The Zestimate that appears on every listing on Zillow is about to get a little closer to the expected sales price of a house, and all the online real estate giant had to do is give away more than $1 million.
More than 18 months ago, Zillow launched a contest it called the “Zillow Prize.” The goal of the contest was to see if anyone could create an algorithm that could beat the one powering Zillow’s “Zestimate,” the online real estate giant’s property value estimation tool.
If anyone could beat Zillow’s benchmark model, they’d score themselves $1 million.
And Wednesday, Zillow announced that a team of amateurs succeeded in creating a model that bested the Zestimate.
According to Zillow, the winning team beat the Zillow benchmark model by approximately 13%. The winning team includes data scientists and engineers from around the world: Chahhou Mohamed of Morocco, Jordan Meyer of the United States, and Nima Shahbazi of Canada.
As a result of Mohamed, Meyer, and Shahbazi’s efforts to beat the Zestimate, the team will be awarded $1 million.
Beyond that, Zillow also awarded $100,000 to the second-place team, and $50,000 to the third-place team.
But Zillow is hardly chalking this exercise up as a loss. Just the opposite, in fact.
Zillow said that it will incorporate parts of the winning team’s model, along with other contest entries, to improve the accuracy of the Zestimate that appears on the listings for 110 million homes on Zillow.
According to Zillow, the improvements to the Zestimate will decrease its current nationwide error rate of 4.5% to less than 4%, meaning that half of all Zestimates will be within 4% of the selling price, and half will be off by more than 4%.
Put another way, Zillow claims that on average, the Zestimate is $10,000 off of the actual sale price for a typical home, but with the improvements from the contest, future Zestimates could be approximately $1,300 closer to the sale price.
The Zestimate has long served as a contentious issue between real estate professionals and consumers.
While Zillow describes the Zestimate as a “great starting point” for determining the value of a home, homebuyers and sellers often believe that the Zestimate listed on a home is the true market value of the home.
In Zillow’s defense, it has dramatically improved the accuracy of the Zestimate over time, measured by how close the Zestimate is to the eventual sale price of a home, from 14% in 2006 to its current level of 4.5%.
And soon, that error rate will be less than 4%.
“People are incredibly passionate about their home and understanding its value, and we are amazed by the winning team’s hard work the past two years to make the Zestimate even more precise,” said Stan Humphries, chief analytics officer and creator of the Zestimate.
“We’ve been on a 13-year journey making the Zestimate more accurate, and hosting Zillow Prize allowed us to invite thousands of brilliant data scientists from around the world to join us on this journey,” Humphries continued. “We’re so proud that the winning team’s huge achievement, and the work of all the teams in the competition, will provide millions of homeowners with a better understanding of one of their biggest life investments.”
According to Zillow, the winning team’s algorithm utilized several “sophisticated machine learning techniques, including using deep neural networks to directly estimate home values and remove outlier data points that fed into their algorithm.”
Additionally, the winning team made use of publicly-available, external data including rental rates, commute times, and home prices, among other types of contextual information, like road noise, all of which are variables that factor into a home’s estimated value.
“It’s amazing to know that millions of people will benefit from our ideas,” Shahbazi said. “We brought every novel idea we could to our code and kept experimenting. For every idea that worked, there were a hundred that didn't work. But we kept going.”
All in all, more than 3,800 teams from 91 countries took part in the competition to improve the Zestimate, but only one won.
