventories and Orders (M3) Survey notice of July 20, 2026, under OMB control number 0607-0008, says: "Estimates of new orders serve as an indicator of future production commitments and represent the current sales value of new orders received during the month, net of cancellations." It adds that "Substantial accumulation or depletion of unfilled orders measures excess or deficient demand for manufactured products." The same notice, printed at 91 FR 45236 in the U.S. Government Publishing Office, Federal Register Volume 91, Number 137, records 5,000 respondents reporting voluntarily under Title 13 U.S.C. 131, 182, and 193.

"An indicator of future production commitments" describes the 47 categories with a backlog fairly. For the other 45 it describes nothing the series can do. Where no backlog is collected, new orders is a contemporaneous delivery measure, and a delivery measure cannot lead itself.

The downstream user agrees. The Bureau of Economic Analysis, NIPA Handbook Chapter 3: Principal Source Data lists what it takes in a revealing order: "Data are collected on the value of shipments, on total inventories and inventories by stage of fabrication, and on new orders received and unfilled orders." Shipments first. The uses BEA names are "estimating investment in private equipment, change in private inventories, and nonfarm proprietors' income," all built from delivered value, not commitments. The orders headline moves markets; the shipments line moves the national accounts.

What Would Invalidate This

Four conditions would break the reading above. Three have a monthly test; the fourth arrives with the annual benchmark.

Census starts collecting nondurable backlogs. The claim that 48.9 percent of the $663.6 billion headline is a relabelled shipments figure rests entirely on the bracketed term in the identity being zero for nondurables. If the unfilled orders survey widens past its 47 categories into food, chemicals, or petroleum and coal products, Table 3 gains nondurable rows and the exact match ends. The test is one subtraction: nondurable new orders minus nondurable shipments. The month it stops returning zero, this article is obsolete for nondurables, and all-manufacturing unfilled orders stops equalling durable goods unfilled orders.

The semiconductor treatment changes. The $5,082 million residual is readable as implied semiconductor shipments only while the footnote asymmetry holds: new orders and unfilled orders exclude the industry, shipments do not. If Census restores semiconductor orders data, or drops semiconductor shipments from Table 1, the residual collapses and the inference behind it was wrong. Watch for the sentence "Data on new orders are not available for the semiconductor industry" disappearing from the Table 2 footnotes. A residual that turns negative is the same warning: the gap is being driven by something other than the exclusion.

The Table 7 exclusion list shrinks. If automobiles, light trucks and utility vehicles, or heavy duty trucks begin reporting a backlog, the claim about vehicle assembly inside the transportation equipment orders line stops holding, and the true share of definitionally-shipments dollars falls back toward 48.9 percent rather than sitting above it.

An annual benchmark rewrites the backlog levels. Unfilled orders estimates are ratio-based and benchmarked to annual totals, so a revision can move the durable series with no change in collection. This condition reaches the durable residual, not the nondurable match: a benchmark revises both sides of an equality that holds by construction, so only a change in collection can end it.

What would not invalidate it: a large move in the headline either way, or the advance-to-full revisions, which are an order of magnitude too small to matter. Those change the level, not what the level counts.

Concrete Framework

Four checks, in order, before treating any orders print as forward information.

One. Ask whether the category appears in Table 3. If an industry group has no unfilled orders row, its new orders row equals its shipments row, and the orders print is a delivery print. Nondurables fail this test as a bloc. Fifteen named durable industries fail it individually.

Two. Recompute the identity for the aggregate you are quoting. Shipments plus the change in backlog should reproduce new orders. On the nondurable aggregate it does so exactly, both terms being the same series. On the durable aggregate it leaves $5,082 million in July 2026, and the footnotes say why: the orders side excludes semiconductors, the shipments side does not.

Three. Separate the coverage gap from the definition gap. The panel is roughly 4,700 reporting units representing roughly 3,000 companies, drawn from establishments with $500 million or more in annual shipments. Reported and imputed shipments together cover about 58 percent of the 2022 Economic Census total; reported data alone about 38 percent. The 20-point difference is imputation applied to shipments, which propagates into the derived orders figure. That is separate from, and additional to, the definitional problem here.

Four. Note which reference month you are holding. The advance report covers durable goods only, about 18 working days after month end; the full report adds nondurables at about 23 working days. Two releases, one reference month, and a reader who mixes them will read a revision as a change in the economy. The same trap appears in energy statistics, where one release day carried two reference months.

The base rates on revision are small. Durable goods new orders for July 2026 went from $339,250 million in the advance report to $339,392 million in the full report. June went from $335,653 million to $335,779 million; May was unchanged at $333,842 million. Shipments moved the other way in July, from $334,722 million to $334,609 million. None of these exceeds five hundredths of one percent of the level. The advance estimate is a good estimate of a series that, for half its weight, is not measuring what its name says.

What to Watch Next Week

The U.S. Census Bureau, M3 Release Schedule puts the August 2026 advance report on September 25 and the August 2026 full report on October 2. The advance release now carries an advance total manufacturing line alongside the durable goods detail, which makes the reconciliation check available a week earlier.

Three things to do with the October 2 file rather than the press release. First, pull Tables 1 and 2 and confirm nondurable new orders again equals nondurable shipments to the dollar. If it does not, something changed in the collection, not in demand. Second, recompute shipments plus the change in unfilled orders against published durable new orders and watch the residual. A residual far from July's $5,082 million is telling you something about semiconductor shipments, the one line the orders tables never show. Third, check Table 8 against the U.S. Census Bureau, Monthly Advance Report on Durable Goods Manufacturers' Shipments, Inventories and Orders figures recorded on September 25, keeping advance and full values side by side rather than overwriting one with the other.

Then resist the summary sentence. If the October 2 release says new orders for manufactured goods rose or fell by some percent, roughly half that move came from a column that is definitionally the shipments column, and the durable half arrives with a backlog term whose benchmark survey reaches only 47 of 92 categories. The number is not wrong. It is two numbers, and only one is about the future. Derived series carry this hazard generally, as with a monthly construction figure that comes from an annual forecast. Census prints the identity, prints the exclusion footnotes, and leaves the nondurable rows out of Table 3 where anyone can see the hole.