US Census New Residential Sales - New Single-Family Houses Sold, Priced and Inventoried

Rows verified during the August 2026 research pass, exactly as they land in your warehouse - counts in thousands of units, prices in nominal dollars.

series              | section                      | year | united_states_k | northeast | midwest | south | west
houses sold         | Annual (not seasonally adj.) | 1963 | 560             | 87        | 134     | 199   | 141
houses sold         | Annual (not seasonally adj.) | 1967 | 487             | 77        | 112     | 170   | 128

series              | section | year | median_price_usd | average_price_usd
sales price         | Annual  | 1963 | 18000            | 19300
sales price         | Annual  | 1968 | 24700            | 26600

Two things jump out even in four rows. First, the regional columns already tell a story in 1963: the South alone absorbed 199 thousand of the nation's 560 thousand new single-family sales, a share of the market it has never surrendered since. Second, the gap between median and average price - 18,000 versus 19,300 dollars in 1963, 24,700 versus 26,600 in 1968 - is the distribution skew made visible: mean sits above median whenever a tail of expensive homes drags it up, so tracking the spread is a crude but honest luxury-segment gauge.

Those rows come from just two of the series cuts. The full program adds houses for sale, sold and for-sale inventory by stage of construction, price-range distributions, regional prices, and the financing splits - ten distinct historical series in all, each a different lens on the same new-home market.

Eight documented fields cover the core record: one period identifier, the national value column, the four-region breakdown, and the price, price-band, construction-stage and financing dimensions. Definitions below are verified against the source card, not inferred.

Geography: United States national totals plus the four Census regions - Northeast, Midwest, South and West. Regional detail runs deeper than counts: prices by region, price ranges by region, and the sold-by-financing cut carries regional splits as well. What you will not find below the region line is sub-state geography - this is deliberately a national-and-region program, the complement to permit surveys that go county-deep.

Temporal: annual series stretch from 1963 to the present - six decades spanning every housing cycle on record - with monthly detail available in recent years. Price-range distributions start in 1999 and survive in three band vintages, because the dollar brackets had to move as home prices did.

Granularity: monthly and annual counts and prices at national and regional level, with quarterly price-by-financing cross-tabs layered on top. Each row is a measure-period-geography cell rather than a wide dashboard extract, so any aggregation - trailing-twelve-month sales, region share shifts, median-versus-mean spreads - is one groupby rather than a hand-stitched spreadsheet exercise.

API, files, or your warehouse. Daily, weekly, or hourly.

• Builder pricing strategy - median and average price series by region anchor spec-home price positioning, and price-range distributions show which dollar bands are actually clearing versus sitting in finished inventory.
• Housing-cycle macro research - new-home sales is one of the classic coincident-to-leading indicators for residential investment, and the 1963-onward annual spine gives every modern month a six-decade context line.
• Inventory and absorption analysis - for-sale counts crossed with stage of construction yield months-of-supply style reads: not-started backlog signals forward commitment, completed stock signals unsold risk standing in the ground.
• Mortgage and rate-sensitive modeling - the FHA/VA/conventional financing splits quantify how federal loan programs pulse with rate cycles, and the quarterly price-and-financing cross-tab ties buyer credit type to price band.
• Building-products demand forecasting - each sale today is a fittings, flooring and appliance install over the next two quarters; regional sales counts turn into shipment forecasts by territory.
• Land acquisition diligence - regional absorption trends feed submarket scoring before land bids, especially when paired with permit counts to see whether sales are outrunning new supply.

Investors and quants get the cleanest long-run read on new-home demand, with a distribution view (price bands, financing mix) that single headline numbers throw away. Homebuilding executives get their own industry's scoreboard: what sold, at what price, in whose region, financed how. Market researchers and consultants get citable official figures behind every housing-market sizing deck, reaching back further than any commercial panel. Data scientists and ML engineers get a small, fixed-schema, measure-period-geography table that drops straight into feature stores without parsing gymnastics. Policy analysts and journalists get the receipt for affordability arguments - the actual price bands buyers cleared, not list-price anecdotes.

Three things worth knowing upfront. First, vintages matter: figures are subject to revision as later releases incorporate additional survey returns, and some releases bundle initial estimates for two months in one drop - pin the vintage your models ingest, exactly as you would for any federal series.

Second, the price bands moved in 2024. The May 2024 update re-cut the Table 2 sales-price groupings to match current home-price distributions, recalculating 2020-through-March-2024 on the new bands and archiving 2002-2019 under the former ones. A naive concatenation of band histories produces a step change that is entirely definitional; join on band vintage, not just on band label. The workbook-format migration that started with the January 2024 release is cosmetic by comparison, but worth knowing when older extracts surface.

Third, mind the edges of scope: the program covers new, privately-owned single-family units only - no resales - and months-of-supply style ratios appear in release tables rather than as a native column of the historical series, so we derive them from sold-plus-for-sale counts on request rather than pretending they were always there.