Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

washingtonpost.com

Post reply on HN
Print view View original

191–200 of 265 posts

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#191

Post by Jemaclus »

hbosch wrote: post #146
ProfessorLayton wrote: post #66

At least in the Bay Area, Zillow et al. are decent for getting a decently accurate ballpark figure in mind. A lot of sale prices are clearly put up below market to grab attention, knowing that the sale price is very likely to exceed it by a wide margin. Here in many parts of the Bay, you'll get laughed at and promptly ignored if you try and use these "zestimates" to try and put downward pressure on the price. Sellers…

A friend was looking at houses via Redfin. The agent told him on day 1 that he should expect to spend 10%–20% more than the online quote, as a rule of thumb. Since then I've taken "Zestimates" with ~65,000 to 100,000 grains of salt.

I used to work for Z, and I can confirm that the Zestimate is just that: an estimate. It doesn't know things like whether you did renovations or whether the house next door caught on fire and burned down and things like that. If something about your particular house changed in the last few years, the Zestimate is likely to be way off. That said, it's fairly good as far as comps go, and if you have an average house in the hood, the Zestimate is gonna be pretty close.

Regarding the rule of thumb you mentioned, my wife and I have been looking for a house in the Bay Area, and we've determined that the listing price is just a completely made up number. We made an offer on a house that was 100K over asking price, and the sellers countered with 40K on top of that (so 140K over asking). I was like... what? You listed for X, we offered X+100K, you don't get to counter that. lol

We obviously didn't get the house, and they wound up selling for 175K over asking price (+~25%). We've made several other offers, and sometimes the house goes for 50K over asking and sometimes it goes for 250K over asking. It just seems to be totally random. It's absurd. Instead, we actually rely a lot on the Zestimate and Redfin's estimate to figure out what the "real" value of the house is. Z and R tend to be roughly in the right ballpark for the Bay Area.

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#192

Post by WaxProlix »

DougWebb wrote: post #181
WaxProlix wrote: post #148

Earlier quoted context omitted.

Even without revealing the highest offer, the offers themselves constitute a silent auction, which is where a lot of the information asymmetry comes from. If I could speak to each of the other people and tell them I had X amount to spend we could all agree to walk away or make offers without going through a bunch of hassle and (for 32 of 33 offerers) heartbreak. The process really highlights to me how broken certain…

Why would the seller go along with the idea that the buyers would negotiate amongst themselves to choose a price?

They wouldn't, because it's not in their interest.

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#193

Post by michaelmrose »

devoply wrote: post #186
cmurphycode wrote: post #175

Earlier quoted context omitted.

I know it seems weird to have the "old" data change. But, it's not observed data, it's estimates made at a point in time. The estimate comes from the property itself as well as the algorithm. If they changed the algorithm, do you think it would be better to recalculate the old values, or show a huge swing in your home value, implying some sort of market trend or issue with your house? If I were implementing it, I thi…

They should be more transparent about their algo though. That would then allow users to actually know how they came to the conclusion that they did. Hopefully this lawsuit will reveal that algo.

Why would the lawsuit force them to reveal trade secrets? I don't see how you have any right to know how they arrived at the number they did. I can say publicly that I value your home at $5 and my neighbor bob can say its worth a million dollars.

These are both opinions and anyone seriously considering spending half a million or more can consider spending a modest amount on an appraisal.

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#194

Post by michaelmrose »

ccvannorman wrote: post #176
strictnein wrote: post #19

The Zillow estimate used to be spot on in our neighborhood, for several years even. Recently they all plunged to 10-20% below market value, even though all sales data puts it at the previous mark. The weird thing is that they also went back and recalculated old "Zestimate" data as well. On April 2nd the estimate for our house was $617,453, which is more or less what it would sell for. It had been above $600k since Ju…

If true and you can prove it, this may be valuable information in prosecuting Zillow for some pretty illegal and perverted stuff. I hope you will share this information with someone who can punish Zillow for this.

Punish zillow for what? Seems to be a confusion between something would benefit me/may harm me and the other side having any sort of obligation to consider your position.

An inaccurate estimate from someone who you paid nothing and whom owes you nothing may make your life more difficult but this doesn't imply they owe you anything.

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#195

Post by jdross »

hehheh wrote: post #38
_m8fo wrote: post #16

You would think real estate would prove to be one of the few real life fields where machine learning would be able to predict the prices pretty accurately. You have: 1. List price 2. "Favorites" on the MLS. 3. Square footage 4. Bedrooms 5. Bathrooms 6. Sold price for nearby properties 7. Sold price for nearby properties relative to their list price And the list goes on... Is it that difficult to predict? Literally th…

There's plenty of money in it. Imagine if you could price the house correctly out of the gate, reducing the amount of time the home stays on the market. It is indeed a hard problem to solve using ML. You'd need a few other things: 8. Floor count 9. Size and location of the bedrooms, bathrooms, kitchen 10. In-unit laundry (mostly for condos) 11. Proximity to neighbors 12. Existing tree coverage and other landscaping 1…

It's significantly more complicated, but at Opendoor we act as a market maker for real estate by doing this.

We sell our homes far faster than the market in Phoenix, without sacrificing on price.

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#196

Post by jdross »

vineet wrote: post #57
_m8fo wrote: post #16

You would think real estate would prove to be one of the few real life fields where machine learning would be able to predict the prices pretty accurately. You have: 1. List price 2. "Favorites" on the MLS. 3. Square footage 4. Bedrooms 5. Bathrooms 6. Sold price for nearby properties 7. Sold price for nearby properties relative to their list price And the list goes on... Is it that difficult to predict? Literally th…

Here is a company that is focusing on just that - estimating real estate values (and selling those estimates): https://www.housecanary.com/

At Opendoor we go a step further and actually purchase and sell the homes based on the prices our algorithms produce

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#197

Post by bkjelden »

hehheh wrote: post #38
_m8fo wrote: post #16

You would think real estate would prove to be one of the few real life fields where machine learning would be able to predict the prices pretty accurately. You have: 1. List price 2. "Favorites" on the MLS. 3. Square footage 4. Bedrooms 5. Bathrooms 6. Sold price for nearby properties 7. Sold price for nearby properties relative to their list price And the list goes on... Is it that difficult to predict? Literally th…

There's plenty of money in it. Imagine if you could price the house correctly out of the gate, reducing the amount of time the home stays on the market. It is indeed a hard problem to solve using ML. You'd need a few other things: 8. Floor count 9. Size and location of the bedrooms, bathrooms, kitchen 10. In-unit laundry (mostly for condos) 11. Proximity to neighbors 12. Existing tree coverage and other landscaping 1…

Real estate is not liquid enough that a 'correctly' priced house will sell immediately. You have to wait for a buyer that wants the exact featureset that house is offering.

Real estate has sort of a weird duality where it's viewed as a commodity in aggregate - but it is definitely not a commodity good to the individual buyers looking at houses.

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#198

Post by jdross »

EpicEng wrote: post #110
_m8fo wrote: post #16

You would think real estate would prove to be one of the few real life fields where machine learning would be able to predict the prices pretty accurately. You have: 1. List price 2. "Favorites" on the MLS. 3. Square footage 4. Bedrooms 5. Bathrooms 6. Sold price for nearby properties 7. Sold price for nearby properties relative to their list price And the list goes on... Is it that difficult to predict? Literally th…

So many intangibles. People's emotions (on both sides) are one, but then you have the house with the nice kitchen, or that other house with a not-so-nice kitchen, but a flatter yard. And on and on. Zillow uses plenty of ML, but what's their advertised accuracy? iirc it's something like within 20% 80% of the time?

Roughly 5% median absolute error. 90% within 20% nationwide.

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#199

Post by jdross »

wuliwong wrote: post #78
Spooky23 wrote: post #59

Earlier quoted context omitted.

You don't really need machine learning to do comps. There are literally dozens of tools to do this accurately. Zillow just sucks for a variety of reasons that have more to do with Zillow than anything else.

I think the difficulty is scaling the "art" of estimating prices. Any single home I could sit down and give a good estimate but to come up with a fully automated way to do that seems pretty tough still. I'm also not sure that the really hard parts are solved with ML as well.

I challenge you to come to Opendoor and beat the algorithm over a large set of homes :)

Re: Zillow faces lawsuit over ‘Zestimate’ tool that calculates a house’s worth

#200

Post by Houshalter »

mattkrause wrote: post #25
_m8fo wrote: post #16

You would think real estate would prove to be one of the few real life fields where machine learning would be able to predict the prices pretty accurately. You have: 1. List price 2. "Favorites" on the MLS. 3. Square footage 4. Bedrooms 5. Bathrooms 6. Sold price for nearby properties 7. Sold price for nearby properties relative to their list price And the list goes on... Is it that difficult to predict? Literally th…

If you want to try it, there are a few data sets kicking around, including a) https://www.kaggle.com/c/house-prices-advanced-regression-te... b) http://www.bis.org/statistics/pp_detailed.htm c) https://archive.ics.uci.edu/ml/datasets/housing If you had a really great model, you might be able to make some money with it--find underpriced (per the model) houses, buy them, and then try to sell them at the model's price.

There is a kaggle competition right now to predict house prices in russia. If you can do it better than anyone else, you can win tens of thousands of dollars.
Post reply on HN