The Zestimate, Zillow’s home-value estimate algorithm, is controversial. People love it when their home’s Zestimate is higher than expected and dismiss it when the value falls short. But automated valuation models aren’t going away. In fact, Zestimates and other AVMs are getting smarter and more accurate as machine learning improves.
You have more control than you might think over Zillow’s accuracy. And now Zillow is betting a million bucks that humans can make the machines even smarter.
Are Zestimates accurate?
Today, Zillow says the margin of error is 4.3% nationwide. That means half of Zestimates were within 4.3% of the final selling price, and half were off by more than 4.3%.
So the machines are learning.
How do Zestimates work?
Zestimates are computed using mathematical equations based on publicly available information, including county records of past sales and valuations — even a home’s square footage. Zillow says “millions of data points” are in the secret sauce, including “road networks and neighborhood surroundings, such as views, parks and other amenities.”
And Zillow recently moved all of this data into the cloud, so that it can implement improvements more rapidly and compute Zestimates “in nearly real time.”
Teaching the learning machines
Zestimates are a lot more accurate than tomorrow’s weather forecast. Forget predicting rain; meteorologists get the next day’s high temperature wrong 20% of the time. So a 4% error rate looks pretty good in comparison.
Zillow has put up a million dollars to inspire the ‘brightest scientific minds to improve the Zestimate.’
Yet Zillow has put up a million dollars to inspire the “brightest scientific minds to improve the Zestimate.” The competition, which began in May and recently closed its initial qualifying round, has drawn more than 76,000 entries from 4,350 participants in 78 countries. It’s one of the most popular machine learning contests ever, according to Kaggle, the online platform hosting the competition.
The winning team must develop an algorithm that beats not only Zillow’s current accuracy but also has an error rate lower than that of any other competitor’s formula.
The contest continues through early 2019.
Zestimates are not appraisals
No matter how intelligent automated valuation models grow to be, they aren’t appraisals. They can’t walk through your home, drive through the neighborhood or see the upgrades you’ve made.
And because of the availability of data — or lack of it — AVM accuracy varies by geography. Zillow states that very clearly on its website.
“In some areas, we might not be able to produce a Zestimate at all,” Zillow says on its “How Accurate Is Zillow’s Zestimate?” page. The national median error rate is 4.3%, but it ranges locally from a low of 2.8% in Washington, D.C., to a high of 9% in Dallas-Fort Worth. Zestimates are flat-out unavailable in Houston and St. Louis, among other cities. In fact, numerous counties in some states don’t have the public data necessary to formulate a Zestimate.
AVMs are most accurate in neighborhoods where homes are similar. Master-planned communities are easier for Zestimates to value because they have similar structures, costs and improvements, says Antonio Gonzales, a real estate investor and broker in San Diego. Often there is little variation in views and amenities, so values are more consistent.
Coastal properties and transitioning neighborhoods can be harder for AVMs to value.
“There are certain neighborhoods where, because of the school district or just the development of the community, [you] can go a couple of blocks and you see a drastic change in the prices of homes,” Gonzales says. “Especially in areas that are gentrified.”
How to make your home’s Zestimate more accurate
A couple in Illinois were so dissatisfied with their Zestimates that they proposed a class action lawsuit against Zillow Group. They were trying to sell two properties and the Zestimates for both were well below their listing prices. The judge dismissed the case in August 2017, accepting Zillow’s claims that Zestimates are simply “a starting point in determining a home’s value,” according to press reports.
Thing is, you can “claim your home” on Zillow and correct factual errors. If the Zestimate is based on an incorrect square footage, or the wrong number of bedrooms or bathrooms, you can fix that — and the correction might boost your Zestimate. And it might not. Probably worth a try, though.