Reading the latest movement
In September 2026, housing starts were 1,221 thousand units at an annual rate, down 67 from August 2026 and down 98 from September 2025. Across 813 readings since January 1959, the average is 1,430 thousand units at an annual rate, and the latest reading is higher than 25% of them. The average since January 2017 is 1,380 thousand units at an annual rate.
A difference between periods describes economic change in this edition of the data; a later revision of either observation is a separate event.
What the housing starts tells you
Housing starts count new privately owned housing units on which construction began during the month, reported as a seasonally adjusted annual rate.
Housing starts are a direct measure of new residential construction activity, which feeds into residential investment in GDP and drives demand for lumber, appliances, furnishings and construction labor. Because housing is especially sensitive to mortgage rates, starts are one of the clearest channels through which monetary policy affects the real economy, and economists watch them for early signs of turning points.
Homebuilders, building-product suppliers, lenders and local planners use starts to assess market conditions and to judge whether the country is adding enough housing. Analysts tracking housing shortages compare starts over time with household formation, and the split between single-family and multifamily starts helps distinguish trends in owner-occupied housing from rental supply. Starts also feed directly into estimates of residential investment, so they influence quarterly GDP projections.
How to read it
- Separate single-family from multifamily (structures with five or more units). Multifamily starts are lumpy because a single large project can add hundreds of units in one month.
- Monthly changes often fall within the sampling error. Census reports confidence intervals, and many monthly moves are not statistically significant; use three-month averages or year-over-year comparisons.
- The annual rate shows the pace if that month's activity continued for a year. Unadjusted monthly counts are much smaller and have strong seasonal patterns.
- Regional data (Northeast, Midwest, South, West) can reveal weather effects, since cold or wet months suppress starts in some regions and are later reversed.
What moves it
Mortgage rates and builders' financing costs are the strongest short-run drivers, since they determine affordability for buyers and the viability of new projects. Demand depends on job growth, household formation, migration and the supply of existing homes for sale. Builder costs for land, labor and materials influence how many projects go ahead. Apartment construction responds to rents, vacancy rates and investor appetite. Weather, natural disasters and local zoning or permitting changes cause short-term swings.
Records and long-run averages
Computed from every published monthly reading (thousands of housing units, annual rate) in this edition, January 1959 to September 2026. Records reflect the acquired history only; earlier data published elsewhere is not included.
| Measure | Value | When / basis |
|---|---|---|
| Latest | 1,221 | September 2026 |
| Record high | 2,494 | January 1972 |
| Record low | 478 | April 2009 |
| Long-run average | 1,430 | Median 1,448 · January 1959 onward |
| Average since 2017 | 1,380 | Last ten calendar years |
| Latest percentile | 25th | Share of readings below the latest |
Year-by-year summary
Calendar-year average, range and final reading of the published level (thousands of housing units, annual rate). A year marked partial includes only the months published so far.
| Year | Average | Low | High | Last reading |
|---|---|---|---|---|
| 2026 partial | 1,352 | 1,182 | 1,522 | 1,221 September 2026 |
| 2025 | 1,356 | 1,273 | 1,491 | 1,378 December 2025 |
| 2024 | 1,370 | 1,269 | 1,555 | 1,494 December 2024 |
| 2023 | 1,421 | 1,321 | 1,609 | 1,504 December 2023 |
| 2022 | 1,550 | 1,296 | 1,807 | 1,296 December 2022 |
| 2021 | 1,603 | 1,398 | 1,711 | 1,711 December 2021 |
| 2020 | 1,394 | 936 | 1,614 | 1,614 December 2020 |
| 2019 | 1,292 | 1,121 | 1,548 | 1,548 December 2019 |
| 2018 | 1,247 | 1,095 | 1,357 | 1,095 December 2018 |
| 2017 | 1,205 | 1,146 | 1,273 | 1,177 December 2017 |
| 2016 | 1,177 | 1,068 | 1,313 | 1,252 December 2016 |
| 2015 | 1,107 | 886 | 1,224 | 1,146 December 2015 |
Readings around recent recessions
Values of the published level at each NBER business-cycle peak and trough inside this history, with the range from the peak through twelve months after the trough. Shaded bands on the chart mark the same NBER recession dates.
| Recession (peak to trough) | At peak | At trough | Low in window | High in window |
|---|---|---|---|---|
| February 2020 to April 2020 | 1,549 | 936 | 936 April 2020 | 1,688 March 2021 |
| December 2007 to June 2009 | 1,037 | 585 | 478 April 2009 | 1,103 February 2008 |
| March 2001 to November 2001 | 1,590 | 1,602 | 1,540 October 2001 | 1,829 February 2002 |
| July 1990 to March 1991 | 1,171 | 921 | 798 January 1991 | 1,297 March 1992 |
Recession dates: NBER Business Cycle Dating Committee. A reading at a peak or trough month is not a forecast signal on its own.
Put the latest level in context
These comparisons use published levels in this snapshot, with matching reference periods. Rate differences use percentage points; growth of a rate is not a percentage-point difference. The chart above may use a different transformation.
| Earlier period | Earlier level | Latest level | Absolute difference | Relative change |
|---|---|---|---|---|
| September 2025 | 1,319 | 1,221 | -98 | -7.43% |
| September 2021 | 1,576 | 1,221 | -355 | -22.53% |
| September 2016 | 1,068 | 1,221 | 153 | 14.33% |
How have prices, jobs and output changed since 2019? · Research questions and worked comparisons.
Definition and limitations
Counts housing units, not buildings. The annual rate describes the monthly pace; it is not actual units built that month. Permits, starts and completions are different stages and do not track one fixed cohort. Monthly estimates have sampling uncertainty and are revised.
Percentage growth is calculated from matching series levels. Both endpoints retain the same adjustment basis. Growth computed from rounded source inputs can differ from the agency’s published headline rounding.
Which measure fits your question? Compare definitions and complementary measures before drawing a conclusion.
Release schedule and revisions
The Census Bureau and the Department of Housing and Urban Development jointly publish housing starts in the monthly New Residential Construction release, typically around the 12th business day of the following month. Estimates come from the Survey of Construction. Each release revises the previous months, and annual revisions update seasonal factors.
See the producer's release page for the next publication.
Read it alongside
- Housing building permits: Permits usually precede starts, so they help anticipate where starts are heading in coming months.
- Housing completions: Completions show how much of the construction pipeline is turning into finished units ready for occupancy.
- 10-year Treasury yield: Mortgage rates broadly track the 10-year Treasury yield, the main financial influence on housing demand.
Frequently asked questions
When is a house counted as started?
Census counts a start when excavation begins for the footings or foundation of a residential building. For apartment buildings, all units in the building are counted as started at the same time. Renovations, additions and manufactured (mobile) homes are not counted as housing starts.
Why are housing starts reported at an annual rate?
Monthly seasonally adjusted counts are multiplied by 12 to show the yearly pace implied by that month. This makes months comparable with each other and with annual totals, despite seasonal swings in construction. It does not mean that number of units were started in that single month; the actual monthly count is about one-twelfth of the rate.
Why do housing starts and building permits not match?
Permits and starts measure different stages and cover slightly different areas. Starts include homes built in places that do not require permits, while some permitted projects begin months later or are never started. Timing differences mean a given month's permits and starts reflect different groups of projects.
Transformations
The published level keeps the source unit. Year-over-year change compares with the same period a year earlier (52 weeks for weekly data); previous-period change compares with the preceding observation, in percentage points when the series is already a rate. No frequency conversion, interpolation or hidden rescaling is applied. Formulas and methods.
Source and methodology
Producer documentation · Acquisition endpoint / file · Producer release information
Revisions and archived snapshots
Earlier site acquisitions are archived. Matching unchanged series identities can be compared between versions; this is not a complete agency vintage history. Previous-period readings above come from the same acquired snapshot. The revision comparison separates changes to matching historical observations from new coverage and corrected series identities. Earlier agency vintages remain outside this archive. Read the revision policy and download this frozen snapshot.