A crazy 'Zestimate' history on Baltimore house
But Wonk reader Hilary had a much stranger experience. The new algorithm, applied to the last five years of data, turned her Bolton Hill home's Zestimate into a roller coaster that makes the housing bust look like a kiddie ride.
"I count eight months in which it either gained or lost over $100k, and a couple more that were close to that," Hilary wrote me. "In December 2008, when the financial world was imploding and the housing market seemed to be in freefall, Zillow's new algorithm has my house's value going up from $493k to $671k, and then to $732k the next month. Really: look at the graph and ask yourself: could this possibly describe the value of any actual house?"
(I'm not sharing the web link to Hilary's chart because that page lists her address, but this is indeed what it looked like as of late Tuesday.)
When I asked Zillow about this Zestimate, the company quickly declared it an error and said it would be fixed soon -- likely this week.
The company stood by the current-day Zestimate for Hilary's house. But the previous five years of figures were buggy, a spokeswoman said.
"They knew about this bug already and were working on fixing it," Zillow's Katie Curnutte said of the company's tech staffers.
Why the algorithm change? It's the second major alteration since Zillow launched in 2006 -- the first came in 2008 -- and the company says it's in response to the volatile housing market.
Now, Curnutte said, "We look at more recent transactions and we weight them heavier than we used to. That lets us keep up with a market that's changing really rapidly."
Zestimates are free, and they offer homeowners (and prospective buyers) a glimpse at what properties might be worth -- most homes in the country, not just ones for sale. But they're frequently criticized by appraisers, real estate agents and academics for inaccuracies.
Curnutte said Zestimates aren't an appraisal and aren't meant to be used for anything other than a "starting point" in the quest to determine value. But she says the new calculation method puts them closer to the mark.
Zillow checks its accuracy by comparing Zestimates to arms-length home sales, and it says the typical error in the Baltimore metro area dropped from 12.1 percent with the previous algorithm to 7.9 percent with the new one. (Here's more on the margin of error by region, including the ranges. For instance, Zillow says just over one-third of Baltimore-area Zestimates are within 5 percent of the sales price.)
For Hilary's part, she says she always assumed that Zestimates weren't spot on but could be "somewhat informative about relative changes in housing prices, as long as whatever mistakes they made were more or less consistent over time." The recent zig-zagging threw her for a loop and made her question even that.
The new algorithm was met by "howls" from some homeowners whose Zestimates plummeted as a result, The Wall Street Journal reported recently. But Hilary says she really doesn't care what value estimate is applied to her house.
"I'm not remotely interested in selling it in the foreseeable future," she says.
How did your Zestimate change? Does it look right to you?