海角大神

Zillow鈥檚 Zestimates, the machine-made home value assessments, put听Zillow听on the map. Without these controversial algorithm-produced appraisals, Zillow would be a home search engine just like all the others. By and large, real estate agents hate them and homeowners swear by them 鈥� or at them 鈥� depending on how their home value goes.

How accurate are Zestimates? Here鈥檚 what we found.

Found the perfect house? Now find the best home loan with NerdWallet鈥檚听mortgage rate听tool.

Zestimates: Improving Accuracy听

Zillow says Zestimates currently have a median error rate of 8%. That may seem to be a large margin 鈥� but it has improved. The algorithm began with a 13.6% median error rate in 2006, according to Zillow.

However, error rates can vary widely, depending on where you live. In the top metros, Zillow says accuracy ranges from a median error rate of 5% in Washington, D.C. to 9.5% in Pittsburgh. But in four cities 鈥� Houston, Kansas City, San Antonio and St. Louis 鈥� Zillow is unable to produce a Zestimate at all.

How Zillow Zestimates are computed

Zillow doesn鈥檛 reveal much about what鈥檚 inside the black box of Zestimates, other than to say that the value is calculated three times a week based on 鈥渕illions of public and user-submitted data points.鈥�

However the Zestimates are created, Zillow says between major upgrades new iterations of the model are constantly being devised. And since Zestimates were introduced on the site in 2006, the company has deployed three completely new versions of the algorithm, in 2006, 2008 and 2011.

鈥淗ome characteristics, such as square footage, location or the number of bathrooms, are given different weights according to their influence on home sale prices in each specific geography over a specific period of time,鈥� Zillow says.

The site also mixes in property tax information and prior sale prices for a home, as well as for others nearby.

But the accuracy of the home value estimates is hotly debated. Zillow says it calculates the accuracy of its 鈥減roprietary automated valuation models,鈥� or AVMs, by taking a home鈥檚 final sale price and comparing it to the Zestimate on or before the sale date.

More robo home values are on the way

The Zestimate is now just one of many AVMs available on home search sites and in the real estate industry. The old saying, 鈥淚f you can鈥檛 beat them, join them鈥� seems to apply. Developers are frantically trying to out-do one another by cobbling together new variations of black-box home price estimates.

But there are several areas听where AVMs usually fall short:

• AVMs can鈥檛 determine the actual condition of a house. Has it been recently renovated 鈥� or is it in dire need of an upgrade?
• AVMs account for the number of bedrooms, baths and the total square footage, but these don鈥檛 always correspond to a higher or lower price: Sometimes a bathroom is poorly located, or a bedroom is tiny.
• AVMs听don鈥檛 usually take into account title defects, permitted versus unpermitted structures, and other property nuances.
• Physical characteristics that impact local demand and value 鈥� such as architectural style, year built, and layout and flow of a home 鈥� cannot be factored in.

Zestimates: Better high than low

To see a Zestimate at work, consider the fact that in July of this year Zillow Group CEO Spencer Rascoff listed his听four-bedroom home in Seattle, Washington, for $1.295 million. At the time, the Zillow听Zestimate valued the home at about $1.39 million. That鈥檚 over 7% higher than听the list price, but within the Zillow median margin of error.

Gordon Stephenson, the listing agent for Rascoff, told industry publication听Inman听that Zillow听probably overestimated the value of Rascoff鈥檚 home because 鈥渋ts algorithm might not have听accounted for the home鈥檚 unique floor plan.鈥�

However, he added that if the听Zestimate of the home had been lower than its listing price, buyers might have been听inclined to think the listing was overpriced. But since the Zestimate was higher than the asking price, buyers might think the home was 鈥渁 bargain.鈥�

鈥淚t鈥檚 certainly better for us as a seller to听be $100,000听below the Zestimate than $100,000听above the Zestimate,鈥� Stephenson told Inman. That means there鈥檚 鈥渁 lot less explaining听to do.鈥�

鈥淢y hope is that the public is getting pretty smart about how Zestimates actually relate to home values,鈥� Stephenson added.

Since then, the home鈥檚 price has been discounted $96,000. Meanwhile, the Zestimate has听risen听by $185,000 to $1,575,206. Go figure.

听Hal Bundrick is a staff writer at NerdWallet, a personal finance website. Email:hal@nerdwallet.com. Twitter:听@halmbundrick

This article first appeared in NerdWallet.听