Can a machine — even one loaded with sophisticated algorithms, cloud-computing technology and real-estate market data — accurately estimate the value of your home?
No way, you might say, it takes trained, professional appraisers to inspect and value houses, and even the best of them don’t always get it right. But real estate and technology entrepreneurs strongly believe that relatively accurate automated estimations of value are achievable, and they have been working for years to create systems that do just that.
You’ve probably heard of or used the best known — and most controversial — version to date: Zillow’s “Zestimate” model. It allows you to go online, enter an address and get a nearly instantaneous valuation, along with property data such as square footage, numbers of bedrooms, baths, lot size and the like. It’s available for millions of houses across the country, whether they are on the market for sale or not.
But Zestimates have provoked criticism by owners, sellers and realty agents who are upset at inaccuracies in value conclusions, property data and analysis of neighborhood price trends. Zillow itself acknowledges the problem, disclosing on its website that the “median error rate” nationwide is substantial — 7.9 percent. Error rates are measured by the difference between what the automated system says is the estimated value and the subsequent selling price of the property. Median Zestimate error rates in some exurban markets, however, range much higher — 10 percent to 20 percent and worse.
Never miss a local story.
100 million Number of homes in Zillow database. Number in Redfin database: 40 million
Now there’s a new, noteworthy competitor to Zestimates. Last week, national real estate brokerage Redfin rolled out its own proprietary model, claiming a median error rate of just 1.96 percent on homes listed for sale and 6.23 percent for all other houses. Dubbed the “Redfin Estimate,” it’s available free on 40 million homes in 35 major metropolitan markets.
Redfin CEO Glenn Kelman says his company’s model not only offers the “lowest published error rate” available, but taps into “a treasure trove of data” provided directly by multiple listing services around the country and uses advanced cloud-computing power.
“This gives Redfin information about homes that non-brokerage real estate websites don’t have, like whether a home has a water view or is located on a busy street,” he said.
The Redfin model also employs a weighting system keyed to market preferences for different home features and locations. For example, Kelman said, a waterfront location in the Seattle area may be much more highly valued than in other markets
As you might suspect, Redfin isn’t providing free online estimates purely out of a desire to further the general public’s knowledge. It’s in the business of representing buyers and sellers, and you can’t miss the tie-ins that accompany your Redfin Estimate. When you visit its site (www.redfin.com/redfin-estimate) and log in an address, it displays not only the home photo and value, but has multiple links to agents “who can discuss and refine estimates … as well as list the home for sale.” It’s all voluntary of course — no obligation — but you get the point.
7.9 percent Zestimate median error rate. Redfin median error rate: 1.9 percent on homes listed for sale, 6.23 percent for all other houses
Like Zillow, Redfin discloses median error rates. The lowest rates for estimates on active listings are in Colorado (1.41 percent), Virginia (1.58 percent), Washington state (1.65 percent) and Arizona (1.67 percent). In metropolitan Washington D.C., it’s 1.95 percent. Valuation error rates on houses not listed for sale run considerably higher, such as 10.67 percent in West Virginia, 8.46 percent in Pennsylvania and 8.2 percent in Florida.
So what does Zillow think about this new competition from Redfin? I spoke with Stan Humphries, chief economist and chief analytics officer, and he didn’t mince words. Redfin’s claims of superiority may be “great marketing but there’s not much reality to it,” he said.
Zestimates are available on 100 million homes, more than twice as large a database as Redfin’s, and its median error rate covers both listed and non-listed houses. Non-listed and non-urban homes inevitably are tougher to value, but Redfin limits its exposure by focusing on just 35 metropolitan areas. He scoffed at Redfin’s trumpeting of direct pipelines into Multiple Listing Service data and its use of cloud computing. Zillow uses listing information from MLSs and directly from brokers, and has employed cloud computing for years, said Humphries.
So what to make of this emerging spat between real estate valuation competitors? How about this: Try both systems on one or more houses — maybe your own and one or more you know well. See which performs better. It won’t cost you a cent to do test drives.
Kenneth Harney is executive director of the National Real Estate Development Center.