- Find a Real Estate Professional
- Realtors®
- Mortgage Lenders
- Home Improvement Pros
- Other Real Estate Services
- Review an Agent, Lender or Pro
- Marketing on Zillow
- Real Estate Agent Advertising
- Join the Professional Directory
- Popular
- Real Estate Market Reports
- More
Answers (15)
- sbklein
- Contributions:5
Not sure of who has what biases or comes from what parts of the industry in this exchange - but I can say with certainty that entities that are actually exposed to risk in the valuation of homes place virtually no value on the information provided by tax assessments. As Surbiton indicated, they are heavily biased downward because noone challenges a lower than anticipated rating. There are also other statistical issues, since the tax assessment uses much of the same information the AVM uses - but with a bias introduced by the potential for the owner to provide cherry-picked comparables that have issues the assessment entity may not be aware of. So, if tax assessments are used (by lenders or insurers), they are given very little weight.
- Dunes....
- Contributions:3894
Pasa that's a bunch of Crap..
- Pasadenan
- Contributions:21466
The National Association of Realtors® is the party that intentional bankrupted millions of people in the U.S. and almost Bankrupted over 100 countries in the world.
They are the largest political lobbying organization in the U.S., and it was their lobbying that created the housing bubble and foolish tax giveaways. All bubbles have to correct. But some people only look at short term gains without regard to long term consequences.
Complaining on one or two information sites is not going to address the problem. If the monoply is not broken, the problems for the consumers will only get worse.
Those that complain about websites like Cyberhomes.com and Eppraisal.com are really saying they want organizations like the National Association of Realtors® to rip off the public, and that they want the first amendment of the U.S. Constitution revoked and only "mega business" to dictate all prices in all market sectors.
They are the largest political lobbying organization in the U.S., and it was their lobbying that created the housing bubble and foolish tax giveaways. All bubbles have to correct. But some people only look at short term gains without regard to long term consequences.
Complaining on one or two information sites is not going to address the problem. If the monoply is not broken, the problems for the consumers will only get worse.
Those that complain about websites like Cyberhomes.com and Eppraisal.com are really saying they want organizations like the National Association of Realtors® to rip off the public, and that they want the first amendment of the U.S. Constitution revoked and only "mega business" to dictate all prices in all market sectors.
- Pasadenan
- Contributions:21466
Well, here is a link to a thread from 2007, complaining that the Zestimate was $200k low...
Zillow says this home has dropped by over 200,000 in the last month, wow I want it.
But, checking the sold prices for the property, it turned out that Zillow was right, and the Realtor that set the listing price was wrong...
Obviously, if it was worth more to the seller, it would have just been taken off the market instead.
Though there may be "price persistence", the market dynamics are subject to a lot of changes that are not visible to the average person. The machine algorithms and modeling are much more sensitive to these issues.
I've given links to a number of other free estimating sites... which one(s) do you think is better? Why?
Zillow says this home has dropped by over 200,000 in the last month, wow I want it.
But, checking the sold prices for the property, it turned out that Zillow was right, and the Realtor that set the listing price was wrong...
Obviously, if it was worth more to the seller, it would have just been taken off the market instead.
Though there may be "price persistence", the market dynamics are subject to a lot of changes that are not visible to the average person. The machine algorithms and modeling are much more sensitive to these issues.
I've given links to a number of other free estimating sites... which one(s) do you think is better? Why?
- Pasadenan
- Contributions:21466
Well, here is a link to a thread from 2007, complaining that the Zestimate was $200k low...
Zillow says this home has dropped by over 200,000 in the last month, wow I want it.
But, checking the sold prices for the property, it turned out that Zillow was right, and the Realtor that set the listing price was wrong...
Obviously, if it was worth more to the seller, it would have just been taken off the market instead.
Though there may be "price persistence", the market dynamics are subject to a lot of changes that are not visible to the average person. The machine algorithms and modeling are much more sensitive to these issues.
Zillow says this home has dropped by over 200,000 in the last month, wow I want it.
But, checking the sold prices for the property, it turned out that Zillow was right, and the Realtor that set the listing price was wrong...
Obviously, if it was worth more to the seller, it would have just been taken off the market instead.
Though there may be "price persistence", the market dynamics are subject to a lot of changes that are not visible to the average person. The machine algorithms and modeling are much more sensitive to these issues.
- Pasadenan
- Contributions:21466
No, you don't hand tweak a computer model for individual cases randomly brought to attention that have no basis in the modeling data used... to do so would completely defeat the purpose of automatically creating market indexes, and would show a bias toward properties listed by obnoxious Realtors®.
It is easier and more meaningful to explain the uniqueness of an estimate in the space provided.
Now, if one could capture improvement data from the county records; that would be a much better solution. I have some ideas... but not detailed enough to lobby Zillow to try to implement. (But I did drop Stan a few hints a couple times).
It is easier and more meaningful to explain the uniqueness of an estimate in the space provided.
Now, if one could capture improvement data from the county records; that would be a much better solution. I have some ideas... but not detailed enough to lobby Zillow to try to implement. (But I did drop Stan a few hints a couple times).
- Pasadenan
- Contributions:21466
Obviously they use as much public data as they can for the modeling. And obviously, the weighing depends on the learning factors when modeling to the recently sold.
Last sold price adjusted for last sold date is good for capturing the amenities, views, traffic, noise levels, roof quality, ceiling heights, casework, attic space, insulation levels, roof overhangs, mature trees... but it doesn't capture changes since purchase at all.
The square footage and number of bedrooms and number of bathrooms may capture some revisions since purchase, but never to the actual level of the improvement.
The tax assessed value is the best way presently to capture changes since purchase, but as the assessed value only increases by the change amount, it still isn't that good for that purpose either. Another public records variable is needed for that purpose, and so far, Zillow doesn't have one from the data collection companies they pay.
Last sold price adjusted for last sold date is good for capturing the amenities, views, traffic, noise levels, roof quality, ceiling heights, casework, attic space, insulation levels, roof overhangs, mature trees... but it doesn't capture changes since purchase at all.
The square footage and number of bedrooms and number of bathrooms may capture some revisions since purchase, but never to the actual level of the improvement.
The tax assessed value is the best way presently to capture changes since purchase, but as the assessed value only increases by the change amount, it still isn't that good for that purpose either. Another public records variable is needed for that purpose, and so far, Zillow doesn't have one from the data collection companies they pay.
- sbklein
- Contributions:5
Data quantity is not a problem at the entity where I do my work, so volatility is not an issue. It's pretty clear that the answer on this has to be the tax assessment, which is for less than the house was purchased for by the previous owner 10 years ago and significantly less than what I purchased it for 1 year ago. The tax assessment took place almost 2 years ago, which makes me wonder why it's affecting the algorithm now when it had not previously - perhaps they don't refresh that data often? What is particularly interesting/frustrating is that the addition of the tax assessment is overpowering the recent purchase price (a purchase which occurred when the index for this zip was lower than it is now), and a fair number of recent comparables. Basically, it is acting as though the recency of the assessment information (which seems to be very recent to the algorithm, even though it's actually older than the purchase price information) has led the model to essentially fully discount the purchase price and comparables - because the new Zillow price is $50K or so above assessment and $400K or so below purchase price and comparables. That seems like bad modeling to me - particularly since it places such heavy reliance on what is probably the least valuable piece of information available - the tax assessment (rather than the purchase price, recent comparables and price index changes since the most recent purchase(s)).
- Pasadenan
- Contributions:21466
Still wondering about the claims of the original poster about doing their own AVM's with no fluctuations....
How does one overcome sparse sales, random nature of specific sales, and missing data fields?
Remember, if you are looking for 10 coefficients, you need at least 10 data points... and for more reliability and usability, you need a lot more than 10 coefficients. There are only two possibilities to increase the number of data points... 1) increase the area the model covers, or 2) increase the time the sales could have occurred prior to the estimate. Either of these methods will mean it won't track the specific market area as well or as closely.
(By the way, tweaking individual estimates manually by adding data not included in the modeling does nothing to improve the market tracking of an area, and actually is counter productive to that goal. On the other hand, pointing out errors in "recently solds" in the vicinity does help improve the learning model).
How does one overcome sparse sales, random nature of specific sales, and missing data fields?
Remember, if you are looking for 10 coefficients, you need at least 10 data points... and for more reliability and usability, you need a lot more than 10 coefficients. There are only two possibilities to increase the number of data points... 1) increase the area the model covers, or 2) increase the time the sales could have occurred prior to the estimate. Either of these methods will mean it won't track the specific market area as well or as closely.
(By the way, tweaking individual estimates manually by adding data not included in the modeling does nothing to improve the market tracking of an area, and actually is counter productive to that goal. On the other hand, pointing out errors in "recently solds" in the vicinity does help improve the learning model).
- Pasadenan
- Contributions:21466
"(among other things, I build AVM models) - and am still stumped as to how this fact pattern can be possible in an automated environment." -
It appears that one of Zillow's concerns in giving too many details on their proprietary methods is that there are many that would like to "copy", "steal" or "borrow" Zillow's unique methods to improve tolerance range. Certainly, that seems to be why CoreLogic has sued Zillow and 10 other AVM companies for alleged patent infringement of U.S. Patent #5,361,201 (That, plus the patent expires next year becoming public domain, and the patent holders would like to make some money off of that patent before it expires).
But actually, I'm stumped how any AVM could possibly NOT have the fluctuations apparent in Zillow's trends given the random nature of what sells at any given point in time, the wide fluctuation in prices people are willing to accept for a sale, the sparse sales data when taking a small enough area and small enough time frame, and missing data fields that is so prevalent in much of the public and agent/owner provided data.
Perhaps you can share what you are doing for your Automated Valuation Models to avoid these fluctuations?
It appears that one of Zillow's concerns in giving too many details on their proprietary methods is that there are many that would like to "copy", "steal" or "borrow" Zillow's unique methods to improve tolerance range. Certainly, that seems to be why CoreLogic has sued Zillow and 10 other AVM companies for alleged patent infringement of U.S. Patent #5,361,201 (That, plus the patent expires next year becoming public domain, and the patent holders would like to make some money off of that patent before it expires).
But actually, I'm stumped how any AVM could possibly NOT have the fluctuations apparent in Zillow's trends given the random nature of what sells at any given point in time, the wide fluctuation in prices people are willing to accept for a sale, the sparse sales data when taking a small enough area and small enough time frame, and missing data fields that is so prevalent in much of the public and agent/owner provided data.
Perhaps you can share what you are doing for your Automated Valuation Models to avoid these fluctuations?
- Pasadenan
- Contributions:21466
As an outside observer that is very interested in the marketing trends and analysis, all I can say is they weight the tax assessed value adjusted for local rules and the last sold price adjusted for last sold date pretty heavily now.
Since it is a "learning model"; it appears to have something to do with when some of the recently solds were previously purchased. Zillow statisticians have specifically stated they don't use linear regression for their modeling, thus the learning model may tend to amplify some effects for some groups of housing more than others and more than one would expect.
Zillow has had 3 major modeling algorithms over it's history, but the fluctuations have always been present with each, and they have always been unique to the individual properties, canceling each other out when you select enough of them.
I've been requesting the ability to add and to average and to chart together for comparing the Zestimate trends of a select group of houses for sometime, and always told it was way down on the priority list. But I did find out this past week that it is possible to download the data underlying the chart; but the method is a bit obscure and not advertised.
But just using graphic scaling and overlaying of charts in a photo program, here is a comparison of the fluctuations of the methods used for one property:
present estimating method (green)
previous estimating method (red)
2007 estimating method (used until January 2008) (magenta):
Web address for full size for better legibility:
http://photos3.zillow.com/is/image/i0/i7/i6554/ISa21mbr9m9sir.jpg
Since it is a "learning model"; it appears to have something to do with when some of the recently solds were previously purchased. Zillow statisticians have specifically stated they don't use linear regression for their modeling, thus the learning model may tend to amplify some effects for some groups of housing more than others and more than one would expect.
Zillow has had 3 major modeling algorithms over it's history, but the fluctuations have always been present with each, and they have always been unique to the individual properties, canceling each other out when you select enough of them.
I've been requesting the ability to add and to average and to chart together for comparing the Zestimate trends of a select group of houses for sometime, and always told it was way down on the priority list. But I did find out this past week that it is possible to download the data underlying the chart; but the method is a bit obscure and not advertised.
But just using graphic scaling and overlaying of charts in a photo program, here is a comparison of the fluctuations of the methods used for one property:
present estimating method (green)
previous estimating method (red)
2007 estimating method (used until January 2008) (magenta):
Web address for full size for better legibility:
http://photos3.zillow.com/is/image/i0/i7/i6554/ISa21mbr9m9sir.jpg
- sbklein
- Contributions:5
Thanks - appreciate the responses! The longer ownership status suggestion doesn't seem like a likely explanation, because we purchased the house a year ago. The recent sales explanation is hard for me to believe because it should have had a similar impact on neighboring homes, which are mostly similar in basic metrics (except for land - I have the largest plot in the neighborhood). The updates thing is a possibility, though we have not made any updates - so the only way that could have had an impact is if a data cleanup somehow removed an update that had previously been in the system. Tax assessed value is an interesting suggestion. The tax assessment is oddly low (which was pitched as a selling point). Apparently the previous owners managed to successfully challenge the assessment (which had been close to the price we purchased it at). I'm not sure how, given that comparables tend to be significantly higher than the assessed value. But it's certainly possible that had an impact (though that assessment took place almost 2 years ago now.
In any case - I'm not selling now or any time soon, so it's not an urgent issue. But - I'm in the mortgage finance business (among other things, I build AVM models) - and am still stumped as to how this fact pattern can be possible in an automated environment.
In any case - I'm not selling now or any time soon, so it's not an urgent issue. But - I'm in the mortgage finance business (among other things, I build AVM models) - and am still stumped as to how this fact pattern can be possible in an automated environment.
- Pasadenan
- Contributions:21466
As for why yours and not the neighbors; it is because the modeling method used is not a comparables method. It is closer to a neural network method as used in optical character recognition.
The reason yours moved different than the neighbors is most likely the last sold price adjusted for the last sold date. The longer a property is owned, the more likely it is to be affected by sales that neighboring properties are minimally affect by.
Similarly, the tax assessed value may be having an impact.
The reason yours moved different than the neighbors is most likely the last sold price adjusted for the last sold date. The longer a property is owned, the more likely it is to be affected by sales that neighboring properties are minimally affect by.
Similarly, the tax assessed value may be having an impact.
- Pasadenan
- Contributions:21466
Since the "opinions of approximate value" are "editorials" and noted as such, there is nothing to sue over.
However, if you do believe that some recently "sold" data was used in error (they intentionally exclude distressed sales..., but the data collection company for given areas do not always copy the county records correctly), use the "report a problem" button on the "edit" pull down tab, under the main photo on the home details page to have the issue looked into.
In most cases, it is just way too few sales in the area recently to be able to do good modeling. If that is the case, the next sale will likely cause it to swing the other direction.
Zillow does want to address the wild fluctuation issue, so any specifics that can be brought to their attention on the issue will help them develop better algorithms to filter for that and account for it in the modeling.
However, if you do believe that some recently "sold" data was used in error (they intentionally exclude distressed sales..., but the data collection company for given areas do not always copy the county records correctly), use the "report a problem" button on the "edit" pull down tab, under the main photo on the home details page to have the issue looked into.
In most cases, it is just way too few sales in the area recently to be able to do good modeling. If that is the case, the next sale will likely cause it to swing the other direction.
Zillow does want to address the wild fluctuation issue, so any specifics that can be brought to their attention on the issue will help them develop better algorithms to filter for that and account for it in the modeling.
- Dana Scanlon, "Bethesda Specialist"
- Contributions:480
Hi,
Zestimates can be frustrating, and don't always make sense. You can't blame human error, because a human isn't involved. Its all based on complex and secret algorithms in any given geographic area.
Lets say a nearby home which is much smaller and less updated than yours sold recently. Bingo, that would affect the zestimate, even though it has absolutely no bearing on your property's value.
I always tell my clients, whether they are buyers or sellers, to pay no attention to these figures. Zillow's computers do not know who has an upgraded kitchen and bathrooms with all the bells and whistles, versus the vintage pink tile bathroom from the 1950s.
Zestimates can be as accurate or inaccurate as tax assessments!
Even if you are planning to sell your home soon I wouldn't be too concerned because any good realtor will tell their buyer the same thing.
If you want to try to contact zillow to see if they would look at the data and make a change you can try listingsupport@zillow.com.
Zestimates can be frustrating, and don't always make sense. You can't blame human error, because a human isn't involved. Its all based on complex and secret algorithms in any given geographic area.
Lets say a nearby home which is much smaller and less updated than yours sold recently. Bingo, that would affect the zestimate, even though it has absolutely no bearing on your property's value.
I always tell my clients, whether they are buyers or sellers, to pay no attention to these figures. Zillow's computers do not know who has an upgraded kitchen and bathrooms with all the bells and whistles, versus the vintage pink tile bathroom from the 1950s.
Zestimates can be as accurate or inaccurate as tax assessments!
Even if you are planning to sell your home soon I wouldn't be too concerned because any good realtor will tell their buyer the same thing.
If you want to try to contact zillow to see if they would look at the data and make a change you can try listingsupport@zillow.com.
Sharp Home Value Change for no apparent reason.
Incidentally, the home was just appraised, for a refinance, at slightly over the earlier Zillow estimate, and a significantly smaller house, built at a similar time, with a much smaller plot and on a heavily trafficked nearby street, just sold for $50K more than the new estimate.
So what happened and how do I get it fixed?
Stating a discriminatory preference in an advertisement for housing is illegal. If you think this content is discriminatory or otherwise inappropriate and feel it should be removed from Zillow, please let us know by completing the information above.
We will review this content. Thanks for helping make the site more useful to everyone. To learn more, read Zillow's Good Neighbor Policy.