Prolepsia forecastYes / no

Will US business applications in 2027 exceed 5.5 million, according to the Census Bureau?

As of 16 November 2026, Prolepsia puts the chance of yes at 95%.

95%chance of yes

Yes 95%No 5%
Prolepsia forecast made

How Prolepsia reasons

The report Prolepsia wrote with this forecast.

How forecasts are made · Track record

TL;DR

The forecast is 94.9% for Yes and 5.1% for No. Current application levels provide a substantial cushion above 5.5 million, but the outcome requires that elevated filing activity persist through a year for which no observations are yet available.

Key evidence

  • The current level is well above the threshold. Unadjusted applications totaled 5,360,944 in January–October 2026, compared with 4,832,972 in the same period of 2025. The trailing twelve-month total was 6,224,714. These counts establish the starting position; none carries over into the separate 2027 total (Census-sourced application series).
  • Even the weaker latest month remains above the required pace. October recorded 501,609 seasonally adjusted applications, down 6.0% from September. Sustaining October's pace would imply an annualized benchmark of 6,019,308—not a forecast of the resolving total (Census release; seasonally adjusted series).
  • The headline decline overstates the breadth of the retreat. Retail accounted for about 79% of September–October's fall. Meanwhile, growth in January–October unadjusted applications was predominantly outside retail (retail seasonally adjusted data; retail unadjusted data).
  • The definition matters. Census's internet-sales exclusions affected high-propensity and corporate applications, not total BA. Those exclusions should not be applied again to the series that resolves this question (Census methodology).

Uncertainties

The main downside is sustained normalization in retail combined with deterioration elsewhere. Falling planned-wage applications warn that strong total filings do not imply equally strong employer creation (planned-wage series).

Administrative changes are another risk. An audit described planned January 2027 controls on daily EIN issuance, but deployment and the effect on qualifying applications remain unconfirmed (TIGTA audit). Historical comparisons also use revised data, whereas resolution uses the designated first release containing December 2027. The forecast concerns qualifying applications—not distinct businesses, successful launches, or payroll employers (Census definitions).

Read the full report

TL;DR

The forecast is 94.9% for Yes and 5.1% for No. The elevated application level and broad growth outside retail support exceeding 5.5 million in 2027. The decisive risk is a sustained reversal across several sources of applications, not a single weak monthly release.

Context

The question concerns qualifying applications for Employer Identification Numbers, not businesses that become operational or hire workers. The relevant Census Business Applications series covers the 50 states and Washington, D.C., and excludes specified nonbusiness applications and territories. High-propensity applications, corporate applications, planned-wage applications, and business formations are different measures (Census definitions).

Resolution uses the twelve not seasonally adjusted national BA observations labeled January–December 2027. Exactly 5,500,000 resolves No. The controlling figures are those in the first official release containing December 2027, expected in January 2028; subsequent revisions do not change the outcome. An official annual total takes precedence only if it has the same definition and coverage. Census publishes the relevant monthly data through its BFS data service.

Evidence

The historical record gives two different messages. Exceeding this threshold has been rare, but today's starting level is much higher than it was through most of the available history. Of the 21 complete years from 2005–2025, only 2025 exceeded 5.5 million. The supplied annual observations show the scale of the change in application activity (Census-sourced unadjusted series):

Year National BA, unadjusted
2015 2,795,897
2016 2,948,338
2017 3,182,674
2018 3,486,829
2019 3,498,990
2020 4,356,498
2021 5,390,816
2022 5,064,093
2023 5,471,404
2024 5,228,063
2025 5,696,742

This history argues against treating 5.5 million as an easy threshold in every environment. It also argues against using the unconditional historical frequency as the forecast. Application activity has moved to a different level. The relevant comparison is how much that level would need to fall before the 2027 total misses the threshold.

Historical comparisons at the same forecasting horizon support persistence. Across 20 October starting points from 2005–2024, the following calendar year's unadjusted total never fell more than 10% below the preceding August–October seasonally adjusted annualized pace. The weakest ratio was 0.9266, for 2023–2024. These comparisons use the Census-sourced adjusted and unadjusted series. They support Yes, but they are not a hard bound: the sample is small, largely reflects a rising trend, and uses revised historical data.

The current unadjusted counts provide more direct evidence. January–October 2026 totaled 5,360,944 applications, compared with 4,832,972 in January–October 2025—an increase of about 11%. The full supplied monthly history for 2026 is below (unadjusted BA series):

Month in 2026 National BA, unadjusted
January 548,819
February 506,166
March 535,628
April 650,032
May 499,268
June 497,361
July 635,310
August 478,271
September 582,422
October 427,667

The trailing twelve-month total reached 5,509,645 through October 2025, rose to 6,339,310 through September 2026, and eased to 6,224,714 through October 2026. The latest total is 724,714 above the question's threshold. That is evidence of an elevated filing level, not a floor for 2027: every application counted in the target year remains unobserved (unadjusted BA series).

Seasonally adjusted data help separate the underlying pace from monthly calendar effects. Applications were 530,216 in August, 533,637 in September, and 501,609 in October 2026. Census reported October's decline as 6.0% (adjusted series; official release). Annualizing October gives 6,019,308; annualizing the August–October average gives 6,261,848. Reaching the threshold would require declines of about 9% and 12%, respectively, from those benchmarks. These are scale comparisons, not alternative forecasts or substitutes for the unadjusted resolution measure.

The sector evidence changes the interpretation of October's retreat. Retail accounted for about 79% of the September–October decline. Yet retail contributed only 29,692 of the 527,972 additional unadjusted applications in January–October 2026 relative to the same period of 2025. Applications outside retail increased by 498,280. The inference is that retail explains much of the recent volatility, but not most of the year's broader increase (retail adjusted series; retail unadjusted series; total unadjusted series).

Employer intent supplies a counterweight. January–October seasonally adjusted applications with planned wages fell from 419,011 in 2025 to 359,498 in 2026. October 2026 recorded 34,131, the lowest observation in the supplied January 2025–October 2026 window. This warns against describing the total-application increase as a uniform employer-startup boom. It does not establish that total BA will fall proportionately (planned-wage series).

Classification changes do not erase the elevated total. Census removed certain internet-sales applications from HBA and CBA, not total BA. It also revised industry assignments and seasonal factors in its November 12, 2026 annual update. The correct reading is to keep the current BA definition and data vintage consistent, rather than subtract applications that remain eligible for the target measure (Census methodology; BFS updates).

Two possible drivers deserve attention without receiving a quantified effect. AI-enabled entrepreneurship is a plausible explanation for some growth, but the research does not establish causation or a durable addition to national BA. Evidence of rising likely-nonemployer applications without a clear corresponding increase in self-employment as a main job reinforces that distinction (EIG research). Separately, changes to platform tax reporting could alter seller formalization incentives, but a reporting obligation does not automatically require a new EIN, and its effect on applications is unmeasured (IRS guidance).

What's non-obvious

The quality of applications and the count of applications are different questions. A rise in side businesses or filings with little payroll potential can still count fully toward this threshold. Weak planned-wage activity therefore matters as a signal about persistence, not as grounds for removing applications from BA. Likewise, Census's internet-sales exclusions in employer-oriented subsets do not mechanically reduce the resolving total (definitions; methodology).

The other distinction is between the source of volatility and the source of growth. Retail dominates the conspicuous recent decline, while most year-to-date growth came from elsewhere. That weakens both extreme narratives: extrapolating earlier retail strength is too optimistic, but extrapolating October's entire headline decline across the economy is too pessimistic. The forecast rests on the elevated overall level and its breadth, not on continued rapid growth.

Uncertainties

The largest risk is joint deterioration: further retail normalization alongside a sustained decline outside retail. The October drop is real. Its concentration limits what it says about the whole economy, but does not make it harmless. Further monthly totals and industry decompositions would show whether the retreat is spreading.

Administrative changes create a separate risk. A TIGTA audit described planned January 2027 programming to enforce permitted daily EIN issuance limits. Deployment remains unconfirmed, and the audit does not quantify a resulting reduction in BFS-eligible applications. A confirmed implementation notice and measured effects on eligible filings would close that gap (TIGTA audit).

Historical evidence is also thinner than its apparent consistency suggests. Revised seasonal factors make past comparisons different from forecasts made with information actually available at the time. First-release annual totals can differ from later revised history. Census documents both revisions and administrative-data limitations, so historical persistence cannot establish certainty (Census methodology).

Finally, BFS months use whole-week statistical periods. The explicit resolution rule—the twelve labeled monthly observations—controls, rather than a separate reconstruction of daily filings (Census statistical calendar). None of 2027 has yet been observed. The saved forecast remains 94.9% for Yes and 5.1% for No, with the downside covering sustained normalization, economic or administrative shocks, and uncertainty in the designated first release.

Forecast history

How Prolepsia’s forecast has movedOne forecast so far: 95% on 16 Nov 2026.
DateForecast
95% chance of yes

Sources Prolepsia read

  1. Business Formation Statistics - Definitionswww.census.gov
  2. Business Applications: Total for All NAICS in the United States (BABATOTALNSAUS) | FRED | St. Louis Fedfred.stlouisfed.org
  3. AI and Entrepreneurship: What is Really Happening?eig.org
  4. IRS provides transition relief for third-party settlement organizations; Form 1099-K threshold is $5,000 for calendar year 2024 | Internal Revenue Servicewww.irs.gov
  5. Further Improvements Are Needed to Strengthen the Controls for Issuing Employer Identification Numberswww.tigta.gov
  6. AI Is Enabling More Entrepreneurship | Nasdaqwww.nasdaq.com
  7. www2.census.govwww2.census.gov

The question

This question asks whether the total number of business applications filed in the United States during calendar year 2027 (January 1 through December 31, 2027) will exceed 5.5 million, as reported by the U.S. Census Bureau's Business Formation Statistics (BFS) program. The relevant measure is the national Business Applications (BA) series, which tracks applications for Employer Identification Numbers (EINs) submitted primarily for business purposes. The measure excludes certain categories of EIN applications, including those associated with trusts, estates, tax liens, and specified nonbusiness activities. It measures business applications rather than the number of businesses actually established. The geographic scope includes the 50 U.S. states and Washington, D.C., excluding Puerto Rico and other U.S. territories. As of November 16, 2026, the latest Census Bureau release, published November 12, 2026, reported 501,609 seasonally adjusted business applications for October 2026, a 6.0% decrease from September 2026. This indicates a monthly application rate that, if sustained for 12 months, would exceed 6 million annual applications. The Census Bureau publishes monthly BFS data approximately 11–12 days after each month ends, making the complete 2027 figures likely available in January 2028. The question will be evaluated using the sum of the 12 monthly, not seasonally adjusted, national BA observations for 2027. Sources: U.S. Census Bureau Business Formation Statistics (https://www.census.gov/econ/bfs/index.html), November 12, 2026 release (https://www.census.gov/econ/bfs/current/index.html), and BFS definitions (https://www.census.gov/econ/bfs/definitions.html).

How it resolves

The question resolves YES if the sum of the U.S. Census Bureau's not seasonally adjusted national Business Applications (BA) figures for January through December 2027 is strictly greater than 5,500,000. The question resolves NO if the corresponding sum is 5,500,000 or fewer. The primary resolution source will be the U.S. Census Bureau's official Business Formation Statistics monthly data, available through https://www.census.gov/econ/bfs/data.html and https://www.census.gov/data/tables/time-series/econ/bfs/business-formation-statistics.html. The calculation will use the national BA series, not the High-Propensity Business Applications (HBA), Business Applications with Planned Wages (WBA), Business Applications from Corporations (CBA), or Business Formations series. Monthly figures must be not seasonally adjusted, since the objective is to count applications actually recorded during the calendar year rather than estimate seasonally adjusted application activity. The outcome will be determined using the figures available in the first official Census Bureau monthly BFS release that includes December 2027, expected in January 2028. If the Census Bureau publishes an official annual national BA total for 2027 in the same release, that total will take precedence over a manually calculated sum, provided it uses the same geographic coverage and unadjusted BA definition. Otherwise, the 12 monthly values will be summed. If the December 2027 release is delayed, resolution will wait until the complete data become available, up to June 30, 2028. If the necessary official data remain unavailable by that date, the question will be annulled. The question will also be annulled if a methodological change makes it impossible to obtain a comparable national BA total for 2027.

Fine print

The threshold is strictly greater than 5.5 million; exactly 5,500,000 applications resolves NO. Applications are counted according to the Census Bureau's BA methodology, regardless of whether they ultimately lead to operational businesses or employers with paid workers. The question concerns applications recorded during 2027, not businesses formed during 2027 or applications submitted in earlier years that result in business formation during 2027. The geographic scope excludes Puerto Rico, Guam, the U.S. Virgin Islands, American Samoa, and the Northern Mariana Islands. Revisions published after the designated resolution release will not change the outcome. The Census Bureau periodically revises historical data and classifications, but the question will use its official BA definition and data as published at the time of resolution. If both seasonally adjusted and not seasonally adjusted figures are provided, only the latter will count. If an annual total conflicts with the sum of monthly observations because it uses a different statistical definition or geographic scope, the monthly national BA series will control. The question does not require 5.5 million distinct businesses or individuals, since multiple EIN applications may not correspond to distinct new businesses.