A decade ago, the then CEO of Zillow, Spencer Rascoff, sold his home in Seattle. Zestimate, Zillow's online home valuation tool, misvalued the property by an embarrassing margin, according to a GeekWire report from that time.
"Rascoff ... sold a home in the city’s Madison Park neighborhood for $1.05 million at the end of February," said GeekWire. "On March 1, the day after the sale, the Zestimate for the home reached $1.75 million. On Zillow’s website today [May 23, 2016], it’s come down to $1.575 million."
It's not fair today to highlight a 10-year-old, 40% miss by Zestimate. No doubt, Zillow has invested many millions since then in improving and refining its service. But, still today, Zillow says that Zestimate "is not an appraisal and can't be used in place of an appraisal."
And many real estate agents must be weary of arguing with clients who believe online home valuation tools before expert local opinion.
Online Home Valuations Lack Nuance
One agent currently raising questions about online automated home valuations (aka automated valuation models or AVMs) is Kerry Ramage, a managing broker and Realtor with RE/MAX Elite in Melbourne, Florida.
"Online home valuations have become a common starting point for many homeowners ...," Ramage wrote in HelloNation's recently published spring edition. "These automated estimates appear quick and convenient, and they offer a general sense of the current market.
"The challenge is that the tools rely on broad data rather than on-the-ground knowledge, so their numbers often shift by tens of thousands of dollars," Ramage continued. "I tell clients that an algorithm can estimate trends, but it cannot understand the small features that drive real value in a local neighborhood."
Megan Micco, a certified Green Realtor in Berkeley, California, concurs. "These models rely purely on historical data and never actually visit your property," she writes on her website.
"They’re backward-looking by design, using what sold yesterday to predict what might sell tomorrow, and while an algorithm can tell you that your home has three bedrooms, it cannot tell you that the primary suite has stunning morning Bay Bridge views that makes buyers fall in love."
That's where AVMs can misstep. They deal only in hard data.
While the buyers who make up a housing market may take notice of those number-based factors, they often focus on other things: condition, views, amenities and the features that make them fall in love with a home. And AVMs can't take those into account.
How AVMs work
Of course, Zestimate isn't the only AVM. Redfin Estimate is another big one, and there are plenty of others. Experian, one of the Big 3 credit bureaus, explains how they work:
"Each AVM uses its own statistical methods, but all analyze data from similar sources to arrive at a price estimate for a given property. A property's physical characteristics, including square footage, lot size, number and type of rooms, and amenities such as fireplaces, garages, patios and decks, are taken into consideration, as are recent sales figures for similar properties nearby. (How 'similar' and 'nearby' properties are defined is part of the statistical approach that distinguishes one AVM from another.)"
But Experian goes on to describe the things AVMs can't take into account. Those include:
- The property's condition — How well it's been maintained.
- Interior improvements — Does the home have refinished floors, new carpets, a finished basement, recently replaced appliances, and so on?
- Curb appeal — How attractive the home is when you pull up outside, including things like lot orientation, landscaping, and where it's located on its block.
How Accurate Are AVMs?
Within the limitations described above, AVMs can be pretty accurate. But even probably the most famous, Zestimate, acknowledges error rates.
"The nationwide median error rate for the Zestimate for on-market homes is 1.74%, while the Zestimate for off-market homes has a median error rate of 7.20%," it says.
So, those not selling their homes who check Zestimate to assess their net worth — much as they might look at their stock portfolios — may be seeing over- or underestimates of 7.2%. That's an average, so it could be appreciably more or less.
And that varies by metro area. In San Francisco, the most extreme example, nearly 8% of Zestimates are adrift by 10% or more of the actual sales price. And 1.43% are out by 20% or more, according to Zillow's figures.
Does All This Matter?
If someone is just idly wondering how much their home is worth, AVM error rates aren't important. That's especially true for those who understand what these algorithms miss, and can make mental adjustments.
The danger comes when people rely on the figures for practical purposes. That could happen when someone is pricing a home that they plan to market as "for sale by owner" (FSBO). They could understate the home's value and sell it for less than its worth. Or they could overstate it, and be left with a home that takes a long time to sell, usually after price reductions.
There could be similar dangers in estate planning. Someone wants to leave their home to one child, and to adjust the cash that child receives so as not to advantage or disadvantage their other kids. Relying on AVMs might foil that objective.
AVMs are a good place to start when valuing one's home. But the moment one needs to rely on a valuation is the time to access some local human expertise.
