Comment by TimPC | Hacker News Reader

Comment by TimPC | Hacker News Reader

Parent Full thread

It’s very hard to do good real estate pricing models from raw data. You can get all sorts of metrics on the property statistics but one important factor in the cost of the home is the state of renovation. Since most listings don’t have a historical listing of work done in the home your best hope of accurately assessing this comes from something like computer vision on the listing pictures. Assessing the quality of work, recency of work, quality of materials and factoring that into a house price is extremely hard. It would not surprise me at all if Zillow paid near market value for a whole bunch of dated homes that hadn’t kept up with the times and would sell for below the market of similar homes in the neighborhood according to typical metrics. The stories make it seem like their model had even more issues than that. Still I’m surprised anyone actually tried this idea given how poor the algorithms are and how obvious it is that key data points are not available.

What was stopping Zillow from having a human look at the property and adjust the price based on factors that their algorithm didn't consider?

Nothing, but that would require them hiring an appraiser on each transaction, which would be humbling for them to have to do (and while biased as an appraiser myself, I would not trust anyone else other than a licensed appraiser to do that type of adjustment work, because it’s a daily part of our jobs).

Or they could have an internal team of appraisers doing that type of work, but the business might not like it when the appraisers don’t validate their pricing structure because real data does not support it.

It's wild that they didn't have human appraisers review at least a sample of the houses to groundtruth the algorithm.

Greed and wanting to be ahead of others will blind you to truth standing right in your face. While everyday workers at Zillow didn’t have a say in company policy, you can’t tell me that the executives and project managers didn’t know what they were doing or why they were doing it. They didn’t want to be told their idea was bad or couldn’t work (which it did work until it didn’t) early on, so they got the debt they refused to pay through due diligence.

And Zillow has been around for almost 20 years, so it’s not like they didn’t have awareness of how the home market works or how to validate pricing metrics. From the outside it fells like willful ignorance, but perhaps they suffer from the same ills any large company trying to “move fast and break things”.

What stopped them is that the entire rational for Zillow's approach is that human appraisal isn't needed. An algorithm, they think, should be able to do a good enough job. Any individual person who, when presented with this scenario, thought they should have brought in a human to look at the house does has been selected out of working for Zillow.

Reply on news.ycombinator.com