Zillow and their "zestimates"

[Allen W.]

Hello all,

In your experience have you found Zillow or any other site's online estimate of a property to be fairly accurate once you had a formal appraisal completed?

If not, how far off was it? I know they can't account for any upgrades you completed before the appraisal.

[Steve K.]

According to Zillow themselves (https://www.zillow.com/z/zesti...), Zestimates have a median error rate of 2% for listed properties and 7% for off market properties nationwide. Their website breaks down their accuracy rate in different parts of the country and in my market they claim to have a less than 38% chance of being within 20% of the actual home value. Because the median home price is north of $1.5M here, that means we can only hope for Zestimates to be within about $300k of the true market value on average, which is too far off to be valuable data. 

Zillow is probably a lot more accurate in areas where properties change hands more often, where they have more recent data, more comps and less variation street to street, block to block etc. and I suppose the margin of error is less of a big deal in lower-value areas where the margin amounts to merely tens of thousands instead of hundreds of thousands. Overall, not very helpful in my market.

Anecdotally, they are currently off on all of my personal properties (in Denver metro) by 200,000-500,000. Ever since they famously lost half a billion big ones during the second half of last year by using their own algorithms to buy properties through their ibuyer program, I've been cross-referencing their Zestimates every time I run comps out of curiosity. They're usually off by several hundred thousand in my experience. On the deal I'm working on now, the Zestimate is $1.5M and we're under contract for $2.5M (off market deal, no competitive bidding). It appraised at $2.55M, so Zillow was off by $1.05M according to the appraiser on that one, which is kind of a big difference. Yet, the Zestimate still comes up in every listing appointment, so obviously everyone looks at it. A high percentage of FSBO's are indeed suspiciously priced exactly at the Zestimate, to the detriment of those sellers.

 For these reasons, I think it's a fair question to ask whether statistical/AI-based price prediction causes more harm than good. People should know that property pricing is based on many things besides number of beds and baths in a home/ the few data points readily available online, and that algorithms don't know what the difference in price is on one side of the street vs. the other, one side of the tracks vs. the other, city or county address, one school district or another, updated tastefully inside or a hack-job remodel by a first-time flipper (or not updated in 75 years)... what the place smells like, if it's on a busy street, on a hill with a nice view or in a hole with drainage issues, near a hot new development or in an area experiencing terminal post-industrial decline etc.,etc., etc. I think most people do know this but I suppose it's just so much quicker and easier to look at a Zestimate than to pull comps, that's impossible to resist. But many people probably don't realize how these online estimates have rapidly gotten even less accurate just these past few months, with such dynamic conditions in the market recently. Prices have been rising so much, so fast that many are worth a hundred thousand or more than they were just last month. CMA's created a few weeks ago are no longer accurate. Sales from just a few months ago need to be adjusted up by as much as 20%. Bidding wars on certain properties are completely unpredictable and can drive sales prices way up above what the recently sold comps support. Because most buyers in my market are moving in from other locations, properties that are move-in ready and have the cute factor that most buyers are looking for are getting 20+ offers within hours of being listed and going under contract for 5-30% above-ask, while those with tenants in them or needing immediate work are not getting nearly as much competition. Algorithms don't understand these market nuances. Plus it's impossible to create an accurate CMA right now without first calling all the listing agents of the comparable properties that are pending to find out what they'll close at, and making that adjustment to the comp. Bots don't do that but appraisers and Realtors do. I wouldn't count on Zillow for anything other than getting very rough, ballpark data, and with the understanding that it may not even be in the ballpark in certain markets. (Some of this reply was repurposed from a similar reply I made to a recent similar thread).