Housing starts estimate new privately owned residential units beginning construction; analysts compare permits, completions, revisions, and sampling error.
Housing starts estimate the number of new privately owned housing units on which construction began during a period. In U.S. statistics, a start occurs when excavation begins for the footings or foundation of a building. The measure counts housing units, not merely construction projects: an apartment building containing 100 units can contribute 100 housing starts.
Housing starts are an early measure of residential construction activity. They help analysts assess the supply pipeline, builders’ activity, demand for construction financing and materials, and the possible direction of future housing completions. They do not directly measure home prices, home sales, affordability, or investment returns.
The Census Bureau defines the start of construction as the point when excavation begins for the footings or foundation. Site planning, land purchase, demolition, grading, financing approval, and permit issuance can all occur earlier without creating a housing start.
The statistic counts housing units within new residential structures. This distinction matters:
The total number of units can therefore rise sharply because several large multifamily buildings started, even if fewer individual buildings broke ground.
U.S. new residential construction statistics cover privately owned housing. They exclude publicly owned housing, group quarters such as dormitories, transient lodging such as hotels, HUD-code manufactured homes, relocated buildings, and units created by converting an existing structure. A privately developed project can still be classified as private even when it receives partial public subsidies.
“Single-family” does not necessarily mean detached. The Census definitions can classify side-by-side attached units, such as some row houses and townhouses, as separate single-family structures when they are separated from ground to roof, have no units above or below, and meet the stated utility and heating criteria.
By contrast, apartments stacked above or below one another are multifamily units. Condominium ownership does not change the physical structure classification: units in an apartment building are classified with the applicable multifamily structure size.
Housing starts are one stage in a longer development process. Comparing the stages helps separate proposed supply from active construction and finished inventory.
| Measure | Trigger | What it indicates | Main limitation |
|---|---|---|---|
| Building permits | A local authority authorizes qualifying construction | Potential future construction in permit-issuing areas | A permitted unit may be delayed, redesigned, or never started |
| Authorized, not started | A permit exists but construction has not begun | Backlog that could enter construction | Timing remains uncertain and non-permit areas are not represented in this series |
| Housing starts | Excavation begins for footings or foundation | New units entering active construction | A start does not guarantee timely completion |
| Under construction | A started unit has not yet been completed | Active construction inventory | Includes projects at very different stages of work |
| Housing completions | Construction reaches the survey’s completion standard | New units reaching the end of the building pipeline | Completion may occur well after the original market and financing decision |
The stages need not move together in one month. Permits may rise while starts fall if builders postpone groundbreaking. Starts may decline while completions rise because builders are finishing an earlier wave of projects. A pipeline view is more informative than treating one series as a complete housing-market signal.
The headline total combines structure types with different financing, construction, and demand characteristics.
Single-family starts generally reflect units in one-unit structures, including qualifying attached homes. Analysts often compare this series with New Home Sales, builder inventory, mortgage rates, and construction costs.
Single-family construction may be built for sale, for rent, by an owner, or by a contractor on an owner’s land. Housing starts therefore have broader coverage than the new-home-sales series, which applies to new single-family houses built for sale and uses its own sale definition.
Multifamily data include units in structures with two or more units, while headline releases commonly highlight buildings with five units or more. A large project can add many units to the series at once, making monthly multifamily estimates particularly uneven.
Multifamily starts are useful for monitoring the future apartment and condominium pipeline, but structure type does not reveal tenure or affordability by itself. A multifamily unit may be intended for rent or sale, and a higher count does not prove that the units will be affordable to a particular household group.
Single-family and multifamily construction can diverge because they respond to different buyers, tenants, lenders, land constraints, project sizes, and completion timelines. Always inspect the components before attributing a change in total starts to the entire housing market.
The Census Bureau’s Survey of Construction (SOC), partly funded by HUD, produces national and regional estimates for housing starts and completions. The survey has two main parts:
Permit totals from the Building Permits Survey help scale the sampled Survey of Construction observations into national and regional estimates. The Census Bureau states that less than 2% of new construction occurs in non-permit areas, but those areas are included through a separate survey process.
This design explains why permits and starts are related but not identical. Permit data record authorization. Survey follow-up establishes whether and when authorized units actually begin construction.
Preliminary estimates are released monthly for the United States and Census regions. The national release also separates important structure categories. Local permit statistics can provide more geographic detail, but national housing-start estimates should not be treated as a precise measure for one city, neighborhood, or property.
Housing-start headlines are usually shown at a seasonally adjusted annual rate, or SAAR. Seasonal adjustment attempts to remove recurring seasonal patterns, such as typical differences between winter and summer construction. Annualization expresses the adjusted monthly pace as a rate for a full year.
Suppose the seasonally adjusted estimate for a month were 120,000 units. The annualized rate would be:
1Illustrative SAAR = 120,000 x 12 = 1,440,000 units
The 1.44 million figure would not mean that 1.44 million units actually started during that month. It would describe the annual rate implied by that month’s seasonally adjusted pace. The Census Bureau explicitly states that SAAR is neither a forecast nor a projection.
| Presentation | Best use | Important caution |
|---|---|---|
| Seasonally adjusted annual rate | Comparing the current pace across months after adjusting for recurring seasonality | Not an actual monthly count and not a forecast |
| Not seasonally adjusted monthly estimate | Understanding estimated activity that occurred in the specific month | Raw month-to-month comparisons can be distorted by normal seasonal patterns |
| Not seasonally adjusted annual total | Reviewing estimated activity over a full calendar year | Available later and less useful for detecting a new monthly turn |
When quoting a figure, label its basis. Calling a SAAR value “starts this month” confuses an annualized, seasonally adjusted rate with the actual monthly unit estimate. Dividing SAAR by 12 also does not necessarily reproduce the published not-seasonally-adjusted count.
Housing starts are estimates from a sample, not a complete real-time count of every foundation excavation. They are subject to both sampling and nonsampling error.
Sampling error exists because the survey observes a sample that represents a larger population. The monthly release provides measures of uncertainty around estimates and changes. A reported increase or decrease can be smaller than its confidence interval, meaning the survey does not provide strong statistical evidence that the underlying rate changed in that month.
The practical rule is simple: read the estimated change and its published margin of error together. Do not describe a noisy one-month move as definitive when the confidence interval includes no change.
Nonsampling error can arise from incomplete coverage, nonresponse, reporting mistakes, classification differences, recording or processing errors, and imputation. A narrow confidence interval addresses estimated sampling variability, not every possible source of error.
The first estimate is preliminary. Under the current SOC methodology:
These rules make the data vintage part of the observation. An analyst should retain the release date and distinguish the first-reported estimate from a later revised series.
A disciplined review moves from measurement to context rather than jumping from one percentage change to an economic conclusion.
Confirm whether the figure is a level, monthly percentage change, year-over-year change, SAAR, or not-seasonally-adjusted count. These values answer different questions.
Determine whether total starts were driven by single-family units, two-to-four-unit buildings, or buildings with five units or more. A multifamily surge does not necessarily imply an equally strong single-family market.
Compare the estimated change with its published margin of error. If the uncertainty is larger than the reported move, use cautious language such as “the estimate was little changed” rather than declaring a confirmed turn.
A strong current estimate may be offset by a downward revision to prior months. Conversely, an apparently weak current month can accompany upward revisions. Compare current and previously published vintages.
Permits can indicate potential future starts, while completions show units reaching the end of construction. The authorized-not-started and under-construction series show the backlog between those stages.
Three-month averages, year-over-year comparisons, and cycle context can reduce the influence of a single volatile observation. They do not remove revisions or sampling error, but they discourage overreaction to monthly noise.
National and regional data can diverge from local conditions. Local employment, land availability, insurance costs, zoning, permit processing, migration, and existing inventory can produce a different pattern in a specific market.
Assume a monthly release shows the following illustrative results:
| Measure | Illustrative change | Initial interpretation |
|---|---|---|
| Building permits | -8% | Fewer units entered the authorization pipeline |
| Housing starts | -3% | Groundbreaking slowed modestly |
| Housing completions | +10% | Builders finished more units from earlier starts |
| Units under construction | -2% | Active construction inventory began to contract |
This pattern does not mean the numbers conflict. It can describe a pipeline moving from an earlier construction wave toward completion while fewer new units replace finished projects.
Before drawing a stronger conclusion, an analyst would ask:
The correct conclusion is conditional. The release suggests a softer forward pipeline only if the permit and start weakness persists and is not explained by sampling noise, revisions, or temporary timing effects.
Housing starts connect physical construction activity with several financial decisions.
Starts can provide context for future construction volume, but they are not the same as a public builder’s orders, backlog, closings, revenue, or margins. Company mix, geography, cancellation rates, incentives, land costs, and accounting recognition still determine results.
For a Construction Loan, a start can indicate that a project has moved into active work. It does not establish that lender conditions have been satisfied, costs remain within budget, draws are supported, liens are controlled, or completion is assured.
Single-family construction can eventually add homes that require purchase financing, while multifamily construction can create demand for development and permanent commercial financing. The timing from start to sale, occupancy, or loan conversion varies, so starts are an early pipeline indicator rather than an immediate measure of mortgage originations.
The Weekly Mortgage Applications Survey measures a different stage: mortgage application activity reported by participating lenders.
Residential construction affects spending on structures and related goods and services. Starts can help analysts assess construction momentum, but permits, construction spending, employment, completions, inventories, and sales provide necessary context.
Starts can signal units entering a future supply pipeline. They cannot determine future prices or rents without information about geography, unit type, completion timing, vacancies, household demand, financing conditions, and competing inventory.
Housing starts can respond to several overlapping forces:
These are possible influences, not a mechanical formula. A national release cannot identify the cause of a monthly change by itself.
Housing starts do not directly measure:
Starts are one data point in a housing-market framework, not a standalone valuation or investment signal.
Before using a release, verify:
Housing-starts data are educational market evidence, not personalized investment, lending, appraisal, legal, or tax advice. Verify the latest official release and use decision-specific local and financial information.