Why Zillow Couldn’t Make Algorithmic House Pricing Work

There was a great episode of The Indicator from Planet Money about this.<p>One aspect not mentioned in this article is that when you offer an average price for a home, you are only going to get the owners of below-average homes taking you up on the offer, i.e. the Lemons. This is a tricky problem for Zillow to resolve given the mismatch in knowledge between the existing owner and the iBuyer.<p><a href="https:&#x2F;&#x2F;www.npr.org&#x2F;2021&#x2F;11&#x2F;08&#x2F;1053689886&#x2F;ibuyers-zillow-and-the-lemons-problem" rel="nofollow">https:&#x2F;&#x2F;www.npr.org&#x2F;2021&#x2F;11&#x2F;08&#x2F;1053689886&#x2F;ibuyers-zillow-and...</a>

This whole issue is fascinating to me and no doubt will be studied in schools in the coming years. One of the lead product designers on Zillow Offers wrote a retrospective yesterday that outlines, broken user journeys, greed and the attempt to algorithmically one-size-fits-all the whole market. He lays a lot of the blame squarely at the feet of incompetent executive leadership<p>From his notes: “They valued old-school growth hacks from 2007 over taking user-centered, data-informed, research-backed approaches. They spent more time flexing and building decks that drove an internal narrative of Sellers being happy, than actually addressing real and persistent product problems that ultimately led most sellers down a path where they didn&#x27;t know what to expect, didn&#x27;t understand the process and were inundated with calls and scheduling.”<p><a href="https:&#x2F;&#x2F;www.linkedin.com&#x2F;posts&#x2F;bardlavens_zillowoffers-zillowfailure-activity-6864367665520234496-yuld" rel="nofollow">https:&#x2F;&#x2F;www.linkedin.com&#x2F;posts&#x2F;bardlavens_zillowoffers-zillo...</a>

Isn&#x27;t another problem that people who would use their automated service to sell their houses will skew towards those whose houses have some value-diminishing problem (which their algorithms have a hard time accounting for) or for other odd reasons have lower than usual market price?<p>People who can fetch a premium for their houses probably wouldn&#x27;t use an automated algorithm, which tends towards the mean?<p>E.g. a self selection of those who would use this service tends to give Zillow a bunch of houses of lower value than average.

This does not mean the zestiamte estimate is flawed, but that flipping homes is inherently risky in independent of the price estimate algo.<p>The huge, v-shaped recovery of the housing market and also stock market and GDP caught a lot of experts and pundits by surprise. The expectation was that, even with stimulus, that things would take years to recover as the US economy clawed its way out from the depths of the pandemic, but then Game Stop stock went up 100x and suddenly a FOMO unlike the likes ever seen took hold. Glad I didn&#x27;t sell, and also I added to some stock positions in April 2020. Suddenly everything went from extreme over-supply to extreme shortages in a year.

Interesting topic but this article is very short on the important details.<p>“Buy and hold for a few months” would have been a good strategy for most of the pandemic, but Zillow appears to have overpaid substantially to the point where that didn’t work. Their predictions must have been badly off. (Also: this is a great example of the “winner’s curse” in first-price auctions!)<p>The article quotes a figure of $65,000 “over market price”, but where “market price” comes from for those houses it bought is unclear.<p>The claim of “market manipulation” by Zillow is ludicrous on its face without substantial evidence, which I’m guessing did not appear in a tiktok video.<p>If all iBuyers in the Phoenix market _combined_ were buying 5% of homes, that doesn’t suggest to me that Zillow _alone_ would be able to exert any pricing pressure.<p>I would love to read a more detailed article than this one if any are available.

I&#x27;d recommend this article for a deeper look at how Zillow and its competitors exhibited different behavior in Phoenix as the market cooled:<p><a href="https:&#x2F;&#x2F;www.mikedp.com&#x2F;articles&#x2F;2021&#x2F;10&#x2F;19&#x2F;ibuying-is-hard-zillow-pauses-new-purchases" rel="nofollow">https:&#x2F;&#x2F;www.mikedp.com&#x2F;articles&#x2F;2021&#x2F;10&#x2F;19&#x2F;ibuying-is-hard-z...</a><p>&gt; As the market cooled between August and September, Opendoor and Offerpad purchased fewer houses, while Zillow purchased more.<p>&gt; The iBuyers also adjusted to changing market conditions by paying less for houses. The median purchase price in Phoenix peaked in August. Opendoor and Offerpad&#x27;s median purchase price also peaked in August before tracking the market and declining in September. But Zillow kept paying more and more.

Lots of general contractors, with enough capital to buy a home, do some repairs, and sell it; They have a very good knowledge of the local market. Can do the repairs themselves.<p>Larger companies probably focus on commercial developments and&#x2F;or condos.

Because models &lt;&gt; reality.<p>That’s it. Full stop. Anyone who thinks otherwise will eventually face this same fate.<p>For instance, from the sub-heading, “But the pandemic messed up its predictions”<p>should really be this instead, “but their algorithms couldn’t handle a situation they didn’t anticipate”.

House price volatility is localized to the point that sample sizes are too small. Prices are generally neighborhood-based and the sales in a neighborhood are too infrequent. One quick divorce sale, or one over-eager buyer can skew the data enough to foil this approach.<p>As an anecdotal example, my development has 115 homes. In the last 12 months, 3 have changed hands. The first sold at 5% over the Zestimate, the second, 10% over, the third 15% over. How would the Zillow algorithm predict the price of the next sale?

After the failure everyone is able to say why it wouldn&#x27;t work.

They have to offer the seller more than the seller thinks the winning bid would’ve been if they sold it without Zillow. And it has to be less (including their transaction and improvement costs) than the actual winning bid will be, when they turn around and sell it themselves.

&gt;They tend to offer lower prices than traditional buyers, but attract sellers by promising faster, all-cash deals.<p>That was the deal?<p>They thought they&#x27;d get a lower price by offering cash and quick closing?<p>Even in a down market (last time I bought) I offered quick closing and all cash.<p>Sellers didn&#x27;t like it. It took around 9 months before I got a call back from one house I wanted saying &quot;Is that offer still good?&quot;<p>In the meantime they had accepted 3 other&#x2F; better offers ...but pending the other buyer selling their house, all of those feel through.<p>And that was in a down market...

2020-2021 was a very weird period. The fact that a given predictive model failed during these years is not really a strong indictment against the model. It makes sense to suspend the algorithmic trading during that time until the situation goes back to normal.

Any metric, once it becomes a target, ceases to be a good metric.<p>Seems to me like the guys at Zillow never heard of this one.

Rumor has it that this tiktok video[0] was the cause of the pause and eventual disbandment of iBuyer program.<p>[0]<a href="https:&#x2F;&#x2F;www.tiktok.com&#x2F;@seangotcher&#x2F;video&#x2F;7007855978309848325" rel="nofollow">https:&#x2F;&#x2F;www.tiktok.com&#x2F;@seangotcher&#x2F;video&#x2F;700785597830984832...</a>

I think Matt Levine&#x27;s take was pretty interesting: <a href="https:&#x2F;&#x2F;www.bloomberg.com&#x2F;opinion&#x2F;articles&#x2F;2021-11-03&#x2F;zillow-is-done-trading-houses" rel="nofollow">https:&#x2F;&#x2F;www.bloomberg.com&#x2F;opinion&#x2F;articles&#x2F;2021-11-03&#x2F;zillow...</a>

Lack of liquidity.<p>Algorithmic pricing works in liquid markets. Housing is not a liquid market.

ctrl-F “adversarial environment” - no hits.<p>Markets are not physics - there is no neutral POV.

The simple answer is they didn&#x27;t fact check with local realtors.