How to read US house price measures: FHFA index and Census medians

In this article

Written by

Kooky

Builder of Shaka, the payment router that pays every agent their commission on closing date.

About Kooky and Shaka →

Within five days of each other in late September 2026, two federal releases appeared to describe two different countries. On 24 September the Census Bureau and the Department of Housing and Urban Development reported that the median sales price of a new house sold in the United States in August 2026 was 5.8 per cent lower than a year earlier. On 29 September the Federal Housing Finance Agency reported that its House Price Index was 2.6 per cent higher in July 2026 than in July 2025.

Neither release is wrong, and neither contradicts the other. They measure different homes, at different moments of a sale, with different arithmetic, and one of them is a survey estimate with a wide margin around it. This guide goes through the official measures one at a time: the three main versions of the FHFA index and the loans each is built from, the way a repeat-sales index is calculated, how long the data take to arrive and why past figures change, the median and average prices in the new home sales release, and the separate constant-quality index the Census Bureau publishes. It gives the latest figure each official page shows, with its date.

FHFA monthly House Price Index report of 29 September 2026; Census Bureau and HUD new residential sales release of 24 September 2026.

Two kinds of number: a change and a level

The first thing to settle about any house price figure is whether it is a level or a change.

The Census Bureau and HUD release gives levels: an amount in dollars, the median or the average price of the new houses sold in a month. A change can be worked out from two levels, and the release does so, but the starting point is a price tag.

The FHFA House Price Index gives no price tag at all. The agency's frequently asked questions, last updated on 7 October 2026, state that the index measures neither the mean nor the median price. It is an index number, and the only thing it is built to report is how much prices have moved between two periods. The same page points readers who want a price level towards the Census Bureau and the National Association of Realtors as alternative sources.

A level depends on which houses happened to sell in the month. An index of the FHFA type is designed, in the agency's words, to isolate price changes from changes in the mix of properties.

What the FHFA index is built from

The FHFA describes its index as a broad measure of single-family house price movements in the United States. It measures the average price change between sales or refinancings of the same properties over time.

The raw material is mortgage data. According to the FHFA, the index uses single-family properties whose mortgages have been purchased or securitised by Fannie Mae or Freddie Mac, the two companies the agency calls the Enterprises, with records going back to January 1975. The FHFA has access to these records as the regulator of the two companies. Its data page speaks of tens of millions of home sales; its question page puts the total at more than 100 million sale and refinance transactions.

The agency is plain about the limit of this source: the Enterprise data do not cover all residential mortgages. For the flagship index, a loan must be both conforming and conventional. The FHFA defines conforming as a loan that meets the Enterprises' guidelines and does not exceed the conforming loan limit, and conventional as a loan that is not insured or guaranteed by the Federal Housing Administration, the Department of Veterans Affairs or another federal entity. Loans that fail either test are removed. A house bought with an FHA or VA loan, or with a loan above the limit, therefore leaves no trace in the flagship index, and neither does a house bought with cash.

Property type narrows the sample again. The FHFA states that it excludes condominiums, cooperatives, multi-unit properties and planned unit developments. It also filters out records that look like errors: extreme values, implausible appreciation, repeat sales inside the same period, and records with incomplete dates or addresses.

Distressed sales are kept. Short sales and bank-owned sales are included in every FHFA index except one, the Distress-Free index, introduced in the second quarter of 2012.

Purchase-only, all-transactions, expanded-data

The name "FHFA House Price Index" covers a family of indexes. Three matter most to a general reader, and they differ in which price observations they accept.

The purchase-only index is the one in the headlines. The FHFA names its seasonally adjusted purchase-only index as the flagship, and it is the standard index of both the monthly and the quarterly news releases. Every observation in it is a price a buyer agreed to pay.

The all-transactions index adds appraisal values to those sale prices, which brings refinancings into the calculation. A refinancing involves no sale: the value comes from an appraiser, not from a buyer and a seller. The FHFA publishes 410 metropolitan indexes in its releases, made up of 373 metropolitan statistical areas and 37 metropolitan divisions, while the purchase-only index is published for the 100 most populous of them.

The expanded-data index goes in the opposite direction: it stays with real sales but widens the pool of them. The FHFA explains that it adds purchase mortgages from other sources, conforming and non-conforming, together with cash sales. The sources are records of FHA-endorsed mortgages and county recorder data from a licensed provider, which the agency says cover most counties but not all. Its data page adds one limit: the added transactions are those with sale prices below the annual loan limit ceiling.

The base period is a small trap. According to the FHFA, most of its series are set to 100 in the first quarter of 1991. The all-transactions series keep older bases: 100 in the first quarter of 1995 for metropolitan areas, and 100 in the first quarter of 1980 for states and census divisions. Two index levels from different versions cannot be laid side by side. Only the percentage changes can.

For that reason the FHFA asks anyone quoting the index to say which version is meant and whether it is seasonally adjusted, in the form "Purchase-Only, seasonally adjusted, nominal". The last word matters too: the index is nominal, which means inflation has not been taken out of it.

How a repeat-sales index works

A repeat-sales index starts from pairs. The FHFA merges the monthly files it receives from Fannie Mae and Freddie Mac and looks for properties that appear more than once. Each property with two recorded transactions yields one price differential: what the same house was worth at two dates.

A worked example, with invented figures: a house sold for US$300,000 in one year and the same house sold for US$390,000 seven years later. The pair says that this house rose by US$90,000, or 30 per cent, over the interval. It says nothing about the price of any other house. An index is produced by combining a very large number of such pairs, each covering its own interval, into one series that fits them best.

The FHFA describes its method as a modified version of the Case-Shiller methodology, using a geometric weighted repeat-sales procedure. Transactions are grouped by month for the monthly index and by quarter for the quarterly index. Because every observation compares a house with itself, the agency says the result is commonly described as a constant-quality index.

The national figure is assembled from the bottom up. The FHFA builds each of the nine census divisions from the growth rates of its states, weighted, and the national index from the divisions. The weights are estimates of each state's share of the country's stock of one-unit detached houses. They are shares of houses, not shares of value or of sales.

One detail of vocabulary avoids a common misreading. In the quarterly report, a "four-quarter" change compares a quarter with the same quarter a year before. The FHFA notes that all the transactions of a quarter are pooled, so the figure is not a comparison between the last day of one quarter and the last day of another.

From closing table to release: lag and revisions

The date attached to each observation is the date the loan was originated, not the date the Enterprise acquired it. Between that date and publication there is a chain of delays, which the FHFA sets out in its answers.

The calendar shows the result. The monthly report published on 29 September 2026 carried data for July. The next one is scheduled for Tuesday 27 October 2026 with data to August, according to the FHFA's release calendar, followed on Tuesday 24 November by the quarterly report for the third quarter with the September monthly tables, and on Tuesday 29 December by the monthly report for October.

Past figures are not fixed, and the FHFA gives three reasons. First, new repeat transactions: a sale recorded today creates a pair with an earlier sale of the same house, and that pair adds information about every period between the two dates. Second, the Enterprises buy seasoned loans, mortgages originated some time ago, whose records enter the data late. Third, the timing with which data become available.

The size of these changes can be read in the reports themselves, which print the earlier estimate next to the current one. In the quarterly report of 25 August 2026, the national change from January to February 2026 was revised from 0.0 per cent to a fall of 0.1 per cent, and the change from December 2025 to January 2026 from a rise of 0.2 per cent to a rise of 0.3 per cent.

What the latest FHFA reports show

The most recent FHFA figures at the date of this guide come from two reports.

The monthly report of 29 September 2026 gives the seasonally adjusted index for July 2026: up 0.3 per cent from June, and up 2.6 per cent from July 2025. The June figure, a change of 0.0 per cent, was left as first published. Across the nine census divisions the monthly change ran from a fall of 0.8 per cent in the Mountain division to a rise of 1.5 per cent in the Middle Atlantic division. Over twelve months the same two divisions marked the range: 0.6 per cent in the Mountain division and 6.3 per cent in the Middle Atlantic.

The quarterly report of 25 August 2026 covers the second quarter. On the seasonally adjusted purchase-only index, national prices were 2.1 per cent higher than in the second quarter of 2025 and 0.3 per cent higher than in the first quarter of 2026; the report gives the quarterly change as 0.35 per cent, or 1.39 per cent at an annualised rate. Prices rose over the year in 46 states and the District of Columbia and fell in four: New Mexico (1.25 per cent), Washington (0.94 per cent), Colorado (0.53 per cent) and California (0.20 per cent). Alaska, at 8.3 per cent, and Vermont, at 7.3 per cent, led the states. Among the 100 largest metropolitan areas, prices rose in 76; the range ran from a rise of 7.73 per cent in Elgin, Illinois, to a fall of 3.73 per cent in Everett, Washington.

The same report shows how two honest figures for nearly the same period can differ. The quarter-on-year-earlier change was 2.1 per cent. The monthly table in that report, comparing June 2026 with June 2025, gives 2.3 per cent. One pools three months against three months; the other compares single months.

New home prices: what the Census Bureau and HUD count

The new residential sales release is published jointly by the Census Bureau and HUD, and it describes a different market by a different route. Its sales count concerns new single-family houses, and its prices are those of new houses sold.

Three definitions come from the Census Bureau's Survey of Construction definitions page; the release itself, as read for this guide, does not name the survey its figures come from.

The first is the sale. A house is considered sold when a sales contract has been signed or a deposit accepted. Houses sold before construction has started are counted, and the survey does not follow the transaction to closing. The FHFA index, by contrast, dates a transaction by the origination of the loan. The two measures are therefore looking at different moments in the life of a deal, the agreement in one case and the financing in the other.

The second is the sales price: the price agreed between purchaser and seller, including the improved lot. The Census Bureau excludes adjustments made later through change orders, and extras or options the purchaser pays for in cash.

The third is the kind of house. The Census Bureau defines houses built for sale as houses built on the builder's land with the intention of selling house and land in one transaction. Houses built on the owner's own land, whether by a general contractor or by the owner acting as contractor, are separate categories; for contractor-built houses the Census Bureau records a contract price that does not include the land. Single-family, in these statistics, takes in detached and semi-detached houses, row houses and townhouses.

The comparison with the FHFA sample is direct. The flagship FHFA index needs a second transaction on the same property and a conforming, conventional loan. The Census definitions, for their part, list conventional, FHA and VA loans as types of financing.

The release gives two summaries of the prices. The median, in the Census Bureau's definition, is the middle point of all reported values, with half above and half below. The average is the arithmetic mean, the sum of the values divided by the number of houses reporting.

The latest new home figures and their margins

The release of 24 September 2026 covers August 2026. Sales of new single-family houses ran at a seasonally adjusted annual rate of 684,000, 6.4 per cent above the July rate of 643,000 and 2.0 per cent below the 698,000 of August 2025.

In one month, from July to August 2026, the median rose 0.4 per cent while the average fell 9.1 per cent. The two summaries of the same set of sales moved in opposite directions. And the average sat US$85,000 above the median in August 2026, against US$134,200 in July: an arithmetic mean is pulled by the most expensive houses in the month, and the median is not.

Then there is the margin printed beside each change. The release attaches a plus-or-minus figure to each change. The 0.4 per cent monthly rise in the median carries a margin of 7.4 percentage points either way, and the 5.8 per cent fall over the year a margin of 8.2 points. Taken at face value, a fall of 5.8 with 8.2 either side spans everything from a fall of 14.0 per cent to a rise of 2.4 per cent. The 6.4 per cent monthly rise in the sales rate carries a margin of 19.5 points. The release page does not explain these margins or state a confidence level.

Why a median can fall when no price has

A median describes the houses that sold. If a different set of houses sells the following month, the median changes even when no individual house is worth a dollar more or less. This is the "mix" the FHFA refers to when it says its index is designed to separate price change from changes in the properties sold.

A worked example with invented figures makes the mechanism visible. Assume a builder with a fixed price list that does not change between two months.

In the first month five houses sell, at US$300,000, US$350,000, US$400,000, US$450,000 and US$500,000. The median is the third value, US$400,000. The average is the total of US$2,000,000 divided by five, also US$400,000.

In the second month five houses sell again from the same unchanged price list, but the selection is different: US$300,000, US$320,000, US$350,000, US$400,000 and one large house at US$900,000. The median is now US$350,000, a fall of 12.5 per cent. The average is US$2,270,000 divided by five, or US$454,000, a rise of 13.5 per cent.

No price has moved, yet the median says prices fell by an eighth and the average says they rose by more than an eighth.

A median tells you what was bought this month. A repeat-sales index tells you what happened to the price of what was bought before.

The Census constant-quality index

The Census Bureau publishes its own answer to the mix problem, separately from the monthly sales release: a set of price indexes for new single-family houses, on a page sponsored by HUD.

According to that page, there are indexes for single-family houses sold and for single-family houses under construction, both based on the Survey of Construction. The index for houses sold incorporates the value of the land, in line with the sales price definition, which includes the improved lot. It is published quarterly for the country and annually for the regions. The index for houses under construction is published monthly at national level.

The current series uses 2005 as its base year, and an earlier historical series uses 1996. The Census Bureau's question page for these indexes, last revised on 2 September 2026, refers to two documents, one on general price index questions and one on the 2005 base year series.

Reading the measures side by side

Put together, the official measures differ on points that each page states for itself.

Three readings follow from the sources.

The two headline figures of late September 2026 do not cover the same month. The FHFA figure is for July and the Census figure for August, because the FHFA receives loan records about two months after origination. Comparing the releases of one week means comparing different periods.

Only the Census figure carries a printed margin. The median's yearly change of 5.8 per cent comes with 8.2 points either way. The FHFA reports read for this guide print no such margin; their uncertainty shows up instead as later revisions.

Geography changes the answer as well. Inside the FHFA's own July 2026 report, twelve-month growth ran from 0.6 per cent in the Mountain division to 6.3 per cent in the Middle Atlantic. A national figure of 2.6 per cent describes neither.

Which measure suits a particular question, a valuation, a listing or a market report, depends on the homes and the period that question is about.

Kooky, from Shaka

Kooky edits Agents Estate and builds Shaka, the payment router he made for real estate professionals. One payment comes in, and every agent, agency and party in the deal receives their signed share on closing date.

About Kooky and Shaka →