Business Applications Rose 8.1% in July. On Our Arithmetic, Retail Trade Accounts for Almost All of the Gain.

by Tomas Ferreira · August 18, 2026

Markets Reporter, US Market Current

The U.S. Census Bureau's Business Formation Statistics for July, published at 10 a.m. on Aug. 12 as release CB26-130, opens with a number that looks like a broad revival of American business creation. "Business Applications for July 2026, adjusted for seasonal variation, were 578,926, an increase of 8.1 percent compared to June 2026," the release states. It is one of the larger one-month moves the series has printed outside a policy shock, and it is the figure that will be quoted.

The industry table inside the same document says the increase was not broad. Of the twenty rows the Census Bureau prints under "Business Applications by Industry," one moved decisively: retail trade, NAICS 44-45, at 141,241 seasonally adjusted applications, up 44.5 percent from June. Twelve of the other nineteen rows rose as well, but no row outside retail moved by as much as 9 percent in either direction.

The release publishes levels for July and month-over-month percent changes, not June levels, so the contribution has to be reconstructed. Doing that from the published percent changes -- our arithmetic, not the Census Bureau's -- and carrying the rounding of those percentages through as a range rather than a point: the national gain works out to roughly 43,100 to 43,600 applications, and the retail trade gain to roughly 43,460 to 43,530. Retail's increase therefore accounts for something between about 99.6 percent and 100.9 percent of the national increase. That band straddles 100 percent in both directions, and it is an artefact of the rounding: the published figures cannot establish that retail was literally the whole of the gain, and they cannot rule out that the rest of the economy contributed a small net amount in either direction. What they do establish is that the non-retail contribution is too small for this release to resolve.

Run the same reconstruction on everything that is not retail and the residual is close to nothing. Applications outside NAICS 44-45 stood at 437,685 in July -- a number the release does not print, and which is the national total less the retail row, our subtraction. The implied June figure lands between roughly 437,500 and 438,100, meaning the non-retail economy filed somewhere between about 400 fewer and about 165 more applications than it did in June. That is flat to the limits of what the published data can resolve.

We checked every row of the sector table, not a sample of it. The twenty rows sum to 578,926, matching the national total exactly, so nothing is missing or double-counted. Twelve rows rose, seven fell, and one -- other services, at 44,741 applications -- carries the code Z, which the release defines as "absolute value < 0.05". The decliners were utilities (566, down 8.0 percent), unclassified (13,152, down 4.6 percent), mining (537, down 3.8 percent), professional, scientific and technical services (81,366, down 2.5 percent), real estate (26,431, down 1.2 percent), arts and entertainment (15,165, down 0.8 percent) and construction (48,972, down 0.5 percent). One caveat applies to the largest of those: the release states that "Industry data, except for Utilities (22), is seasonally adjusted at the national level," so the utilities line is not adjusted on the same basis as the column it sits in and should not be read alongside the others.

That professional services and construction both slipped matters, because those two sectors carry categories that the Census Bureau treats as evidence an application will end up with a payroll. The gains outside retail were concentrated in wholesale trade (9,704, up 6.0 percent), manufacturing (7,842, up 4.3 percent), management of companies (6,782, up 3.0 percent) and accommodation and food services (29,951, up 2.8 percent). The table also prints an agriculture row (3,959, up 3.4 percent), which we set aside: the Census Bureau's own definitions page states that Business Applications exclude "applications with a NAICS sector code of 11 (agriculture, forestry, fishing and hunting) or 92 (public administration)," and we could not reconcile that exclusion with the row's presence in the July table. Nothing here rests on it.

The release publishes four application series, and in July they did not point the same way. Business Applications rose 8.1 percent. High-Propensity Business Applications rose 1.4 percent, to 151,857. Business Applications from Corporations rose 3.2 percent, to 44,738. Business Applications with Planned Wages fell 1.6 percent, to 35,024. Those are all four of the application series in the release.

The four are not parallel measures. The Census Bureau's definitions page describes a Business Application not as a statement of intent but as a filter: "The core business applications series that corresponds to a basic filtering of EIN applications." It takes in every EIN application except a listed set of exclusions -- "applications for tax liens, estates, trusts, certain financial filings, applications outside of the 50 states and DC or with no state-county geocodes, applications with a NAICS sector code of 11 (agriculture, forestry, fishing and hunting) or 92 (public administration), and applications in certain industries (e.g. private households, and civic and social organizations)." High-Propensity Business Applications are "Business Applications that have a high-propensity of turning into businesses with payroll." Applications with Planned Wages are "High-Propensity Business Applications that indicate a first wages-paid date on the IRS Form SS-4"; the page adds that "The indication of a wages-paid date is associated with a high likelihood of transitioning into a business with payroll." Applications from corporations are "High-Propensity Business Applications from a corporation or personal service corporation, based on the legal form of organization stated in the IRS Form SS-4." The last two are each drawn from the high-propensity series, which is itself drawn from all applications; they are not a single chain, and they overlap. So the only series in the release that requires an applicant to name a date on which they intend to start paying someone is the one that went down. Everything the July release says about employer intent, as opposed to paperwork volume, is carried by 35,024 filings out of 578,926.

The high-propensity series is the more revealing check. Reconstructing its June level the same way -- again our arithmetic, not the Census Bureau's -- high-propensity applications rose by roughly 2,020 to 2,170 in July, against a national application increase of roughly 43,100 to 43,600. On the release's own classification, about one in twenty of July's additional applications carried a marker the Census Bureau associates with becoming an employer. The rounding in the published percent changes puts that share somewhere between roughly 4.6 and 5.0 percent, which is not precise enough to say whether it sits above or below one in twenty.

Whether retail applications can be high-propensity at all depends on which part of the sector they came from, and the release does not let an outside reader find out. The Census Bureau's methodology counts an application as high-propensity if it is "(a) from a corporate entity, (b) that indicate they are hiring employees, (c) that provide a first wages-paid date (planned wages); or (d) that have a NAICS industry code in accommodation and food services (72) or in portions of construction (237, 238), manufacturing (312, 321, 322, 332), retail (44, 452), professional, scientific, and technical services (5411, 5413), educational services (6111), and health care (621, 623)." Criterion (d) reaches NAICS 44 and 452, not the whole of 44-45. The published sector row combines 44 and 45 into a single line, so the split between the qualifying codes and the rest of the sector is not disclosed at this level of the release. What is disclosed is that the high-propensity total moved 1.4 percent while the combined retail row moved 44.5 percent.

By our calculation on the July levels, high-propensity applications were 26.2 percent of all applications in July, corporations 7.7 percent, and planned-wages filings 6.0 percent. Retail trade by itself was 24.4 percent of the national total -- more than professional services and construction put together.

The formation side of the release moved barely at all. "Projected Business Formations (within 4 quarters) for July 2026, adjusted for seasonal variation, were 29,959, an increase of 0.7 percent compared to June 2026," the document says. The eight-quarter projection came in at 40,777, down 0.7 percent. Those are the two projected horizons the monthly release leads with, and they moved in opposite directions by the same rounded amount; the full BFS dataset, the Census Bureau notes, carries "all 12 series (four application and eight formation)".

That near-immobility is mechanical rather than mysterious, and it is worth stating plainly rather than reading as a contradiction. Applications and projected formations are different objects: the first counts filings, the second estimates how many of those filings become employer businesses. The Census Bureau's methodology describes using "a Linear Probability Model (LPM)" to estimate "the probability of an application transitioning to an employer," and says the approach "amounts to reweighting each application by the predicted probability that the application becomes an employer business between quarters t and t + k." A surge made up of applications the model scores low will lift the raw count and leave the reweighted count roughly where it was. That is what July looks like.

The projections are also carrying more weight than most readers assume. The release states that "the Census Bureau can only state actual business formations within 4 quarters up to December 2022, and business formations within 8 quarters up to December 2021," and that "Projected business formations within 4 quarters start in January 2023 and projected business formations within 8 quarters start in January 2022 and both end in July 2026." There is, in other words, no observed employer-formation outcome in this release for any application filed in the past three and a half years. Every formation figure covering 2023 through 2026 is model output.

One caution on the reconstruction above. The Census Bureau's methodology states that "Each month adjusted application and formation series are revised on for the prior two months, as well as the current and previous month in the prior year," and that "Factors for seasonal adjustments are recomputed and the seasonally adjusted applications and formations series are revised annually." The June figure sitting behind the published 8.1 percent is therefore a revised June, not the one printed on July 9. The July level and the 8.1 percent come from the same vintage, so the comparison inside this release is internally consistent -- but the June base it implies is not the June the Census Bureau published in July. That is why the contributions here are given as ranges rather than single numbers, and why the June levels are not quoted directly.

The regional table tells a second story about composition. The South filed 257,650 applications, the West 141,947, the Midwest 100,780 and the Northeast 78,549; the four sum to the national total exactly, as do the regional columns for each of the other three application series. The regional totals all rose -- the Midwest by 13.2 percent, the South by 7.8 percent, the West by 6.9 percent and the Northeast by 4.9 percent -- but the high-propensity column split: up 11.7 percent in the Northeast and 1.1 percent in the Midwest, down 0.9 percent in the South and 1.0 percent in the West. The density of employer-type filings varies sharply too. High-propensity applications were 33.6 percent of Northeast filings and 23.8 percent of Midwest filings, by our calculation. Corporate applications were 14.6 percent of Northeast filings and 5.0 percent of the Midwest's -- 11,486 against 5,035, on a Midwest application base about 28 percent larger than the Northeast's. Planned-wages filings, by contrast, sat between 5.6 and 6.5 percent of applications in all four regions.

The release contains no note explaining the retail movement. There is no methodology change flagged for July and no data-processing advisory. What the tables do carry are standing seasonal-adjustment notes -- that industry data other than utilities is adjusted at the national level, and that everything but the industry data is adjusted at the state level -- and nothing else. The Census Bureau published the number and left it there.

The next monthly release, covering August, is scheduled for Sept. 11 at 10 a.m. Two things in it will settle how much of July was signal: whether the retail row holds near 141,000 or gives the gain back, and whether the revision to June -- which the September release will still be making -- shifts the base that produced the 8.1 percent in the first place. Until then, the defensible statement about July is narrow. Filings rose sharply, one sector accounts for very nearly the whole of the rise, and the series built to detect employers barely moved.