4 ms·

Zillow's estimates are absolutely garbage. I'm not sure how they determine estimates but they are too low by about 30%, at least in the Bay Area. If anyone fr

by steven2012 14y ago

Zillow's estimates are absolutely garbage. I'm not sure how they determine estimates but they are too low by about 30%, at least in the Bay Area. If anyone from Zillow is reading this thread, they really need to do a better job with gathering their estimates by doing things like weighting most recent comps better, and taking a notice of how much newer sold houses are being overbid on. They've been around for a few years now, and their estimates are still a joke. They can calibrate their estimates by finding houses that are currently pending, making an estimate, and then seeing what they end up selling for once the deal closes. It appears they don't even bother doing any sort of backtesting on their estimates.

xradionut 14y ago
Zillow's data is very, very bad. It doesn't reflect the market or the tax assessment records in my area.
smackfu 14y ago
It's a very hard problem. * The price data isn't reliable. I see random sales for way under market price in my area, probably family sales. Tax evaluations should be reliable, but they are often years out of date. * The input variables aren't reliable. This condo went for less because it has really high condo fees, but the condo fees aren't in the listing. This other house says it has lakefront AND beachfront AND water views. These two houses have "decks", one is a tiny front porch, the other is a huge thing. These two houses have the same lot size, but one is all grass and the other is straight down a hill. * There isn't much data volume. You are trying to come up with an accurate price for every house in a neighborhood, but only a few percent sell each year. So the one or two houses that sell determine the prices for all the rest, but the data for those houses is unreliable due to the previous two factors.
- steven2012 14y ago
  It is a hard problem, but the fact they are touting their Zestimate, and having the values so ridiculously off, gives me no confidence in Zillow. And it hasn't gotten any better since Zillow first started, which makes me have even less faith in them. If they backtested their data and corrected for it, or had some sort of machine learning algo, I would bet it would get a lot better. To me, it looks like they are using a simple average $/sqft algo and it looks like they go back a year or so to determine this $/sqft without a higher weighting to more recent house prices. If they have Redfin agents bidding in the area, they can pipe in the price their clients bid for certain houses, and use whether or not they got the house as a training data point as well.
- jyu 14y ago
  Sales information also does not take into account market factors such as 'days on market' and 'condition' or if it was a retail sale vs short sale vs foreclosure. There is a lot of pricing fluctuation even within a neighborhood with different house models.
- xradionut 14y ago
  Zillow list properties that are gutted, abandoned as the same as properties that have been maintained, remodeled or expanded. This information in many cases is reflected in the annual property tax assessment. Since these records are available online gratis, you think Zillow could use them as a part of their figures.

xradionut 14y ago

Zillow's data is very, very bad. It doesn't reflect the market or the tax assessment records in my area.

smackfu 14y ago

It's a very hard problem. * The price data isn't reliable. I see random sales for way under market price in my area, probably family sales. Tax evaluations should be reliable, but they are often years out of date. * The input variables aren't reliable. This condo went for less because it has really high condo fees, but the condo fees aren't in the listing. This other house says it has lakefront AND beachfront AND water views. These two houses have "decks", one is a tiny front porch, the other is a huge thing. These two houses have the same lot size, but one is all grass and the other is straight down a hill. * There isn't much data volume. You are trying to come up with an accurate price for every house in a neighborhood, but only a few percent sell each year. So the one or two houses that sell determine the prices for all the rest, but the data for those houses is unreliable due to the previous two factors.

steven2012 14y ago
It is a hard problem, but the fact they are touting their Zestimate, and having the values so ridiculously off, gives me no confidence in Zillow. And it hasn't gotten any better since Zillow first started, which makes me have even less faith in them. If they backtested their data and corrected for it, or had some sort of machine learning algo, I would bet it would get a lot better. To me, it looks like they are using a simple average $/sqft algo and it looks like they go back a year or so to determine this $/sqft without a higher weighting to more recent house prices. If they have Redfin agents bidding in the area, they can pipe in the price their clients bid for certain houses, and use whether or not they got the house as a training data point as well.
jyu 14y ago
Sales information also does not take into account market factors such as 'days on market' and 'condition' or if it was a retail sale vs short sale vs foreclosure. There is a lot of pricing fluctuation even within a neighborhood with different house models.
xradionut 14y ago
Zillow list properties that are gutted, abandoned as the same as properties that have been maintained, remodeled or expanded. This information in many cases is reflected in the annual property tax assessment. Since these records are available online gratis, you think Zillow could use them as a part of their figures.