Every social post image for a national market carries the line “Market stats provided by Redfin and refined by AgentIntel”, and a national market page ends with “Data sourced from Freddie Mac, Indeed Hiring Lab and Redfin Data Center then refined at AgentIntel”. Redfin publishes the figures. AgentIntel puts every place on the same footing, builds the periods Redfin doesn’t publish, and fills in what Redfin leaves out. The same figures are behind national market pages and reports. This page says which numbers are Redfin’s as published, which ones AgentIntel calculates and why, and how those calculations were checked.
What each statistic means is in National statistics.
What Redfin provides
Redfin publishes its Data Center files once a month. For every place it covers, from the whole country down to zip codes and neighborhoods, and for each property type, there is one row per period with the figures a national market page shows: sales, pending sales, new listings, homes for sale, median prices, days on market and the negotiation percentages. Redfin publishes these summary figures only, never the sales behind them.
Three things about the files shape everything AgentIntel does with them.
- The country, states, metros and counties get calendar months. Cities, zip codes and neighborhoods get three-month periods, one ending each month: January through March, then February through April, and so on. Redfin publishes no single-month figure for a zip code.
- Some figures are seasonally adjusted. For the country, the states and about 100 larger metros, Redfin smooths out the usual swing of the seasons before publishing. Counties, cities, zip codes and neighborhoods are published as they happened.
- Each figure covers its own period only. There is no twelve-month figure for a county in the files, and no figure for last month alone in a zip code.
Straight from Redfin
For a single month on a country, state, metro or county market, and for a three-month period on a city, zip code or neighborhood market, every statistic is Redfin’s own figure for that period: Median Sales Price, Median Price per Sqft, Median Asking Price, Median Asking Price per Sqft, Median Days on Market, Contract in First Two Weeks, Asking vs Sales Price, Sold Above Original Asking Price, Months of Inventory, New Listings, Active Listings, Total Active Listings, Pending Listings and Total Sales. The one exception is a place Redfin seasonally adjusts, where the figure is restored to what happened, as described below.
Two things on a market page are never Redfin’s. Appreciation is AgentIntel’s own calculation. And the change shown beside a figure compares the period you chose with the one before it, worked out from the same figures; Redfin’s own month-over-month and year-over-year changes aren’t used.
What AgentIntel calculates
The seasonal adjustment is removed
Where Redfin publishes a seasonally adjusted figure, AgentIntel shows the figure for what happened instead.
Why. AgentIntel is for comparing places with each other and with their own past. A seasonally adjusted state line next to a county’s actual figures would show the county falling behind every winter and racing ahead every spring, when it was only following the seasons like the state around it. The Perspective map, the rankings and a report’s charts only mean something when every place is measured the same way.
How. Redfin has published both the adjusted and the unadjusted version of its history for the country and many metros, so the size of its adjustment in each calendar month is known. From that, AgentIntel keeps a factor for every place, property type, statistic and month that undoes the adjustment. Where Redfin never published both versions, the factors come from the counties that make up the metro or state, which Redfin never adjusts, and from Redfin’s weekly files, which are unadjusted. The factors are refreshed with every monthly release.
Longer periods are built from the months
A national market page offers Last 3 Months, Last Quarter, Last 12 Months and Last Year, and Redfin publishes none of them. AgentIntel builds each from the months in the period, or for a city, zip code or neighborhood from the four non-overlapping three-month periods in the year.
- Total Sales, Pending Listings and New Listings are added up.
- Active Listings is the count at the end of the period, as it is for a month.
- The median prices, Median Days on Market, Months of Inventory and the percentages are averaged. A 12-month Median Sales Price is the average of twelve monthly medians.
- Total Active Listings is left out. Homes for sale at any point in a year can’t be counted from monthly figures, so Indicators shows it only for the period Redfin publishes it for.
Why averaged medians. Redfin never publishes the sales themselves, so the true median of a year’s sales can’t be known. The average of the monthly medians is the closest figure there is, and it moves with the market the same way. A period with a gap in Redfin’s history shows no figure rather than one built from fewer months.
Cities, zip codes and neighborhoods get a monthly figure
Redfin publishes three-month periods for these places, and a market page shows Last Month for them too. The three counts, Total Sales, Pending Listings and New Listings, are split into their months. Every other statistic, including the prices, Median Days on Market and the Active count beside the gauge, is Redfin’s three-month figure unchanged, because a median or a percentage can’t be split.
How. Each of Redfin’s three-month figures is the sum of three months, and each month sits inside three of Redfin’s periods. AgentIntel finds the monthly figures that add up to Redfin’s periods and follow the month-by-month pattern of the place’s county, which Redfin publishes monthly. If the county’s sales ran 30% higher in March than in February, the zip code’s March is placed above its February by about as much, and its three-month totals still match Redfin’s. Redfin’s periods don’t always add up exactly, so where they disagree the difference stays in the months around it rather than carrying through the whole history. A place without county figures has each period split evenly.
Why the county. It is the smallest place Redfin publishes monthly, and the same months are busy or quiet across a county. Splitting each period evenly would show every zip code with the same sales in January as in March.
Appreciation
AgentIntel’s own figure: the percent change between the Median Sales Price for the twelve months ending with the period and the twelve months before it, the same way MLS markets measure it. A single month’s year-over-year change swings with which homes happened to sell that month, especially in a smaller market.
Which places get a market
A place gets a national market, and a property type gets its own version of it, when Redfin reports at least 30 sales there in the latest year and five years of monthly history. That is rechecked with every release. A place’s history is used back to the first gap in Redfin’s reporting, and a month in which Redfin published an impossible figure, such as a ten-billion-dollar asking price, is left out.
How the calculations were checked
Removing the seasonal adjustment
Each way of working out the factors was tested against the places where Redfin’s own adjustment is known from both versions of its history. The factors built from county figures matched within about 1% for prices and ratios, within about half a day for Median Days on Market, and within about a point for Sold Above Original Asking Price. For the counts, the adjustment is read directly from Redfin’s county figures, which are never adjusted and add up to the metro or state. Every release, each place’s new factors are compared with its previous ones, and a large move raises an alert.
Splitting three-month periods into months
The split was checked against MLS records, which count every sale by the month it closed, for zip codes covered by both Redfin and an MLS AgentIntel works with: Unlock MLS in Austin, NTREIS in Dallas–Fort Worth, SCWMLS in south-central Wisconsin and CRMLS in California. The check covers the last three years in the 616 zip codes with at least 30 sales in a month, about 10,500 zip-months in all.
- Total Sales landed within 3.6% of the MLS’s own count in a typical month, and within 10% in 85% of months. Redfin’s own three-month figure is within 2.3% of the MLS’s three-month count, so most of that gap is the two sources counting different homes, not the split.
- Pending Listings and New Listings are counted differently by Redfin and an MLS, so even their three-month figures differ by about a tenth. The monthly figures stay about as close to the MLS as the three-month ones.
- Dividing each period evenly instead would have put Total Sales within 10% in fewer than half of the months.
Definitions
Redfin’s published definitions were checked against MLS records for the same places and months. Where the data and the definition disagreed, National statistics follows the data: Sold Above Original Asking Price is measured against the original asking price, not the final one, and Pending Listings counts every home that went under contract in the period, however long it had been listed.
MLS figures against national figures
Where an MLS Connection and Redfin cover the same county or zip code, AgentIntel compares its MLS statistics with its national statistics after any change to how either is calculated. The two never match exactly, because they count different homes, as MLS and national data explains. The check is that the ratio between them stays steady from place to place and month to month; a ratio that moves points to a calculation to look into.